Recent advances in understanding the effects of climate change on the world’s oceansHollowed, Anne B; Barange, Manuel; Garçon, Véronique; Ito, Shin-ichi; Link, Jason S; Aricò, Salvatore; Batchelder, Harold; Brown, Robin; Griffis, Roger; Wawrzynski, Wojciech
doi: 10.1093/icesjms/fsz084pmid: N/A
Abstract In June 2018, >600 scientists from over 50 countries attended the Fourth International Symposium on the Effects of Climate Change on the World’s Oceans (ECCWO-4). ECCWO-4 provided a forum for scientists to share information, build understanding, and advance responses to climate impacts on oceans and the many people, businesses and communities that depend on them. Seven Key Messages emerging from the symposium and relevant information from recently published literature are summarized. Recent scientific advances are improving our ability to understand, project, and assess the consequences of different levels of 21st century climate change for ocean ecosystems and ocean dependent communities. Outcomes of the symposium highlighted the need for on-going engagement with stakeholders, communities, and managers when considering the trade-offs associated with tactical and strategic opportunities for adaptation to climate change. Science informed adaptation frameworks that engage the public in their development are needed for effective management of marine resources in a changing climate. The summary provides a brief overview of the advances in climate-ocean science emerging from the symposium and provides context for the contributed papers within the broader socio-ecological advances of the discipline. Introduction The Fourth International Symposium on the Effects of Climate Change on the World’s Oceans (ECCWO-4) was jointly convened in Washington, USA in 2018 by the International Council for the Exploration of the Sea (ICES), the North Pacific Marine Science Organization (PICES), the Intergovernmental Oceanographic Commission (IOC) of the United Nations Educational, Scientific and Cultural Organization (UNESCO); the Food and Agriculture Organization of the United Nations (FAO), and the U.S. National Oceanic and Atmospheric Administration (NOAA). Previous symposia were held in Gijón, Spain, in 2008 (Valdés et al., 2009), Yeosu, South Korea, in 2012 (2nd ICES/PICES/IOC Effects of Climate Change on the World's Oceans. 2013. ICES JMS 70:915–1054) and Santos, Brazil, in 2015 (Barange et al., 2016). ECCWO-4 occurred at a key juncture for the scientific community’s understanding of the implications of climate impacts on ocean systems and societies responses to these impacts. National and international assessments confirm that the Earth’s climate and oceans are rapidly changing. The impacts are already evident in some regions and more impacts are expected with continued changes in the planet’s climate system (Jay et al., 2018; IPCC, 2018). Several critical milestones have been achieved since the last ECCWO symposium. The publication of the IPCC Fifth Assessment Report (AR5) laid the foundation for a landmark international agreement to curb carbon emissions (the Paris Agreement, agreed at the 21st session of the Conference of the Parties of the UN Framework Convention on Climate Change or “COP 21” in December 2015). The adoption by the UN General Assembly in 2015 of the 2030 Agenda for Sustainable Development and its 17 Sustainable Development Goals (SDGs) provided a roadmap for global efforts to incorporate climate change impacts within the global socio-ecological system. These developments underscore the worldwide recognition that changing oceans will have significant impacts on a range of issues spanning health, food security, gender equality and women’s empowerment, sanitation, energy, sustainable economic growth, ensuring sustainable consumption and production, mitigating and adapting to climate change, protecting biodiversity, and maintaining peaceful dialogues among nations. The symposium organizers recognized that future efforts to inform the public of the risks associated with these changes would benefit from the products, outcomes and lessons that emerged from this symposium [website (https://meetings.pices.int/meetings/international/2018/climate-change/background)]. These factors provided the motivation to organize ECCWO-4 to review and evaluate new science relevant to the ECCWO-4. More specifically, the timing of the symposium was designed to provide scientists an opportunity to prepare peer-reviewed publications that could be considered by national and international climate change assessment teams tasked with informing policy makers of the observed and projected impacts of climate change on the world’s oceans and trade-offs of different adaptation options with respect to ocean systems. Key messages and their relevance to articles in this issue Convening a symposium of over 600 scientists from over 50 nations to discuss the consequences of different levels of 21st century climate change for ocean ecosystems and ocean dependent communities [for details see symposium scope (https://meetings.pices.int/meetings/international/2018/climate-change/scope)] required contributions from many scientists. To minimize the chance of missing Key Messages, we used a multi-pronged approach that included input from session and workshop conveners [see list of conveners (https://meetings.pices.int/meetings/international/2018/climate-change/program)], the scientific steering committee [see Organizers (https://meetings.pices.int/meetings/international/2018/climate-change/organizers)], the Plenary Speakers (https://meetings.pices.int/meetings/international/2018/climate-change/speakers), and participating scientists, to identify and compile Key Messages that emerged from the symposium and that represent the current state of the discipline. Session conveners were asked to prepare a summary slide that identified the main “takeaways” from their session and any new scientific advances that have emerged since ECCWO-3. We synthesized input from these sources into the following seven Key Messages. We also, identify the articles in this symposium volume that highlight one or more of these Key Messages. Climate change is already affecting the oceans and the socio-ecological systems that depend on them Many of the sessions at the symposium provided evidence that oceans are changing in ways that impact humans. Rapid ecosystem changes were reported in high latitude systems (Overland et al., 2018; Stevenson and Lauth, 2019). Previous studies found evidence of trophic level differences in responses to climate change in marine ecosystems (Perry et al., 2010; Friedland et al., 2012; Kortsch et al., 2015; Ullah et al., 2018). Friedland et al. (in press) also detected such differences in trophic responses in his study of the northeast shelf of the United States. These transitions have far reaching implications for weather, national security, food security, transport and commerce (Larsen et al., 2014; Hoegh-Guldberg et al., 2014; IPCC, 2018; Free et al., 2019). The Quo Vadimus “graphic novel” article provides a pictorial summary of key findings from the symposium based on selected cartoons by Bas Kohler (Link et al., in press). This graphic novel describes the future landscape of our world under changing climate conditions. Using clever and humorous cartoons the artist conveys the message that oceans are changing and there is much at stake if these changes continue. The cartoons depict the urgent need for scientists to advance our understanding of these changes and the risks associated with them. The graphic novel concludes with a reminder that finding solutions to deal with climate induced change and their consequences will require two-way communication with managers and stakeholders to co-design effective adaptation pathways. The full set of cartoons from the Symposium are available on the ECCWO-4 website (https://www.flickr.com/photos/pices/sets/72157669719136128). Technical advances in existing and new observation networks are improving our understanding of key ocean processes, but our ability to project future ocean conditions at seasonal to multi-decadal time scales is still incomplete To monitor changes in ocean environments, scientists are developing new technologies to collect more and targeted measurements to better understand the oceanic carbon cycle and to minimize uncertainties for both short-term prediction and long-term projection of carbon uptake, ocean acidification, and deoxygenation. The continuation and enhancement of ocean observing using diverse technologies (satellite, moorings, ships, profiling floats, gliders, etc.) is the foundation for detecting and understanding the implications of climate change on ocean systems. Several international groups are developing initiatives to enhance global ocean observing systems. For example, in response to this need, the Chinese Academy of Sciences created a new Center for Mega Science [Fang Wang Symposium Video (https://meetings.pices.int/publications/video#2018-ECCWO)]. The UN launched the United Nations Decade of Ocean Science for Sustainable Development (https://en.unesco.org/ocean-decade) (2021–2030), which is designed to collect the information necessary to understand and project ocean responses to multiple stressors. Several scientists reported on extreme ocean events and their ecosystem and societal impacts. Marine heat waves have been observed in several regions and they are expected to be more common and persistent in the future [Alistair Hobday Symposium Video (https://meetings.pices.int/publications/video#2018-ECCWO)]. The development of a coherent standardized framework for reporting ocean heat waves is advancing this field of research (Hobday et al., 2018a). It was noted that these extreme events may allow researchers to rehearse their responses to persistent anomalous warming events [a “stress test,” see Francisco Werner Symposium Video (https://meetings.pices.int/publications/video#2018-ECCWO)]. For decades scientists have been striving to identify leading indicators of anomalous climate events. Advances in the collection and availability of ocean observations and new analytical methods for rapid synthesis of large data sets using models have improved our ability to predict anomalous ocean conditions on seasonal scales. However, predictions of future ocean states at decadal time scales may not yet be possible in all regions. The symposium provided an opportunity for scientists to assess short- and medium-term ocean forecasting skill (Tommasi et al., 2017; Payne et al., 2017; Frölicher and Laufkötter, 2018; Frölicher et al., 2018). In several ocean regions there is strong evidence that short-term (3–9 months) forecasting of oceanographic processes is now feasible. This finding opens new opportunities for the advancement of ecosystem-linked stock assessments. New methods and frameworks are being developed to detect regime shifts and utilize early detection systems of changing climate conditions to inform management (Hobday et al., 2016; Payne et al., 2017; Hobday et al., 2018b; Oliver et al., 2018). Hobday et al. (in press) tackle the ethical responsibilities of scientists who are developing forecasting systems for use in management and industry planning. Their article includes reviews of seven examples of forecast systems serving a range of stakeholders. They identify a variety of issues that should be considered during the scoping, development, delivery and evaluation phases of climate forecasting systems. The authors provide ten principles, which serve as a useful guide for forecasting teams. The paper by Bisagni et al. (in press) explores inter-annual variability in Gulf Stream warm-core rings (WCRs) in a Western Boundary Current ecosystem. This paper is another example of the importance of collecting and maintaining time series of ocean observations. This study utilizes a rich set of observations to evaluate time trends in the interannual variability in the frequency and location of strong WCRs over a 44-year time frame. The authors then used this long time series to identify a statistically significant relationship between time trends in the abundance of longfin squid (Doryteuthis pealeii) over the U.S. Northeast Shelf Large Marine Ecosystem and the spatially averaged annual mean WCR encounter area. Mochizuki and Watanabe (in press) found a distinct subdecadal variation in the 2000s over the tropical Pacific that was rarely observed in other decades. The authors compared observed patterns of ocean-heat-content to decadal hindcasts and found that subdecadal variations were difficult to predict. This demonstrates one of the comparative studies needed to fully interpret and understand patterns in the predictive skill of models that was called for in Hobday et al. (in press). Hermann et al. (in press) provide a good demonstration of the technical advancements in our ability to project multi-decadal ocean trajectories. Multiple publications (Sundby et al., 2016; Cheung, 2018; Overland et al., 2018; Stevenson and Lauth, 2019) provide evidence that high latitude ecosystems are changing rapidly and that widespread change is projected in the future under different emission scenarios (e.g. RCP 4.5 and RCP 8.5). Hermann et al. (in press) present results from the Alaska Climate Integrated Modeling project. In their study, outputs from the Fifth Coupled Model Inter-comparison Project are downscaled for use in projecting future ocean conditions in the eastern Bering Sea to 2100. The paper considers both scenario (representative concentration pathways) and structural (between model differences) uncertainty in projections. The authors use multivariate statistical methods to explore the modes of variability and covariability across variables to characterize the dominant relationships among bio-physical features. They demonstrate that these dominant modes can be used to rapidly estimate the regional ecosystem responses to large ensembles of forcing scenarios. These approaches reveal the spatial extent and magnitude of ecosystem change expected to occur in the region by the end of the century, and locations where high levels of variability are expected. Despite significant gaps, our understanding of socio-ecological systems has improved sufficiently to enable us to contrast the ecological and societal impacts of different future scenarios An exciting advance since ECCWO-3 has been the development of fully coupled socio-ecological models capable of projecting future scenarios based on global responses to the challenges of climate change and regional responses to natural resource management. New modelling tools and insights into climate-impacts allow us to contrast futures under alternative societal responses to climate change (Lotze et al., 2019; Skogen et al., 2018; Hermann et al., in press). The uncertainty associated with ecological projections of climate change impacts on marine ecosystems is being incorporated by contrasting outcomes from a range of ecosystem models with varying levels of complexity. Kaplan et al. (2019) explored the relative importance of incorporation of mechanistic linkages in stock projection models. Understanding and predicting how ecological changes will impact human societies, institutions and economies will be critical for effective adaptation (Karp et al., in press). Some marine organisms exhibit a capacity to adapt to climate change, but there are energetic and physiological costs, as well as limits Research continues to reveal complex energetic and physiological trade-offs associated with individual or species adaptation to changing environmental conditions [see Widdicombe Symposium Video (https://meetings.pices.int/publications/video#2018-ECCWO)]. There are several publications in this issue focused on the ecological effects of climate change. The research findings from the current efforts build on previous ECCWO symposia and they reveal complex and sometimes unexpected biological and ecological responses at different life stages, with lagged effects. Important lessons can be learned by examining the variations in adaptive responses and consideration of the full suite of environmental stressors within the socio-ecological system (Hobday et al., in press). Friedland et al. (in press) conducted a retrospective examination of the role of ocean temperature on phytoplankton, zooplankton, fish and macroinvertebrate distributions along the northeast US continental shelf large marine ecosystem. They found that species and marine communities responded differently to thermal change. The authors conclude that the different responses may be linked to the organism’s capacity to adapt to novel thermal regimes. They hypothesized that spatial distributions of lower trophic level organisms were less responsive to thermal change because of their ability to integrate seasonal thermal changes, whereas, the responses of higher trophic level species depended on both the availability of lower trophic level organisms and environmental conditions. Two papers in this volume reveal the costs and tradeoffs of species adaptation to environmental stress. Long et al. (in press) explore the implications of ocean acidification on the respiration and feeding of juvenile red and blue crabs in the Bering Sea. The results show that crabs can adapt to ocean acidification exposure if feeding ration is increased, but there are energetic and physiological costs to this adaptation. Crawford et al. (in press) utilize a 38-year time series of the abundance and diets of three South African sea bird species to assess the effects of climate on forage availability. The results of these papers illustrate the need for multi-dimensional considerations when evaluating the adaptive capacity of sea birds to changing environmental conditions. Options available for societal adaptation are more limited if current trends of greenhouse gas emissions continue Coupled socio-ecological models and vulnerability assessments have been used to evaluate the trade-offs associated with different societal responses. Results from these models and assessments show that adaptation options are more limited when the higher greenhouse gas emission scenarios are considered. Recent studies show that as CO2 increases, there are fewer and less effective adaptation and repair options for ocean systems (Gattuso et al., 2015, IPCC, 2018). Understanding these trade-offs within an integrated socio-ecological framework helps to inform options for human responses to climate driven changes to marine ecosystems (Holsman et al., in press). Tactical and strategic opportunities for societal adaptation to climate change have also been revealed through engagement with institutions and dependent communities eager to plan their own future Several case studies where researchers engaged coastal communities in efforts to develop effective climate adaptation strategies were presented at the symposium. Some of these case studies have been extensively documented in a recent FAO report (Barange et al., 2018). The symposium highlighted that there is an urgent and on-going need for scientists to interact with dependent communities to assist in the development and evaluation of realistic adaptation pathways that are based on the best available scientific information. Commitments to an on-going process of science product delivery to resource managers and users will enable more intelligent decision-making by stakeholders, which is necessary to keep up with the changing ocean ecosystems (Wise et al., 2014; Hobday et al., in press). Coastal communities are seeking adaptation options and a guide to good governance to plan for the future. Many coastal communities are turning to aquaculture, marine ranching, and fish attraction technologies to fill critical needs for food security and dietary demands. While technological advancements in fish capture and fish production are emerging, holistic studies that assess the long-term implications of these adaptation responses will be needed to guide future developments to sustain food resources and preserve human and planetary health (Willett et al., 2019). Whether operating at the global or local level, it is important to engage respectfully in adaptation by involving all stakeholders, supporting their progress and recognizing the importance of local knowledge [see Merle Sowman Symposium Video (https://meetings.pices.int/publications/video#2018-ECCWO) and Colenbrander and Sowman, 2015]. It is important that stakeholder involvement in adaptation planning be ongoing and iterative to address obstacles that arise along the way. Communicating and documenting these challenges could provide valuable lessons for similar initiatives. In this volume, Karp et al. (in press) reviewed the conservation and management challenges encountered when U.S. fisheries target species that are shifting their distributions and/or productivity. The authors identify six tactical steps that scientists and managers can take to address shifting spatial distributions and abundance of managed species. Among these, the authors encourage continued close collaboration and communication among scientists, managers, and stakeholders as a key step in the development of adaptation options to support sustainable fisheries management in a changing world. Adaptive management frameworks are urgently needed to address climate-driven policy issues Recent studies suggest that coordinated adaptation frameworks that are capable of incorporating flexibility and insights from past experiences, will lead to wiser decision making and more effective long-term management of impacts (Poulain et al., 2018). Cooperative fishery management frameworks will be needed for effective management of fisheries targeting species with shifting spatial distributions (Pinsky et al., 2018). Two contributed papers in this volume considered adaptive frameworks at different levels of community, manager, stakeholder, and government engagement. Pinnegar et al. (in press) applies a vulnerability and an adaptive capacity analysis of the fisheries sector of Dominica. They consider the implications of long-term climate change and the occurrence of extreme weather events. The paper identifies which parishes in Dominica are most vulnerable to climate change impacts on fish and fisheries. The authors conclude that the vulnerability framework provides useful information for government and development agencies that can be used to enhance resilience and build adaptive capacity. Holsman et al. (in press) also consider pathways to build climate-informed management portfolios. They define climate-resilient management as a mix of dynamic, adaptive and fixed management approaches. The authors address the trade-offs of these different approaches and their relative utility for managers considering current, near future and distant future management of fisheries resources. Holsman et al. (in press) propose a nested adaptation framework that applies different tools for short, medium, and long-term climate impact planning horizons may be needed to address the climate change issues on marine ecosystems. They identify suites of key research activities that could be adopted to improve the adaptation of management frameworks to develop climate-informed policies that foster science-management-stakeholder dialogs. Summary and next steps This brief introduction provides a brief glimpse of the exciting new developments happening in the ocean and climate science arena. ECCWO-4 made it clear that integrated multidisciplinary research programs are appearing across the globe in response to the societal need for information on the impacts of climate change on the world’s oceans. ECCWO-4 included numerous opportunities for the discussion of socio-ecological systems with two way feedbacks between human responses and the ecosystem change [see additional ECCWO-4 session summaries in PICES newsletter (https://meetings.pices.int/publications/pices-press/volume26/PPJul2018.pdf)]. As we look forward to the ECCWO-5 symposium, we anticipate that the scientific community will strive ever harder to provide sound scientific information to inform climate actions that preserve life under water while providing food, jobs, and other services contributing to human well-being into the future. This symposium series informs the public of recent advances in climate science and engages the community in thoughtful consideration of the societal choices we all face with respect to climate change implications on, and consequences for, the world’s oceans. We trust that the papers presented in this symposium issue, not only represent a useful snapshot of the state of climate science research with respect to the ocean, but will also help direct the community toward future advances in assessing the impacts and effects of changing climate on the world’s oceans. Acknowledgements We appreciate the organizing committee, session conveners, presenters, and participants of the Fourth International Symposium on the ECCWO-4. We express our gratitude to the Royal Norwegian Embassy in Washington D.C. and the fourteen co-sponsoring organizations for their active interest and support. We acknowledge that numerous national and international scientific organizations provided critical support that helped to sustain the pace of scientific advancement noted in this symposium. We also thank the student volunteers who helped to make the symposium a success. Anne Hollowed’s participation was supported by the Fisheries and the Environment (FATE), Stock Assessment Analytical Methods (SAAM) and the North Pacific Regimes and Ecosystem Productivity (NPCREP) programs with the National Marine Fisheries Service, and NOAA’s Research Transition Acceleration Program. Shin-ichi Ito’s participation was supported by the Japan Society for the Promotion of Science KAKENHI project, grants JP15H05823 and JP18H03956. The scientific ideas, views, and opinions presented in this paper are solely those of the authors and do not represent the views of ICES, PICES, IOC-UNESCO, FAO, or NOAA. References Barange M. , Bahri T., Beveridge M., Cochrane K., Funge-Smith S., Poulain F. 2018 . Impacts of climate change on fisheries and aquaculture – synthesis of current knowledge, adaptation and mitigation options. In Fisheries and Aquaculture Technical Paper. FAO , Rome . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Barange M. , King J., Valdés L., Turra A. 2016 . The evolving and increasing need for climate change research on the oceans . ICES Journal of Marine Science , 73 : 1267 – 1271 . Google Scholar Crossref Search ADS WorldCat Bisagni J. , Nicols O., Pettipas R. in press. Inter-annual variability of Gulf Stream warm core ring interactions without continental shelf and effects on longfin squid abundance, 1981–2004 . ICES Journal of Marine Science . OpenURL Placeholder Text WorldCat Cheung W. W. L. 2018 . The future of fishes and fisheries in the changing oceans . Journal of Fish Biology , 92 : 790 – 803 . Google Scholar Crossref Search ADS PubMed WorldCat Colenbrander D. R. , Sowman M. R. 2015 . Merging socioeconomic imperatives with geospatial data: a non-negotiable for coastal risk management in South Africa . Coastal Management , 43 : 270 – 300 . Google Scholar Crossref Search ADS WorldCat Crawford R. , Sydeman W. J., Thompson S. A., Sherley R., Makhado A. in press. Food habits of an endangered seabird indicate recent poor availability of abundant forage resources . ICES Journal of Marine Science , 76 : 1344--1352. OpenURL Placeholder Text WorldCat Free C. M. , Thorson J. T., Pinsky M. L., Oken K. L., Wiedenmann J., Jensen O. P. 2019 . Impacts of historical warming on marine fisheries production . Science , 363 : 979 – 983 . Google Scholar Crossref Search ADS PubMed WorldCat Friedland K. D. , Stock C., Drinkwater K. F., Link J. S., Leaf R. T., Shank B. V., Rose J. M., et al. . 2012 . Pathways between primary production and fisheries yields of large marine ecosystems . PLoS ONE , 7 : e28945. Google Scholar Crossref Search ADS PubMed WorldCat Friedland K. D. , McManus M. C., Morse R. E., Link J. S. in press. Event scale and persistent drivers of fish and macroinvertebrate distributions on the Northeast US Shelf . ICES Journal of Marine Science , 76 : 1316--1334. OpenURL Placeholder Text WorldCat Frölicher T. L. , Fischer E. M., Gruber N. 2018 . Marine heatwaves under global warming . Nature , 560 : 360 – 364 . Google Scholar Crossref Search ADS PubMed WorldCat Frölicher T. L. , Laufkötter C. 2018 . Emerging risks from marine heat waves . Nature Communications , 9 : 650. Google Scholar Crossref Search ADS PubMed WorldCat Gattuso J. P. , Magnan A., Billé R., Cheung W. W. L., Howes E. L., Joos F., Allemand D., et al. . 2015 . Contrasting futures for ocean and society from different anthropogenic CO2 emissions scenarios . Science , 349 : aac4722. Google Scholar Crossref Search ADS PubMed WorldCat Hermann A. J. , Gibson G. A., Cheng W., Ortiz I., Aydin K., Wang M., Hollowed A. B., et al. . in press. Projected biophysical conditions of the Bering Sea to 2100 under multiple emission scenarios . ICES Journal of Marine Science , 76 : 1280--1304. OpenURL Placeholder Text WorldCat Hobday A. J. , Hartog J. R., Manderson J. P., Mills K. E., Oliver M. J., Pershing A. J., Siedlecki S. in press. Ethical considerations and unanticipated consequences associated with ecological forecasting for marine resources . ICES Journal of Marine Science , 76 : 1244--1256. OpenURL Placeholder Text WorldCat Hobday A. J. , Alexander L. V., Perkins S. E., Smale D. A., Straub S. C., Oliver E. C. J., Benthuysen J. A., et al. . 2016 . A hierarchical approach to defining marine heatwaves . Progress in Oceanography , 141 : 227 – 238 . Google Scholar Crossref Search ADS WorldCat Hobday A. J. , Oliver E. C. J., Gupta A. S., Benthuysen J. A., Burrows M. T., Donat M. G., Holbrook N. J., et al. . 2018a . Categorizing and naming MARINE HEATWAVES . Oceanography , 31 : 162 – 173 . Google Scholar Crossref Search ADS WorldCat Hobday A. J. , Spillman C. M., Eveson J. P., Hartog J. R., Zhang X., Brodie S. 2018b . A framework for combining seasonal forecasts and climate projections to aid risk management for fisheries and aquaculture . Frontiers in Marine Science , 5 , Article 137. doi:10.3389/fmars.2018.00137 Google Scholar OpenURL Placeholder Text WorldCat Hoegh-Guldberg O. , Cai R., Poloczanska E. S., Brewer P. G., Sundby S., Hilmi K., Fabry V. J., et al. . 2014 . The ocean. In Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part B: Regional Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change , pp. 1655 – 1731 . Ed. by Barros V. R., Field C. B., Dokken D. J., Mastrandrea M. D., Mach K. J., Bilir T. E., Chatterjee M., Ebi K. L., Estrada Y. O., Genova R.C., Girma B., Kissel E. S., Levy A. N., MacCracken S., Mastrandrea P. R., White L. L. Cambridge University Press , Cambridge, United Kingdom and New York, NY, USA . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Holsman K. , Hazen E., Haynie A., Gourguet S., Hollowed A., Borgrad S., Samhouri J., et al. . in press. Toward climate-resiliency in fisheries management . ICES Journal of Marine Science , 76 : 1368--1370. OpenURL Placeholder Text WorldCat IPCC. 2018 . Summary for policymakers. In Global Warming of 1.5°C. An IPCC Special Report on the Impacts of Global Warming of 1.5°C above Pre-Industrial Levels and Related Global Greenhouse Gas Emission Pathways, in the Context of Strengthening the Global Response to the Threat of Climate Change, Sustainable Development, and Efforts to Eradicate Poverty , p. 32 . Ed. by Masson-Delmotte V., Zhai P., Pörtner H-O., Roberts D., Skea J., Shukla P. R., Pirani A., Moufouma-Okia W., Péan C., Pidcock R., Connors S., Matthews J. B. R., Chen Y., Zhou X., Gomis M. I., Lonnoy E., Maycock T., Tignor M., Waterfield T. World Meteorological Organization , Geneva, Switzerland . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Jay A. , Reidmiller D. R., Avery C. W., Barrie D., DeAngelo B. J., Dave A., Dzaugis M., et al. . 2018 . Overview. In impacts, risks, and adaptation in the United States. In Fourth National Climate Assessment , pp. 33 – 71 . Ed. by Reidmiller D. R., Avery C. W., Easterling D. R., Kunkel K. E., Lewis K. L. M., Maycock T. K., Stewart B. C. U.S. Global Change Research Program , Washington, DC, USA . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Kaplan I. C. , Francis T. B., Punt A. E., Koehn L. E., Curchitser E., Hurtado-Ferro F., Johnson K. F., et al. . 2019 . A multi-model approach to understanding the role of Pacific sardine in the California Current food web . Marine Ecology Progress Series , 617–618 : 307 – 321 . Google Scholar Crossref Search ADS WorldCat Karp M. A. , Peterson J. O., Lynch P. D., Griffis R. B., Adams C. F., Arnold W. S., Barnett L. A. K., et al. . in press. Accounting for shifting distributions and changing productivity in the development of scientific advice for fishery management . ICES Journal of Marine Science , 76 : 1305--1315. OpenURL Placeholder Text WorldCat Kortsch S. , Primicerio R., Fossheim M., Dolgov A. V., Aschan M. 2015 . Climate change alters the structure of arctic marine food webs due to poleward shifts of boreal generalists . Proceedings of the Royal Society B , 282 : 20151546. Google Scholar Crossref Search ADS PubMed WorldCat Larsen J. N. , Anisimov O. A., Constable A. J., Hollowed A. B., Maynard N., Pestrud P., Prowse T. D., et al. . 2014 . Polar regions. In Part B: Regional Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change , pp. 1567 – 1612 . Ed. by Barros V. R., Field C. B., Dokken D. J., Mastrandrea M. D., Mach K. J., Bilir T. E., Chatterjee M., Ebi K. L., Estrada Y. O., Genova R. C., Girma B., Kissel E. S., Levey A. N., MacCracken S., Manstrandrea P. R., White L. L. Cambridge University Press , New York . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Link J. S. , Kohler B., Brady M. M. P., Griffis R., Ito S-I., Garçon V., Hollowed A., et al. . in press . A graphic novel from the 4th International Symposium on the Effects of Climate Change on the World’s Oceans . ICES Journal of Marine Science , 76 : 1221--1243. Google Scholar OpenURL Placeholder Text WorldCat Long W. , Pruisner P., Swiney K. M., Foy R. in press . Effects of ocean acidification on the respiration and feeding of juvenile red and blue king crabs (Paralithodes camtschaticus and P. platypus) . ICES Journal of Marine Science , 76 : 1335--1343. Google Scholar OpenURL Placeholder Text WorldCat Lotze H. K. , Tittensor D. P., Bryndum-Buchholz A., Eddy T. D., Cheung W. W., Galbraith E. D., Barange M, et al. . 2019 . Global ensemble projections reveal trophic amplification of ocean biomass declines with climate change . Proceedings of the National Academy of Sciences Jun 2019, 201900194 , doi:10.1073/pnas.1900194116. Google Scholar OpenURL Placeholder Text WorldCat Mochizuki T. , Watanabe M. in press. Observed and hindcasted subdecadal variability of the tropical Pacific climate . ICES Journal of Marine Science , 76 : 1271--1279. OpenURL Placeholder Text WorldCat Oliver E. C. J. , Donat M. G., Burrows M. T., Moore P. J., Smale D. A., Alexander L. V., Benthuysen J. A., et al. . 2018 . Longer and more frequent marine heatwaves over the past century . Nature Communications , 9 : 1324. Google Scholar Crossref Search ADS PubMed WorldCat Overland J. , Dunlea E., Box J. E., Corell R., Forsius M., Kattsov V., Olsen M. S., et al. . 2018 . The urgency of Arctic change . Polar Science , doi:10.1016/j.polar.2018.11.008 Google Scholar OpenURL Placeholder Text WorldCat Payne M. R. , Hobday A. J., MacKenzie B. R., Tommasi D., Dempsey D. P., Fässler S. M. M., Haynie A. C., et al. . 2017 . Lessons from the first generation of marine ecological forecast products . Frontiers in Marine Science , 4 : 1 – 15 . Google Scholar Crossref Search ADS WorldCat Perry R. I. , Cury P., Brander K., Jennings S., Möllmann C., Planque B. 2010 . Sensitivity of marine systems to climate and fishing: concepts, issues and management responses . Journal Marine Systems , 79 : 427 – 435 . Google Scholar Crossref Search ADS WorldCat Pinnegar J. K. , Engelhard G. H., Norris N. J., Theophille D., Sebastien R. D. in press. Assessing vulnerability and adaptive capacity of the fisheries sector in Dominica: long-term climate change and catastrophic hurricanes . ICES Journal of Marine Science , 76 : 1353--1367. OpenURL Placeholder Text WorldCat Pinsky M. L. , Reygondeau G., Caddell R., Palacios-Abrantes J., Spijkers J., Cheung W. W. L. 2018 . Preparing ocean governance for species on the move . Science , 360 : 1189 – 1191 . Google Scholar Crossref Search ADS PubMed WorldCat Poulain F. , Himes-Cornell A., Shelton C. 2018 . Methods and tools for climate change adaptation in fisheries and aquaculture. In Impacts of Climate Change on Fisheries and Aquaculture. Synthesis of Current Knowledge, Adaptation and Mitigation Options , pp. 535 – 566 . Ed. by Barange M., Bahri T., Beveridge M. C. M., Cochrane K. L., Poulain F. FAO Fisheries and Aquaculture Technical Paper , Rome, Italy . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Skogen M. D. , Hjollo S. S., Sando A. B., Tjiputra J. 2018 . Future ecosystem changes in the Northeast Atlantic: a comparison between a global and a regional model system . ICES Journal of Marine Science , 75 : 2355 – 2369 . Google Scholar Crossref Search ADS WorldCat Stevenson D. E. , Lauth R. R. 2019 . Bottom trawl surveys in the northern Bering Sea indicate recent shifts in the distribution of marine species . Polar Biology , 42 : 407 – 421 . Google Scholar Crossref Search ADS WorldCat Sundby S. , Drinkwater K. F., Kjesbu O. S. 2016 . The North Atlantic spring-bloom system – where the changing climate meets the winter dark . Frontiers in Marine Science , 3 : 1 – 12 . Google Scholar Crossref Search ADS WorldCat Tommasi D. , Stock C. A., Hobday A. J., Methot R., Kaplan I. C., Eveson J. P., Holsman K., et al. . 2017 . Managing living marine resources in a dynamic environment: the role of seasonal to decadal climate forecasts . Progress in Oceanography , 152 : 15 – 49 . Google Scholar Crossref Search ADS WorldCat Ullah H. , Nagelkerken I., Goldenberg S. U., Fordham D. A. 2018 . Climate change could drive marine food web collapse through altered trophic flows and cyanobacterial proliferation . PLoS Biology , 16 : e2003446. Google Scholar Crossref Search ADS PubMed WorldCat Valdés L. , Peterson W., Church J., Brander K., Marcos M. 2009 . Our changing oceans: conclusions of the first International Symposium on the Effects of Climate Change on the World’s Oceans . ICES Journal of Marine Science , 66 : 1435 – 1438 . Google Scholar Crossref Search ADS WorldCat Willett W. , Rockström J., Loken B., Springmann M., Lang T., Vermeulen S., Garnett T., et al. . 2019 . Food in the anthropocene: the EAT-Lancet Commission on healthy diets from sustainable food systems . The Lancet Commissions , 393 : 447 – 492 . Google Scholar Crossref Search ADS WorldCat Wise R. M. , Fazey I., Stafford Smith M., Park S. E., Eakin H. C., Archer Van Garderen E. R. M., Campbell B. 2014 . Reconceptualising adaptation to climate change as part of pathways of change and response . Global Environmental Change , 28 : 325 – 336 . Google Scholar Crossref Search ADS WorldCat Published by International Council for the Exploration of the Sea 2019. This work is written by US Government employees and is in the public domain in the US. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model) Published by International Council for the Exploration of the Sea 2019. This work is written by US Government employees and is in the public domain in the US.
Ethical considerations and unanticipated consequences associated with ecological forecasting for marine resourcesHobday, Alistair, J;Hartog, Jason, R;Manderson, John, P;Mills, Katherine, E;Oliver, Matthew, J;Pershing, Andrew, J;Siedlecki,, Samantha
doi: 10.1093/icesjms/fsy210pmid: N/A
Abstract Forecasts of marine environmental and ecosystem conditions are now possible at a range of time scales, from nowcasts to forecasts over seasonal and longer time frames. Delivery of these products offers resource managers and users relevant insight into ecosystem patterns and future conditions to support decisions these stakeholders face associated with a range of objectives. The pace of progress in forecast development is so rapid that the scientific community may not be considering fully the impacts on stakeholders and their incentives. Delivery of information, particularly about future conditions and the uncertainties associated with it, involves a range of judgements, or “ethical” considerations, including treatment of forecast failure, inequity in stakeholder response options, and winners and losers in commercial markets. Here, we explore these often unanticipated considerations via a set of case studies spanning commercial fishing, recreational fishing, aquaculture, and conservation applications. We suggest that consideration of ethical issues by scientists and their research partners is needed to maintain scientific integrity and fairness to end users. Based on these case studies and our experience, we suggest a set of ten principles that might be considered by developers and users of ecological forecasts to avoid these ethical pitfalls. Overall, an interdisciplinary approach, and co-production with end users will provide insurance against many unanticipated consequences. Introduction Marine species distributions and abundances are highly dynamic in both space and time, thus management, conservation, and sustainable exploitation is difficult (Keyl and Wolff, 2008; Ritz et al., 2011). Increasing human pressures on the ocean, in particular climate change, are resulting in changes in ecosystem characteristics and dynamics (Merrie et al., 2014). This means historical experience for a range of marine managers and resource users is less reliable when planning future decisions (Hodgkinson et al., 2014; Hobday et al., 2016). At the same time, asynchronies in ecological dynamics, fishery science, and policy development are threatening the effectiveness of management and governance arrangements (Hennessey and Healey, 2000; Pinsky and Fogarty, 2012; Pershing et al., 2015; Pinsky et al., 2018). This ultimately affects the sustainability of fishing businesses and their delivery of seafood, and undermines the credibility of science and governance institutions. If only humans could see the future or at least synoptically view the present, then these trends might be less problematic [While our focus here is on future predictions, nowcasts and short-term synoptic hindcasts of ocean conditions and species distribution are also important to managers and stakeholders. They reveal the spatial evolution of environmental conditions and species patterns which inevitably leads to short-term mental forecasts based on marine users experience and intuition (Eveson et al., 2015). Hindcasts also reveal the importance of past environmental events such as heatwaves and provide insight into important mechanisms affecting occupancy dynamics of fish and fisheries in the present and near future that might be related to those events.]. Information about the future can support proactive, rather than reactive, decision-making (Hobday et al., 2016) and forecasts have been delivered for agricultural sectors for many decades (see Asseng et al., 2012; Marshall et al., 2014). The rise of forecasting for marine resource management has been built on the accessibility of real-time ocean information, such as satellite-based temperature and ocean colour measurements as well as regional in situ mooring and autonomous observations (Hobday et al., 2016; Siedlecki et al., 2016; Payne et al., 2017). Breakthroughs in recent years now offer the prospect of useful spatial and temporal views of the future ocean and its biology generally known as ecosystem forecasts [The ecological forecasts described here have a wide set of uses but at this time do not include age structure or vital rates, in contrast to stock assessments used by fishery managers to set quotas and allocate fishing effort. Most current stock assessments are made without environmental information, the few that are blur the current boundary between environmental forecasts and assessments (Punt et al., 2014).]. For example, regional ocean models provide information on time and space scales that allow near-term forecasts of a range of environmental variables that influence the distribution, abundance, and phenology of marine species (Stock et al., 2015; Tommasi et al., 2017). Progress is also being made towards multi-year predictions (Salinger et al., 2016; Payne et al., 2017). These models predict spatio-temporal patterns in primary environmental variables (e.g. sea surface temperature, bottom oxygen) which can be delivered directly to end users (Spillman and Hobday, 2014; Siedlecki et al., 2016) or incorporated into habitat models representing a proxy for distribution of a species of interest (Hobday et al., 2011; Eveson et al., 2015). New metrics of interest that closer approximate experienced species habitat are also being forecast, including eddies (Hobday and Hartog, 2014), hypoxic volume (Siedlecki et al., 2016; Testa et al., 2017), aragonite severity index (Siedlecki et al., 2016), fine scale flow regimes (Scales et al., 2018), degree heating days (Spillman, 2011; Liu et al., 2018), habitat volume (Brodie et al., 2018), and habitat duration (Champion et al., 2019). Forecasts for other ecosystem impacts that build on primary environmental variables are also being developed, (e.g. harmful algal blooms, Brown et al., 2013; Anderson et al., 2016; noxious jellyfish, Gershwin et al., 2014). Providing information about the future, via forecasts, provides a range of benefits around informed decision-making but comes with a range of risks. Forecasting can lead to decisions that are different from those that would have been made without a forecast. Scientists tend to think such information is valuable, but it can also be disruptive to existing practice and decision-making. A formal ethical understanding of the risks (Lacey et al., 2015) is beyond the scope of this paper, but we recognize a range of judgements are made in the forecast development and delivery process that could be classified as ethical considerations. We use the term “ethical” in a normative sense, as have other ocean researchers (Barbier et al., 2018), to refer to principles of conduct or practice that would be considered good behaviours by other scientists and stakeholders. For example, forecast developers must recognize that not all people want to know about the future, with estimates of up to 90% of people preferring not to know about negative future events, and even 40–70% preferring to be ignorant regarding positive events (Gigerenzer and Garcia-Retamero, 2017). Just as people at risk of medical conditions may choose not to have probabilistic tests, forecast recipients may not always welcome information on future ocean conditions. Such views have been encountered amongst individuals that receive seasonal forecasts, with comments such as “I don’t want to know everything about the future” commonly expressed to the authors. Some fishers, for example, like their business the way it is, and have confidence to manage in the face of environmental variability (Hodgkinson et al., 2014). They may see such future information as removing their competitive advantage over less skilled operators. Information about the future can also be challenging to integrate, disruptive to existing mental models, and lead to “decision regret.” An ethical response may be to respect that choice, or to address the cause of the motivation and seek to illustrate the advantage of information about a rapidly changing ocean in partnership with forecast users. While there are a range of additional motivations for not wanting to know about the future (Gigerenzer and Garcia-Retamero, 2017), we do not explore these here, but simply note that rational explanations exist. The ethical responsibilities of researchers with respect to their methodological choices in climate downscaling, and the potential consequences of these choices have been addressed by Hewitson et al. (2014), however, we see the need for additional consideration of the nature of risk and responsibility at the interface between seasonal forecasting research and operational decisions that these forecasts influence. Just as Hewitson et al. (2014) argue that downscaled climate information must address the criteria of being plausible, defensible, and actionable, forecast developers cannot absolve themselves of their ethical responsibility when informing end users and must, therefore, be diligent in ensuring any information provided does not lead to perverse outcomes (sensuLacey et al., 2018). For example, while most seasonal forecast teams now address plausibility (i.e. consistent with other mechanisms) and defensibility (e.g. skill assessment), substantial interaction between the forecaster and user is often required to understand actionable information (Hobday et al., 2016; Payne et al., 2017). Scientists differ from medical and engineering professionals who have a charter of professional responsibility accompanied by oaths (e.g. do no harm) and recertifications, however, they must still be cognizant of judgements and ethical considerations (Lacey et al., 2018). Forecast developers influence strategic and operational decisions, such as resource allocation and spatial and temporal fishing strategies, and outcomes of these decisions can hinge on the prediction, rather than actual experience. In the following section, we describe ethical issues encountered as early developers of marine ecological forecasts for stakeholders engaged in fisheries, aquaculture, and conservation. These case studies illustrate a range of ethical issues that are consistent with schemes presented by Lacey et al. (2015) and Hewitson et al. (2014). These issues are encountered in key forecast phases defined by Hobday et al. (2016): (i) scoping, (ii) development, (iii) delivery, and (iv) evaluation (Figure 1). We then describe a set of forecasting ethics based on ten principles in each of the four phases. Figure 1. Open in new tabDownload slide Phases of forecast development, modified from Hobday et al. (2016). Evaluation leads to improved development. Figure 1. Open in new tabDownload slide Phases of forecast development, modified from Hobday et al. (2016). Evaluation leads to improved development. Ethical issues—lessons from existing forecast systems We reviewed seven examples of ecological forecast systems from both coasts of the United States and south-east Australia (Table 1). These forecasts served commercial and recreational fishing, aquaculture, and conservation stakeholders. Much of this work was motivated to help marine resource sectors cope with future uncertainty and promote dynamic and sustainable management. Based on review of these examples, we identified ethical issues in a range of categories in each phase of forecast development and delivery (Table 2). We developed these phase categories in a bottom-up iterative manner, based on discussion of our examples and reference to published literature, followed by review and refinement. Each of the examples revealed different issues across the phase categories, as discussed in the following sections. Table 1. Summary of ecological forecasting examples describing the context for each system. Salmon aquaculture forecasts in Tasmania Dynamical seasonal forecasts to predict water temperatures for south-east Tasmanian Atlantic salmon (Salmo salar) farm sites several months into the future are used to manage production risk (Spillman and Hobday, 2014). High summer temperatures pose a significant risk to production systems of these farms. Based on 20 years of historical validation, the model shows useful skill for all months of the year at lead-times of 0–1 months. Model skill is highest when forecasting for winter months, and lowest for December and January predictions. The poorer performance in summer is due to increased variability due to the convergence of several ocean currents offshore from the salmon farming region. Accuracy of probabilistic forecasts exceeds 80% for all months at lead-time 0 month for the upper tercile (warmest 33% of values) and exceeds 50% at a lead-time of 3 months. Industry engagement is high, and delivery of forecasts is ongoing, supplemented by industry–scientist discussions (Hobday et al., 2016). Funding: Industry Institution: CSIRO/Bureau of Meteorology Stakeholders: Salmon Farmers Recreational dolphinfish forecasts in eastern Australia A seasonal forecast of the habitat and density of dolphinfish (Coryphaena hippurus), based on sea surface temperatures, was developed for the east coast of New South Wales (NSW) Australia (Brodie et al., 2017). Two prototype forecast products were created; geographic spatial forecasts of dolphinfish habitat and a latitudinal summary identifying the location of fish density peaks. The less detailed latitudinal summary was designed to limit the resolution of habitat information to prevent potential resource over-exploitation by fishers in the absence of total catch controls. The dolphinfish habitat forecast was accurate at the start of the annual dolphinfish migration in NSW (December) but other months (January–May) showed poor performance due to spatial and temporal variability in the catch data used in model validation. Habitat forecasts for December were useful up to 5 months ahead, with performance decreasing as forecasts were made further into the future. Funding: Researchers (University PhD) Institution: CSIRO/Bureau of Meteorology/UNSW Stakeholders: Recreational Fishers/Managers Commercial tuna forecasts in the Great Australia Bight A habitat forecast system based on a seasonal ocean model and electronic tagging data was developed in collaboration with the Southern Bluefin Tuna (SBT, Thunnus maccoyii) fishery in Australia to overcome challenges posed by novel environmental conditions (Eveson et al., 2015). A dramatic change in the distribution of SBT compromised the ability of the fishery to efficiently locate and harvest the species. In partnership with industry representatives, a seasonal forecasting system was implemented to project the likely distribution of SBT several months into the future (Eveson et al., 2015). Forecasts are delivered daily via an industry-specific website, which has assisted fishers to efficiently catch SBT under variable climatic conditions. There is a cap on catches so this system does not contribute to over-exploitation. Funding: Industry/FRDC Institution: CSIRO/Bureau of Meteorology Stakeholders: Fishing industry Lobster forecasts in Maine The American lobster (Homarus americanus) fishery is currently the highest-valued commercial fishery in the United States, worth nearly $670 million in landed value in 2016. Over 80% of the value is landed in the state of Maine. The 2012 Northwest Atlantic heat wave disrupted this fishery by prompting early warming of the ocean and an earlier-than-normal large influx of lobster landings. Since readiness for the product was not aligned throughout the supply chain, a glut of lobsters developed, resulting in a price collapse (Mills et al., 2013). At that time, managers and key industry members in Maine questioned whether the early onset of the high-landings period could have been predicted, with the expectation that such information would give valuable early notice for the fishery and supply chain to prepare appropriately for the upcoming season. A regression-based seasonal forecast was developed to provide the expected timing of the summer uptick in landings (June–July) based on ocean temperatures in March–April (Mills et al., 2017). This forecast was formally issued to the industry in 2015 and 2016 via a public website. While it proved technically reliable at predicting the summer uptick timing, it also created unexpected challenges and was not perceived as useful to the industry (Pershing et al., 2018). Funding: Research agency Institution: GMRI Stakeholders: Fishing Industry/Managers Northwest Atlantic industry–science collaborative research programme Inefficiencies in fisheries science and governance produce regulatory environments that can lag rapidly changing northwest Atlantic marine ecosystems and fisheries by 2–5 years (Hennessey and Healey, 2000; Pinsky and Fogarty, 2012). This pilot programme uses sustained embedding of collaborative research and ecosystem scale hindcasting and now-casting within operational fisheries, with the goal of developing products accounting for socioecological change in fishery assessments and management (NEFSC, 2014, 2018; Turner et al., 2017). Fishery and ocean data and ocean model output are used in science–industry partnerships that co-develop, evaluate, and refine habitat nowcast models describing occupancy dynamics of species and fishing fleets. The primary goal is to develop models that can be applied in population assessments and tactical management to account for the effects of changing habitat dynamics on observed abundances, fishery landings, and productivity. The programme is also designed to increase fishing precision by reducing costs of harvesting quotas, including externalities such as bycatch and habitat impacts. Sustained engagement with industry experts provides important collateral benefits including the timely transfer of information about the socioecological dimensions currently impacting specific fisheries that are required for practical ecosystem-based fisheries assessment and management. Funding: Research agency Institution: NOAA Stakeholders: Fishing Industry Northeast Pacific Ocean ecological forecasts In the northeast Pacific Ocean, marine resource managers at the state, federal, and tribal levels make decisions on a weekly to quarterly basis, and fishers operate on a similar timeframe. To determine the potential of a support tool for these efforts, a seasonal forecast system known as J-SCOPE (JISAO’s Seasonal Coastal Ocean Prediction of the Ecosystem) has been developed (Kaplan et al., 2016; Siedlecki et al., 2016). This system features dynamical downscaling of regional ocean conditions in Washington and Oregon waters using a combination of a high-resolution regional model with biogeochemistry and forecasts from NOAA’s Climate Forecast System). Model performance and predictability have been determined for sea surface temperature, bottom temperature, bottom oxygen, pH, and aragonite saturation state through model hindcasts, reforecast, and forecast comparisons with observations. Results indicate J-SCOPE forecasts have measurable skill on seasonal timescales (Siedlecki et al., 2016). Funding: Research agency Institution: NOAA Stakeholders: Fishing Industry and Managers Delaware Bay Atlantic sturgeon by-catch forecasts Atlantic sturgeon (Acipenser oxyrinchus) is a long-lived anadromous species found on the east coast of North America (Vladykov and Greeley, 1963). Overfishing for caviar and flesh in the late 19th and early 20th century (Cobb, 1900; Borodin, 1925; Smith and Clugston, 1997), combined with habitat loss and degradation, severely diminished Atlantic sturgeon populations, and there has been little to no recovery despite a moratorium on directed fishing and improved water quality (Billard and Lecointre, 2000). The Delaware River and Bay historically supported the largest spawning population of Atlantic sturgeon in the world as well as the largest fishery. The Delaware River Atlantic sturgeon fishery was short lived; peak landings dropped more than 90% by the turn of the century (Cobb, 1900). Although directed harvest of Atlantic sturgeon ended in 1998, the results of historic overharvest, coupled with habitat change and ongoing issues of bycatch mortality, have resulted in a >99% decline from historic abundance of 360 000 spawning adults (Secor and Waldman, 1999) to <300 spawning adults annually (ASSRT, 2007). As a result, the National Marine Fisheries Service listed Atlantic sturgeon under the Endangered Species Act (ESA) on 6 April 2012 (United States Office of the Federal Registry, 2012) with incidental bycatch (Stein et al., 2004) and vessel strikes (Simpson and Fox, 2009) identified as risk factors for the Delaware River. The ESA listing has the potential to have major impacts on commercial fisheries, shipping, and other industries that interact with Atlantic sturgeon during their coastal migration. These fishers are also motivated to avoid Atlantic sturgeon because interactions severely damage legal fishing gear, resulting in costly downtime. This action created the need for a short-term forecasting system to alert fishers in the Delaware Bay of their potential risk of interacting with an Atlantic sturgeon. The model was developed by fusing remote sensed data and historic acoustic telemetry observations (Breece et al., 2017), and distributes a nowcast, 1, 2, and 3 days statistical forecasts. Recent satellite observations are not always available to constrain the model, so forecasts are flagged with a warning for users. Forecasts are distributed daily via web applications, and via SMS text messages to users. Funding: University Institution: NGO/Government/University Stakeholders: Fishing Industry Salmon aquaculture forecasts in Tasmania Dynamical seasonal forecasts to predict water temperatures for south-east Tasmanian Atlantic salmon (Salmo salar) farm sites several months into the future are used to manage production risk (Spillman and Hobday, 2014). High summer temperatures pose a significant risk to production systems of these farms. Based on 20 years of historical validation, the model shows useful skill for all months of the year at lead-times of 0–1 months. Model skill is highest when forecasting for winter months, and lowest for December and January predictions. The poorer performance in summer is due to increased variability due to the convergence of several ocean currents offshore from the salmon farming region. Accuracy of probabilistic forecasts exceeds 80% for all months at lead-time 0 month for the upper tercile (warmest 33% of values) and exceeds 50% at a lead-time of 3 months. Industry engagement is high, and delivery of forecasts is ongoing, supplemented by industry–scientist discussions (Hobday et al., 2016). Funding: Industry Institution: CSIRO/Bureau of Meteorology Stakeholders: Salmon Farmers Recreational dolphinfish forecasts in eastern Australia A seasonal forecast of the habitat and density of dolphinfish (Coryphaena hippurus), based on sea surface temperatures, was developed for the east coast of New South Wales (NSW) Australia (Brodie et al., 2017). Two prototype forecast products were created; geographic spatial forecasts of dolphinfish habitat and a latitudinal summary identifying the location of fish density peaks. The less detailed latitudinal summary was designed to limit the resolution of habitat information to prevent potential resource over-exploitation by fishers in the absence of total catch controls. The dolphinfish habitat forecast was accurate at the start of the annual dolphinfish migration in NSW (December) but other months (January–May) showed poor performance due to spatial and temporal variability in the catch data used in model validation. Habitat forecasts for December were useful up to 5 months ahead, with performance decreasing as forecasts were made further into the future. Funding: Researchers (University PhD) Institution: CSIRO/Bureau of Meteorology/UNSW Stakeholders: Recreational Fishers/Managers Commercial tuna forecasts in the Great Australia Bight A habitat forecast system based on a seasonal ocean model and electronic tagging data was developed in collaboration with the Southern Bluefin Tuna (SBT, Thunnus maccoyii) fishery in Australia to overcome challenges posed by novel environmental conditions (Eveson et al., 2015). A dramatic change in the distribution of SBT compromised the ability of the fishery to efficiently locate and harvest the species. In partnership with industry representatives, a seasonal forecasting system was implemented to project the likely distribution of SBT several months into the future (Eveson et al., 2015). Forecasts are delivered daily via an industry-specific website, which has assisted fishers to efficiently catch SBT under variable climatic conditions. There is a cap on catches so this system does not contribute to over-exploitation. Funding: Industry/FRDC Institution: CSIRO/Bureau of Meteorology Stakeholders: Fishing industry Lobster forecasts in Maine The American lobster (Homarus americanus) fishery is currently the highest-valued commercial fishery in the United States, worth nearly $670 million in landed value in 2016. Over 80% of the value is landed in the state of Maine. The 2012 Northwest Atlantic heat wave disrupted this fishery by prompting early warming of the ocean and an earlier-than-normal large influx of lobster landings. Since readiness for the product was not aligned throughout the supply chain, a glut of lobsters developed, resulting in a price collapse (Mills et al., 2013). At that time, managers and key industry members in Maine questioned whether the early onset of the high-landings period could have been predicted, with the expectation that such information would give valuable early notice for the fishery and supply chain to prepare appropriately for the upcoming season. A regression-based seasonal forecast was developed to provide the expected timing of the summer uptick in landings (June–July) based on ocean temperatures in March–April (Mills et al., 2017). This forecast was formally issued to the industry in 2015 and 2016 via a public website. While it proved technically reliable at predicting the summer uptick timing, it also created unexpected challenges and was not perceived as useful to the industry (Pershing et al., 2018). Funding: Research agency Institution: GMRI Stakeholders: Fishing Industry/Managers Northwest Atlantic industry–science collaborative research programme Inefficiencies in fisheries science and governance produce regulatory environments that can lag rapidly changing northwest Atlantic marine ecosystems and fisheries by 2–5 years (Hennessey and Healey, 2000; Pinsky and Fogarty, 2012). This pilot programme uses sustained embedding of collaborative research and ecosystem scale hindcasting and now-casting within operational fisheries, with the goal of developing products accounting for socioecological change in fishery assessments and management (NEFSC, 2014, 2018; Turner et al., 2017). Fishery and ocean data and ocean model output are used in science–industry partnerships that co-develop, evaluate, and refine habitat nowcast models describing occupancy dynamics of species and fishing fleets. The primary goal is to develop models that can be applied in population assessments and tactical management to account for the effects of changing habitat dynamics on observed abundances, fishery landings, and productivity. The programme is also designed to increase fishing precision by reducing costs of harvesting quotas, including externalities such as bycatch and habitat impacts. Sustained engagement with industry experts provides important collateral benefits including the timely transfer of information about the socioecological dimensions currently impacting specific fisheries that are required for practical ecosystem-based fisheries assessment and management. Funding: Research agency Institution: NOAA Stakeholders: Fishing Industry Northeast Pacific Ocean ecological forecasts In the northeast Pacific Ocean, marine resource managers at the state, federal, and tribal levels make decisions on a weekly to quarterly basis, and fishers operate on a similar timeframe. To determine the potential of a support tool for these efforts, a seasonal forecast system known as J-SCOPE (JISAO’s Seasonal Coastal Ocean Prediction of the Ecosystem) has been developed (Kaplan et al., 2016; Siedlecki et al., 2016). This system features dynamical downscaling of regional ocean conditions in Washington and Oregon waters using a combination of a high-resolution regional model with biogeochemistry and forecasts from NOAA’s Climate Forecast System). Model performance and predictability have been determined for sea surface temperature, bottom temperature, bottom oxygen, pH, and aragonite saturation state through model hindcasts, reforecast, and forecast comparisons with observations. Results indicate J-SCOPE forecasts have measurable skill on seasonal timescales (Siedlecki et al., 2016). Funding: Research agency Institution: NOAA Stakeholders: Fishing Industry and Managers Delaware Bay Atlantic sturgeon by-catch forecasts Atlantic sturgeon (Acipenser oxyrinchus) is a long-lived anadromous species found on the east coast of North America (Vladykov and Greeley, 1963). Overfishing for caviar and flesh in the late 19th and early 20th century (Cobb, 1900; Borodin, 1925; Smith and Clugston, 1997), combined with habitat loss and degradation, severely diminished Atlantic sturgeon populations, and there has been little to no recovery despite a moratorium on directed fishing and improved water quality (Billard and Lecointre, 2000). The Delaware River and Bay historically supported the largest spawning population of Atlantic sturgeon in the world as well as the largest fishery. The Delaware River Atlantic sturgeon fishery was short lived; peak landings dropped more than 90% by the turn of the century (Cobb, 1900). Although directed harvest of Atlantic sturgeon ended in 1998, the results of historic overharvest, coupled with habitat change and ongoing issues of bycatch mortality, have resulted in a >99% decline from historic abundance of 360 000 spawning adults (Secor and Waldman, 1999) to <300 spawning adults annually (ASSRT, 2007). As a result, the National Marine Fisheries Service listed Atlantic sturgeon under the Endangered Species Act (ESA) on 6 April 2012 (United States Office of the Federal Registry, 2012) with incidental bycatch (Stein et al., 2004) and vessel strikes (Simpson and Fox, 2009) identified as risk factors for the Delaware River. The ESA listing has the potential to have major impacts on commercial fisheries, shipping, and other industries that interact with Atlantic sturgeon during their coastal migration. These fishers are also motivated to avoid Atlantic sturgeon because interactions severely damage legal fishing gear, resulting in costly downtime. This action created the need for a short-term forecasting system to alert fishers in the Delaware Bay of their potential risk of interacting with an Atlantic sturgeon. The model was developed by fusing remote sensed data and historic acoustic telemetry observations (Breece et al., 2017), and distributes a nowcast, 1, 2, and 3 days statistical forecasts. Recent satellite observations are not always available to constrain the model, so forecasts are flagged with a warning for users. Forecasts are distributed daily via web applications, and via SMS text messages to users. Funding: University Institution: NGO/Government/University Stakeholders: Fishing Industry Open in new tab Table 1. Summary of ecological forecasting examples describing the context for each system. Salmon aquaculture forecasts in Tasmania Dynamical seasonal forecasts to predict water temperatures for south-east Tasmanian Atlantic salmon (Salmo salar) farm sites several months into the future are used to manage production risk (Spillman and Hobday, 2014). High summer temperatures pose a significant risk to production systems of these farms. Based on 20 years of historical validation, the model shows useful skill for all months of the year at lead-times of 0–1 months. Model skill is highest when forecasting for winter months, and lowest for December and January predictions. The poorer performance in summer is due to increased variability due to the convergence of several ocean currents offshore from the salmon farming region. Accuracy of probabilistic forecasts exceeds 80% for all months at lead-time 0 month for the upper tercile (warmest 33% of values) and exceeds 50% at a lead-time of 3 months. Industry engagement is high, and delivery of forecasts is ongoing, supplemented by industry–scientist discussions (Hobday et al., 2016). Funding: Industry Institution: CSIRO/Bureau of Meteorology Stakeholders: Salmon Farmers Recreational dolphinfish forecasts in eastern Australia A seasonal forecast of the habitat and density of dolphinfish (Coryphaena hippurus), based on sea surface temperatures, was developed for the east coast of New South Wales (NSW) Australia (Brodie et al., 2017). Two prototype forecast products were created; geographic spatial forecasts of dolphinfish habitat and a latitudinal summary identifying the location of fish density peaks. The less detailed latitudinal summary was designed to limit the resolution of habitat information to prevent potential resource over-exploitation by fishers in the absence of total catch controls. The dolphinfish habitat forecast was accurate at the start of the annual dolphinfish migration in NSW (December) but other months (January–May) showed poor performance due to spatial and temporal variability in the catch data used in model validation. Habitat forecasts for December were useful up to 5 months ahead, with performance decreasing as forecasts were made further into the future. Funding: Researchers (University PhD) Institution: CSIRO/Bureau of Meteorology/UNSW Stakeholders: Recreational Fishers/Managers Commercial tuna forecasts in the Great Australia Bight A habitat forecast system based on a seasonal ocean model and electronic tagging data was developed in collaboration with the Southern Bluefin Tuna (SBT, Thunnus maccoyii) fishery in Australia to overcome challenges posed by novel environmental conditions (Eveson et al., 2015). A dramatic change in the distribution of SBT compromised the ability of the fishery to efficiently locate and harvest the species. In partnership with industry representatives, a seasonal forecasting system was implemented to project the likely distribution of SBT several months into the future (Eveson et al., 2015). Forecasts are delivered daily via an industry-specific website, which has assisted fishers to efficiently catch SBT under variable climatic conditions. There is a cap on catches so this system does not contribute to over-exploitation. Funding: Industry/FRDC Institution: CSIRO/Bureau of Meteorology Stakeholders: Fishing industry Lobster forecasts in Maine The American lobster (Homarus americanus) fishery is currently the highest-valued commercial fishery in the United States, worth nearly $670 million in landed value in 2016. Over 80% of the value is landed in the state of Maine. The 2012 Northwest Atlantic heat wave disrupted this fishery by prompting early warming of the ocean and an earlier-than-normal large influx of lobster landings. Since readiness for the product was not aligned throughout the supply chain, a glut of lobsters developed, resulting in a price collapse (Mills et al., 2013). At that time, managers and key industry members in Maine questioned whether the early onset of the high-landings period could have been predicted, with the expectation that such information would give valuable early notice for the fishery and supply chain to prepare appropriately for the upcoming season. A regression-based seasonal forecast was developed to provide the expected timing of the summer uptick in landings (June–July) based on ocean temperatures in March–April (Mills et al., 2017). This forecast was formally issued to the industry in 2015 and 2016 via a public website. While it proved technically reliable at predicting the summer uptick timing, it also created unexpected challenges and was not perceived as useful to the industry (Pershing et al., 2018). Funding: Research agency Institution: GMRI Stakeholders: Fishing Industry/Managers Northwest Atlantic industry–science collaborative research programme Inefficiencies in fisheries science and governance produce regulatory environments that can lag rapidly changing northwest Atlantic marine ecosystems and fisheries by 2–5 years (Hennessey and Healey, 2000; Pinsky and Fogarty, 2012). This pilot programme uses sustained embedding of collaborative research and ecosystem scale hindcasting and now-casting within operational fisheries, with the goal of developing products accounting for socioecological change in fishery assessments and management (NEFSC, 2014, 2018; Turner et al., 2017). Fishery and ocean data and ocean model output are used in science–industry partnerships that co-develop, evaluate, and refine habitat nowcast models describing occupancy dynamics of species and fishing fleets. The primary goal is to develop models that can be applied in population assessments and tactical management to account for the effects of changing habitat dynamics on observed abundances, fishery landings, and productivity. The programme is also designed to increase fishing precision by reducing costs of harvesting quotas, including externalities such as bycatch and habitat impacts. Sustained engagement with industry experts provides important collateral benefits including the timely transfer of information about the socioecological dimensions currently impacting specific fisheries that are required for practical ecosystem-based fisheries assessment and management. Funding: Research agency Institution: NOAA Stakeholders: Fishing Industry Northeast Pacific Ocean ecological forecasts In the northeast Pacific Ocean, marine resource managers at the state, federal, and tribal levels make decisions on a weekly to quarterly basis, and fishers operate on a similar timeframe. To determine the potential of a support tool for these efforts, a seasonal forecast system known as J-SCOPE (JISAO’s Seasonal Coastal Ocean Prediction of the Ecosystem) has been developed (Kaplan et al., 2016; Siedlecki et al., 2016). This system features dynamical downscaling of regional ocean conditions in Washington and Oregon waters using a combination of a high-resolution regional model with biogeochemistry and forecasts from NOAA’s Climate Forecast System). Model performance and predictability have been determined for sea surface temperature, bottom temperature, bottom oxygen, pH, and aragonite saturation state through model hindcasts, reforecast, and forecast comparisons with observations. Results indicate J-SCOPE forecasts have measurable skill on seasonal timescales (Siedlecki et al., 2016). Funding: Research agency Institution: NOAA Stakeholders: Fishing Industry and Managers Delaware Bay Atlantic sturgeon by-catch forecasts Atlantic sturgeon (Acipenser oxyrinchus) is a long-lived anadromous species found on the east coast of North America (Vladykov and Greeley, 1963). Overfishing for caviar and flesh in the late 19th and early 20th century (Cobb, 1900; Borodin, 1925; Smith and Clugston, 1997), combined with habitat loss and degradation, severely diminished Atlantic sturgeon populations, and there has been little to no recovery despite a moratorium on directed fishing and improved water quality (Billard and Lecointre, 2000). The Delaware River and Bay historically supported the largest spawning population of Atlantic sturgeon in the world as well as the largest fishery. The Delaware River Atlantic sturgeon fishery was short lived; peak landings dropped more than 90% by the turn of the century (Cobb, 1900). Although directed harvest of Atlantic sturgeon ended in 1998, the results of historic overharvest, coupled with habitat change and ongoing issues of bycatch mortality, have resulted in a >99% decline from historic abundance of 360 000 spawning adults (Secor and Waldman, 1999) to <300 spawning adults annually (ASSRT, 2007). As a result, the National Marine Fisheries Service listed Atlantic sturgeon under the Endangered Species Act (ESA) on 6 April 2012 (United States Office of the Federal Registry, 2012) with incidental bycatch (Stein et al., 2004) and vessel strikes (Simpson and Fox, 2009) identified as risk factors for the Delaware River. The ESA listing has the potential to have major impacts on commercial fisheries, shipping, and other industries that interact with Atlantic sturgeon during their coastal migration. These fishers are also motivated to avoid Atlantic sturgeon because interactions severely damage legal fishing gear, resulting in costly downtime. This action created the need for a short-term forecasting system to alert fishers in the Delaware Bay of their potential risk of interacting with an Atlantic sturgeon. The model was developed by fusing remote sensed data and historic acoustic telemetry observations (Breece et al., 2017), and distributes a nowcast, 1, 2, and 3 days statistical forecasts. Recent satellite observations are not always available to constrain the model, so forecasts are flagged with a warning for users. Forecasts are distributed daily via web applications, and via SMS text messages to users. Funding: University Institution: NGO/Government/University Stakeholders: Fishing Industry Salmon aquaculture forecasts in Tasmania Dynamical seasonal forecasts to predict water temperatures for south-east Tasmanian Atlantic salmon (Salmo salar) farm sites several months into the future are used to manage production risk (Spillman and Hobday, 2014). High summer temperatures pose a significant risk to production systems of these farms. Based on 20 years of historical validation, the model shows useful skill for all months of the year at lead-times of 0–1 months. Model skill is highest when forecasting for winter months, and lowest for December and January predictions. The poorer performance in summer is due to increased variability due to the convergence of several ocean currents offshore from the salmon farming region. Accuracy of probabilistic forecasts exceeds 80% for all months at lead-time 0 month for the upper tercile (warmest 33% of values) and exceeds 50% at a lead-time of 3 months. Industry engagement is high, and delivery of forecasts is ongoing, supplemented by industry–scientist discussions (Hobday et al., 2016). Funding: Industry Institution: CSIRO/Bureau of Meteorology Stakeholders: Salmon Farmers Recreational dolphinfish forecasts in eastern Australia A seasonal forecast of the habitat and density of dolphinfish (Coryphaena hippurus), based on sea surface temperatures, was developed for the east coast of New South Wales (NSW) Australia (Brodie et al., 2017). Two prototype forecast products were created; geographic spatial forecasts of dolphinfish habitat and a latitudinal summary identifying the location of fish density peaks. The less detailed latitudinal summary was designed to limit the resolution of habitat information to prevent potential resource over-exploitation by fishers in the absence of total catch controls. The dolphinfish habitat forecast was accurate at the start of the annual dolphinfish migration in NSW (December) but other months (January–May) showed poor performance due to spatial and temporal variability in the catch data used in model validation. Habitat forecasts for December were useful up to 5 months ahead, with performance decreasing as forecasts were made further into the future. Funding: Researchers (University PhD) Institution: CSIRO/Bureau of Meteorology/UNSW Stakeholders: Recreational Fishers/Managers Commercial tuna forecasts in the Great Australia Bight A habitat forecast system based on a seasonal ocean model and electronic tagging data was developed in collaboration with the Southern Bluefin Tuna (SBT, Thunnus maccoyii) fishery in Australia to overcome challenges posed by novel environmental conditions (Eveson et al., 2015). A dramatic change in the distribution of SBT compromised the ability of the fishery to efficiently locate and harvest the species. In partnership with industry representatives, a seasonal forecasting system was implemented to project the likely distribution of SBT several months into the future (Eveson et al., 2015). Forecasts are delivered daily via an industry-specific website, which has assisted fishers to efficiently catch SBT under variable climatic conditions. There is a cap on catches so this system does not contribute to over-exploitation. Funding: Industry/FRDC Institution: CSIRO/Bureau of Meteorology Stakeholders: Fishing industry Lobster forecasts in Maine The American lobster (Homarus americanus) fishery is currently the highest-valued commercial fishery in the United States, worth nearly $670 million in landed value in 2016. Over 80% of the value is landed in the state of Maine. The 2012 Northwest Atlantic heat wave disrupted this fishery by prompting early warming of the ocean and an earlier-than-normal large influx of lobster landings. Since readiness for the product was not aligned throughout the supply chain, a glut of lobsters developed, resulting in a price collapse (Mills et al., 2013). At that time, managers and key industry members in Maine questioned whether the early onset of the high-landings period could have been predicted, with the expectation that such information would give valuable early notice for the fishery and supply chain to prepare appropriately for the upcoming season. A regression-based seasonal forecast was developed to provide the expected timing of the summer uptick in landings (June–July) based on ocean temperatures in March–April (Mills et al., 2017). This forecast was formally issued to the industry in 2015 and 2016 via a public website. While it proved technically reliable at predicting the summer uptick timing, it also created unexpected challenges and was not perceived as useful to the industry (Pershing et al., 2018). Funding: Research agency Institution: GMRI Stakeholders: Fishing Industry/Managers Northwest Atlantic industry–science collaborative research programme Inefficiencies in fisheries science and governance produce regulatory environments that can lag rapidly changing northwest Atlantic marine ecosystems and fisheries by 2–5 years (Hennessey and Healey, 2000; Pinsky and Fogarty, 2012). This pilot programme uses sustained embedding of collaborative research and ecosystem scale hindcasting and now-casting within operational fisheries, with the goal of developing products accounting for socioecological change in fishery assessments and management (NEFSC, 2014, 2018; Turner et al., 2017). Fishery and ocean data and ocean model output are used in science–industry partnerships that co-develop, evaluate, and refine habitat nowcast models describing occupancy dynamics of species and fishing fleets. The primary goal is to develop models that can be applied in population assessments and tactical management to account for the effects of changing habitat dynamics on observed abundances, fishery landings, and productivity. The programme is also designed to increase fishing precision by reducing costs of harvesting quotas, including externalities such as bycatch and habitat impacts. Sustained engagement with industry experts provides important collateral benefits including the timely transfer of information about the socioecological dimensions currently impacting specific fisheries that are required for practical ecosystem-based fisheries assessment and management. Funding: Research agency Institution: NOAA Stakeholders: Fishing Industry Northeast Pacific Ocean ecological forecasts In the northeast Pacific Ocean, marine resource managers at the state, federal, and tribal levels make decisions on a weekly to quarterly basis, and fishers operate on a similar timeframe. To determine the potential of a support tool for these efforts, a seasonal forecast system known as J-SCOPE (JISAO’s Seasonal Coastal Ocean Prediction of the Ecosystem) has been developed (Kaplan et al., 2016; Siedlecki et al., 2016). This system features dynamical downscaling of regional ocean conditions in Washington and Oregon waters using a combination of a high-resolution regional model with biogeochemistry and forecasts from NOAA’s Climate Forecast System). Model performance and predictability have been determined for sea surface temperature, bottom temperature, bottom oxygen, pH, and aragonite saturation state through model hindcasts, reforecast, and forecast comparisons with observations. Results indicate J-SCOPE forecasts have measurable skill on seasonal timescales (Siedlecki et al., 2016). Funding: Research agency Institution: NOAA Stakeholders: Fishing Industry and Managers Delaware Bay Atlantic sturgeon by-catch forecasts Atlantic sturgeon (Acipenser oxyrinchus) is a long-lived anadromous species found on the east coast of North America (Vladykov and Greeley, 1963). Overfishing for caviar and flesh in the late 19th and early 20th century (Cobb, 1900; Borodin, 1925; Smith and Clugston, 1997), combined with habitat loss and degradation, severely diminished Atlantic sturgeon populations, and there has been little to no recovery despite a moratorium on directed fishing and improved water quality (Billard and Lecointre, 2000). The Delaware River and Bay historically supported the largest spawning population of Atlantic sturgeon in the world as well as the largest fishery. The Delaware River Atlantic sturgeon fishery was short lived; peak landings dropped more than 90% by the turn of the century (Cobb, 1900). Although directed harvest of Atlantic sturgeon ended in 1998, the results of historic overharvest, coupled with habitat change and ongoing issues of bycatch mortality, have resulted in a >99% decline from historic abundance of 360 000 spawning adults (Secor and Waldman, 1999) to <300 spawning adults annually (ASSRT, 2007). As a result, the National Marine Fisheries Service listed Atlantic sturgeon under the Endangered Species Act (ESA) on 6 April 2012 (United States Office of the Federal Registry, 2012) with incidental bycatch (Stein et al., 2004) and vessel strikes (Simpson and Fox, 2009) identified as risk factors for the Delaware River. The ESA listing has the potential to have major impacts on commercial fisheries, shipping, and other industries that interact with Atlantic sturgeon during their coastal migration. These fishers are also motivated to avoid Atlantic sturgeon because interactions severely damage legal fishing gear, resulting in costly downtime. This action created the need for a short-term forecasting system to alert fishers in the Delaware Bay of their potential risk of interacting with an Atlantic sturgeon. The model was developed by fusing remote sensed data and historic acoustic telemetry observations (Breece et al., 2017), and distributes a nowcast, 1, 2, and 3 days statistical forecasts. Recent satellite observations are not always available to constrain the model, so forecasts are flagged with a warning for users. Forecasts are distributed daily via web applications, and via SMS text messages to users. Funding: University Institution: NGO/Government/University Stakeholders: Fishing Industry Open in new tab Table 2. Ethical issues encountered in the scoping, development, delivery, and evaluation phases of ecological forecasts for marine resources across seven case studies in three domains (X) as described in Table 1. Example (and domain) Scoping Development Delivery Evaluation Conflicts of interest Ecosystem health Skill assessment (inadequate) Representation of uncertainty (inadequate) Delivery of products Engagement and education Delivery failure Equity for users Unintended consequences Review of performance of the whole system 1. Tasmanian salmon (aquaculture) X X X X CC? 2. Eastern Australia dolphinfish (fisheries) X X X 3. Great Australia Bight tuna (fisheries) X X 4. Maine lobster (fisheries) X X X X 5. Northwest Atlantic fishers (fisheries) X X X X X X 6. Northeast Pacific environments (fisheries) X X X 7. Delaware Bay sturgeon (conservation) X X X X X Example (and domain) Scoping Development Delivery Evaluation Conflicts of interest Ecosystem health Skill assessment (inadequate) Representation of uncertainty (inadequate) Delivery of products Engagement and education Delivery failure Equity for users Unintended consequences Review of performance of the whole system 1. Tasmanian salmon (aquaculture) X X X X CC? 2. Eastern Australia dolphinfish (fisheries) X X X 3. Great Australia Bight tuna (fisheries) X X 4. Maine lobster (fisheries) X X X X 5. Northwest Atlantic fishers (fisheries) X X X X X X 6. Northeast Pacific environments (fisheries) X X X 7. Delaware Bay sturgeon (conservation) X X X X X Other issues may have been possible, but were not evident due to circumstance or practise. Open in new tab Table 2. Ethical issues encountered in the scoping, development, delivery, and evaluation phases of ecological forecasts for marine resources across seven case studies in three domains (X) as described in Table 1. Example (and domain) Scoping Development Delivery Evaluation Conflicts of interest Ecosystem health Skill assessment (inadequate) Representation of uncertainty (inadequate) Delivery of products Engagement and education Delivery failure Equity for users Unintended consequences Review of performance of the whole system 1. Tasmanian salmon (aquaculture) X X X X CC? 2. Eastern Australia dolphinfish (fisheries) X X X 3. Great Australia Bight tuna (fisheries) X X 4. Maine lobster (fisheries) X X X X 5. Northwest Atlantic fishers (fisheries) X X X X X X 6. Northeast Pacific environments (fisheries) X X X 7. Delaware Bay sturgeon (conservation) X X X X X Example (and domain) Scoping Development Delivery Evaluation Conflicts of interest Ecosystem health Skill assessment (inadequate) Representation of uncertainty (inadequate) Delivery of products Engagement and education Delivery failure Equity for users Unintended consequences Review of performance of the whole system 1. Tasmanian salmon (aquaculture) X X X X CC? 2. Eastern Australia dolphinfish (fisheries) X X X 3. Great Australia Bight tuna (fisheries) X X 4. Maine lobster (fisheries) X X X X 5. Northwest Atlantic fishers (fisheries) X X X X X X 6. Northeast Pacific environments (fisheries) X X X 7. Delaware Bay sturgeon (conservation) X X X X X Other issues may have been possible, but were not evident due to circumstance or practise. Open in new tab Phase 1—Scoping The first phase of forecast development can be inwardly focused, motivated by a search for system understanding, or outwardly, in response to end user needs. Both motivations were revealed by the participants in these case studies. Issues associated with conflicts of interest and ecosystem health were recognized across the case studies. Conflicts of interest Enthusiasm for the technical challenge in developing a forecast should not ignore the perspectives of user groups that can emerge as conflicts of interest. This might result in stakeholders being co-opted to a programme that is not in their best interests. However, in the northwest Atlantic (Example 5, Table 1), industry partners willingly engaged in the pilot programme and had the time to collaborate constructively with scientists. They were prepared to apply advanced technologies to both improve ecosystem-based fisheries science and enhance fishing efficiency. Many of these fishers operate innovative, sophisticated, successful businesses and are leaders in the industry. In this case, industry collaborators are full partners in the programme and helped shape the approaches adopted and products developed, an important feature of co-production (Cvitanovic et al., 2015; Djenontin and Meadow, 2018). There may, however, be conflicts with other users—fishers involved in this programme are already “industry winners” and may be reaping additional competitive advantages associated with participation in the pilot programme, including privileged access to the environmental products. Thus, decisions regarding participants vs. non-participants represent an ethical dilemma when delivering future information (see Equity for users). Conflicts of interest can also be suggested based on the perceived viewpoints of stakeholders. For example in the Delaware Bay Atlantic sturgeon conservation programme (Example 7, Table 1), conservation groups believe that gillnet fishers may use Atlantic sturgeon forecasts to illegally target, rather than avoid, Atlantic sturgeon. Illegally targeting sturgeon is a problem in the northwest United States, and conservation groups project these same motivations onto Delaware Bay fishers, thus assuming this is the reason fishers participate in the bycatch programme. Alternatively, fishers may be nervous about participating in this programme because it could be used by regulatory bodies to develop time-area closures that would affect their livelihood, even if they are not interacting with Atlantic sturgeon. In both cases, the delivery of these spatial forecasts is seen as a potential threat because of another group’s perceived motivations. Ecosystem health A second ethical issue in the scoping phase is consideration of ecosystem health, which was an issue for the dolphinfish and northwest Atlantic forecasting programmes (Examples 2 and 5, Table 2). Modelling technologies and information may confer efficiencies upon users, such as fishers, that eliminate the spatial and/or temporal refugia prey require for maintaining their populations. The east coast Australia recreational fishery for dolphinfish has no effort cap or required reporting, and the forecast development team was conscious of not contributing to over-exploitation and actively considered options to limit impacts to ecosystem health. Initially, a latitude-only forecast was provided to limit its use as a fish finder, such that it represented general distribution only (Brodie et al., 2017). At the same time, the project team sought to explore improvements in catch management that could accompany forecast delivery and reduce risks from over-harvesting (Example 2, Table 1). Similar issues were confronted in the northwest Atlantic example (Example 5, Table 2), where it was recognized that efficiency as a result of model-guided fishing can result in declining ecosystem health. Ethical concerns about increased efficiency are being discussed with industry–science participants with two main aims. First, development of products to improve the accuracy of population assessments and support sustainable fishing is recognized as the primary goal of the industry–science partnerships. Second, efforts to increase catch efficiency that seek to reduce fishing costs must consider the externalities related to collateral damage to ecosystem services. Addressing these ethical considerations and engaging the fishing industry as full partners builds support for more accurate assessments and coherent regulations. It also provides industry partners with a deeper understanding of the science process, a greater acceptance of scientific results, as well as an increased sense of stewardship for the fish and ecosystems supporting their livelihoods. While these two examples confronted this issue, commercial “fish forecasting” services designed to enhance fisher search efficiency have existed for decades (e.g. https://atlantniro.ru, http://www.catsat.com/, https://www.roffs.com/), and we hope these providers also consider the impact on ocean environments. As ocean monitoring, modelling, and information sharing technologies rapidly advance and become available at lower costs, ethical concerns regarding ecosystem health outcomes and delivery of forecasting products should be examined more generally. Phase 2—Development The ethical issues in the development phase were associated with technical judgements about the system. If these issues are not actively considered, then the end user may be misled, or not provided with sufficient context to evaluate the value of the forecast in their decision-making. These judgements are often based on experience, but can also be made on efficiency grounds, which we consider problematic. Skill assessment An important step in developing a forecast system is to understand the model performance. One measure of performance is model skill—defined as the ability of the model to outperform climatological or persistence forecasts (Hobday et al., 2011; Hewitson et al., 2014; Stock et al., 2015). In evaluating skill, it is possible to unconsciously bias a forecast by restricting the set of models or explanatory variables, by varying the length of the sample that is fitted, by deciding to include or suppress influential observations, by focusing on short-term trends rather than long-term trends, and so on (Hewitson et al., 2014). True skill assessment in a forecast system should use forward out-of-sample test data, such that the same test is being used as will be required when the system is forecasting the (unknown) future—or a true forecast (Kaplan et al., 2016; Tommasi et al., 2017). Hindcast data sets must be long enough (e.g. 10 years or more) that the performance of the forecast system can be evaluated under a range of conditions (e.g. El Niño and La Niña periods). Use of weak skill assessments (e.g. randomly dividing data sets, short time periods) or those based on absolute values rather than anomalies can lead to inflated skill estimates that are misleading to an end user. The skill assessment should not only give the forecasters confidence in the system but it should be used to inform the uncertainties of the system—no system can forecast everything. Knowing a system’s strengths and weaknesses is vital to providing a good forecast. Strategies to improve performance include running a multi-model ensemble, applying a correction to potential bias in selected model, paying close attention to residual diagnostics, using out-of-sample validation, determining the relevant forecasting horizon, and taking into account the plausibility of the assumptions that underlie a given forecasting model. If forecast skill cannot be evaluated based on past performance due to an absence of historical data, explicit discussion with stakeholders is critical to explain potential risks in using the forecast. The Delaware Bay Atlantic sturgeon system provides both 0 and 3 days forecasts and a climatological forecast based on 15 years of observations. Because Atlantic sturgeon distribution has a strong seasonal signal, the climatological forecast sometimes performs better than the forecast based on the latest satellite observations. However, forecast skill is difficult to assess in real time, therefore it is based on comparison of past forecasts with historical in situ Atlantic sturgeon observations. Past forecast analysis efforts or re-forecasts are an essential part of building a forecast system (Kaplan et al., 2016; Siedlecki et al., 2016), but cannot replace a true forecast. Overall, the ethical issue here is to ensure that the forecast team is applying best practice for the situation, rather than adequate practice. Representation of uncertainty Forecasts are typically probabilistic—as a result presenting information on the associated uncertainty with any forecasts is complicated. As many stakeholders may be unfamiliar with representation of uncertainty, it is tempting to eliminate this confusion by discarding information on uncertainty. We consider this approach to be ethically flawed, even if it is defended on the basis of reducing complexity to enhance understanding. Complexity in communicating uncertainty confronted the project team delivering dolphinfish forecasts (Example 2, Table 2). Skill declined and uncertainty generally increased over time, but the pattern varied over the annual cycle. In the presentation of different levels of predictability for different months, the project team could not provide a consistent lead time. Forecast periods occurred when skill did not decline and uncertainty was lower at longer lead times, which could not be explained. The project team considered removing the representation of uncertainty from the forecasts being provided, but ultimately included it with considerable warning to the end user about the perceived problem. Uncertainty can be due to the use of research products, which are works in progress, as in the case of the northwest Atlantic system (Example 5, Table 1). Industry collaborators are involved in model development and evaluation and thus the programme is fully transparent about limits of resolution, accuracy, and utility of the models. Components underlying the habitat predictions (e.g. species niche models, ocean models, and observations) are evaluated individually. Predictions are also assessed qualitatively by industry collaborators and quantitatively using out of sample statistical evaluation techniques and fishery dependent and independent data. Uncertainties of models applied in stock assessments are computed and presented as required by the assessment science process. Finally, forecasts are labelled “for research only” which indicates less confidence in model results compared to operational systems. Limitations in the forecast model that result in uncertainty, are important to stress when delivering information to stakeholders. In the J-SCOPE system used in the northeast Pacific (Example 6, Table 1), the transition from the summer upwelling season to autumn conditions (that typically happens sometime in September) is not well forecasted from the April-initialized forecast. This is mainly because the fall transition is driven primarily by storm events, which has low predictive skill on seasonal time scales (Siedlecki et al., 2016). In this case, the project team explicitly communicates this in text attached to forecasts—and provides additional information on the project website where evaluation of past performance suggests this portion of the forecast should be disregarded. Storms and cloud cover also contribute to uncertainty in the Delaware Bay Atlantic sturgeon system. Storm events create large time gaps in the satellite record, therefore statistical reconstructions of satellite observations that are used to predict the occurrence of Atlantic sturgeon are poor. In this case, a forecast is still issued with a degradation warning, and users are directed to a climatological prediction as the best estimate. Phase 3—Delivery After a forecast system has been developed and tested, forecasts of upcoming conditions at a range of time scales are delivered to the user community, via a range of methods. This can create an expectation for ongoing delivery of products, requires education and engagement, raises issues around delivery failures, and can also have unintended consequences (Table 2). Delivery of products The development of a forecast system is typically a research endeavour, with a finite funding period to accomplish the work. Stakeholders involved in such projects may have expectations about the ongoing delivery of information, unless this is clearly ruled out by the project team. In development of applications in Australia, project teams have sought to build systems that can be maintained with little intervention after a project ends (Hobday et al., 2016). The ultimate solution is to pass the forecast system to an operational system, such as a national weather service (e.g. coral reef bleaching—Spillman, 2011). In the northeast Pacific (Example 6, Table 1), experimental forecasts were delivered to improve representation of uncertainty and build confidence with a user group as discussed earlier. While the experimental forecasts were visible to the public (via J-SCOPE website) or announced to a user community (e.g. bulletin or email), delivery of experimental forecasts enabled a dialogue that shaped the form of the forecast product into a useable decision support tool. These forecasts were clearly labelled as experimental. The first true forecast was issued on the website in 2013, however, scientific papers describing these forecasts were not published until 2016 (Kaplan et al., 2016; Siedlecki et al., 2016). This early delivery of forecasts—prior to peer review—was criticized by some in the scientific community, although industry welcomed the information. Unfortunately, the risk of providing a bad forecast—defined here as one that ends up proving false can be costly to the experimental system in terms of end user trust and impacts, even if skill has been explained. For the northeast Pacific system, one example occurred in early June of 2018. A seemingly widespread low oxygen event caused Dungeness crab fishers to catch dead crabs in their traps. The Ocean Observatories Initiative Coastal Endurance array real-time observations of bottom oxygen near Cape Elizabeth, WA suggested that the event began in early June and lasted for over a week. The forecasts initialized in January indicated the potential for such an event, but the subsequent April-initialized forecast did not. All prior forecast performance statistics indicated the April-initialized forecast should perform better than the January-initialized forecasts for the onset of hypoxia at this location. The project team could explain this after the event, and learned more about the forecast system, including aspects of the delivery of uncertainty through this event. While delivery of a less mature forecast system to the public domain is a reputational risk, the timescale for full scientific rigour may not be fast enough to match end user needs. The lessons learned in the process of providing true research forecasts also provide valuable feedback to the forecast team, speeding the rate of learning. Because of the lag between utility and understanding, it is essential that forecasters communicate the likelihood of improvements in understanding as well as the technical limitations in the existing forecasts through clear estimates of uncertainty during a research forecasting period. Stakeholder expectations regarding the ongoing delivery of a forecast and the performance quality of the product may also need to be explicitly managed. In the case of the Tasmanian salmon and Great Australia Bight tuna forecasts (Examples 1 and 3), a new underlying seasonal physical forecast model will replace the tested system in 2019. Should forecast delivery be discontinued for several years until new skill assessments and uncertainty treatments are resolved (scientifically correct), or should stakeholder expectations (ongoing forecast delivery) be given primacy? Forecast delivery systems must also recognize that users do not all access information in the same ways, which can raise ethical issues related to fair and equitable delivery of information. In the Delaware Bay Atlantic sturgeon system, forecast delivery was designed for both managerial and on-the-water users. The forecast is primarily a web-based mapping application featuring low-, medium-, and high-risk regions for Atlantic sturgeon interaction. However, these maps are not well transmitted to users that are out actively fishing, or those without Internet access. Therefore, SMS text messages were used to communicate to make sure that the forecast delivery system allowed access across different user types. Education and engagement Traditional education and engagement involves working with end users of forecasts to build their capacity to interpret information. An insufficient commitment to work with end users can be considered as ethically irresponsible, even if this is not the primary function of a forecast development team. Use of industry representatives or knowledge brokers can be considered to build stakeholder capacity and maintain long-term relationships (Eveson et al., 2015; Cvitanovic et al., 2016). Different levels of interest in the forecast often require a range of products, without dumbing down the messages and complexity. In the case of the tuna forecasts (Example 3, Table 1), the project team was tempted to interpolate the relatively coarse model grid to make the maps easier for stakeholders to compare to satellite-based products, before ultimately deciding this was not scientifically responsible, as it hid the true model resolution and may have led to higher confidence in the products than was warranted. Issues of scale were also confronted in the Gulf of Maine seasonal lobster forecasts (Example 4, Table 1). Users found the state-wide scale of the forecast information disconnected from their local experiences. Users found it hard to relate “normal” for their location to the state-wide “normal” start of the high-landings period and, moreover, to apply the forecasted offset from “normal” to their local experience. Based on discussions, the project team now plans to make these forecasts more spatially explicit so that there is a greater ability to act on locally relevant information. An ecological forecast is not the only source of information about future conditions, and end users should be made aware of these alternative sources of information. Climate modes and extreme events also influence the overall ocean patterns, but may not be represented well in current forecast systems. Thus, there is a need to communicate how output from a single forecast system fits the landscape of available information, and to present any contradictions that may exist. For example, the forecast package provided to Tasmanian salmon farmers (Example 1, Table 1) includes seasonal El Nino-Southern Oscillation (ENSO) forecasts based on the core model (Predictive Ocean Atmosphere Model for Australia). The predictions can be inconsistent with other models from around the world. The forecast team alerts users to this inconsistency and the potential implications, and a higher level of risk management may be used by the stakeholders until greater consistency emerges. Extreme events or processes that are not reflected in the model system can also reduce the accuracy of a single forecast. In such cases, rather than delivering a forecast as if nothing were of concern, expert interpretation can be provided to end users. Stakeholders can then look to other sources of information, such as in situ monitoring to inform their decision-making. Delivery failure A forecast team may have achieved success and industry support as a result of forecast delivery, which can be a barrier to action when there are delivery failures. In the Tasmanian salmon system (Example 1, Table 1), despite mature data delivery practices that underpinned forecast delivery, unanticipated errors in model products were uncovered in late 2016. Forecasts were not consistent with project team expectations and pointed to data assimilation problems in underlying models. These errors could not be resolved quickly, and the team grappled with the potential loss of confidence as a result of halting forecast delivery. It would have been unethical to continue forecast delivery, and despite reputational risk, delivery was halted until the errors could be resolved. A team might also proactively quantify the sensitivity of the forecast products to missing data, because dissemination streams for observed (i.e. remotely sensed or directly sampled) data can experience delays or gaps (Welch et al., 2018). This will inform the forecast team as to the data situation for which forecasts can still be delivered, or halted. Equity for users Delivery of forecasts to one group may advantage or disadvantage them relative to another, and consideration of equity arose in several case studies. The Gulf of Maine lobster experience revealed different outcomes for users across the supply chain (Example 4, Table 1). Experiences in 2016 indicated that the forecast may have influenced winners and losers in the system (Pershing et al., 2018). In 2012, early, high catches of lobsters and an unprepared supply chain meant that harvesters felt the impact of a rapid drop in price, as the value of the product brought to the docks barely covered fishing expenses. Some dealers also incurred increased costs associated with transporting and storing lobsters. This motivated the development of the forecast system. In 2016, the March forecast for expected early high catches impacted directly on the dealers as they tried to sell existing inventory and establish contracts for the remainder of the year (Pershing et al., 2018). However, if dealers cleared existing inventory (even at lower prices), it may have made space in the supply chain for product as it came in over the course of the summer, a move that may have supported the higher prices later in the season which had benefits across the industry, particularly for harvesters. This revealed trade-offs in benefits and costs associated with forecast information for these two distinct industry groups. Fishing is a competitive occupation, and maintaining confidentiality about fishing and business practices is essential for maintaining the trust between industry and scientists that is so difficult to develop and easy to lose. In the northwest Atlantic programme (Example 5, Table 1), nowcast models based on underlying information co-developed with individual industry collaborators are considered proprietary and not shared with other industry partners except in aggregate “crowd-sourced” form. Knowledge of business practices and fishery monitoring information are never shared. This can create inequity among businesses, however, new participants are welcomed. With regard to Tasmanian salmon (Example 1, Table 1), access to forecasts is now restricted to companies that pay for the service. While this is a private arrangement, other companies are aware that forecasts are possible, and they could seek involvement if they desired. Equity can be sought via education and training in the use of forecasts, and this was the approach amongst these case studies when the target group is small. However, if forecasts are made widely available, contact with all end users is impossible. The ethical solution is to be transparent and “equitable” with regard to forecast interpretation, but as with any knowledge system, some users will make more of the information than others, and there may be winners and losers at a range of time scales such that system change is needed (Bell et al., 2013). Unintended consequences Evaluation of the case studies revealed some unanticipated consequences of forecast development and delivery that do not fit within the above categories. These surprises can pose a range of challenges, for which forecast developers may be unprepared or ill-equipped to handle. Unanticipated consequences with wide reaching implications were encountered by the Gulf of Maine lobster forecast team (Example 4, Table 1) as result of anomalous years. A heatwave in 2012 created a difficult year for the industry (Mills et al., 2013) and motivated development of the forecast (Mills et al., 2017). When the forecast was initially issued, it was discussed in the media and in public venues relative to 2012, with sensational headlines such as “2015-On track for 2012 molt replay?” (Crowe, 2015), even though 2015 was expected to be a “normal” year. As the forecasts were issued weekly over a 2-month period each year (which enables users to track uncertainty in the forecast over the forecasting period), there were frequent opportunities for new media stories, that tended to highlight risks to the industry. This public discussion added stress to the industry. Industry users encouraged the forecast team to consider issuing a forecast only if an extremely early lobster molt year was expected in hopes that such a change would reduce ongoing discussion of whether a major disruption may arise in a “normal” year. In addition, in both 2015 and 2016, the lobster forecast was released in early March at the Maine Fishermen’s Forum, an important event attended by many in the lobster fishery. In 2016, this event occurred a few days before a large regional seafood exposition where many seafood purchases for the upcoming season are negotiated. Buyers were not familiar with the forecast, only with the news coverage, which suggested the industry would face a 2012-like price collapse again in 2016. This created a real price effect for dealers who needed to sell product and commit to future deliveries at the seafood exposition. The price for lobster declined in March 2016, a month in which it typically increases, after which it recovered and remained higher than expected (given the high landings) over the course of the year (Pershing et al., 2018). Clearly, disseminating the forecast through a public website led to an unanticipated price effect in this year, and stakeholders asked that forecast information be communicated only to certain segments of the industry. However, distributing the forecast in industry publications or even via a subscription service would not ensure that access to the information would remain restricted (see Equity for users). Further, non-traditional stakeholders who would not be considered part of the industry, such as culinary tour operators, also used the forecasts. Ultimately, the project team decided that information created with public grant funding should not be provided only to select users; instead, if it is issued, it should remain available to all potential users, particularly since unexpected user groups emerged once the forecasts became available. In the case of the tuna forecasting system (Example 3, Table 1), while there was a low risk to ecosystem health due to a quota system (Eveson et al., 2015), an unintended consequence was an decrease in time at sea fishing, and hence an increase in economic efficiency due to higher certainty about fish location. The project team was surprised to hear of impacts on social benefits—fish are caught faster, and as wages are higher at sea, total crew wages declined and they could be considered as losers from the forecasts. This issue may be overstated, as crew are generally employed in other activities by the fishing companies, such as working on the grow-out facilities. With regard to salmon forecasts (Example 1, Table 1), community concerns around expansion of salmon farming (see van Putten et al., 2018) has seen interest groups seeking to obtain forecasts to show that the industry is threatened by warming waters. Thus, forecasts of warm conditions, instead of helping an industry adapt (Hobday et al., 2016), could be used by others to argue against continuation of that industry. A positive unintended consequence in the northwest Atlantic system (Example 5, Table 1), was that development of nowcast models offered the opportunity for fishers and scientists to discuss empirical patterns occurring at scales finer than was known from traditional reporting systems. This had the effect of improving ecosystem understanding for both parties, and might ultimately help fishers to reduce bycatch and minimize trawl impacts. Sometimes, an improvement in information can lead to an unanticipated and rapid change in system understanding. In the same region, work with the Atlantic mackerel fleet to account for distribution shifts of adults and juvenile in fishery independent survey indices contributed to a revision of stock status from unknown to overfished. Not all stakeholders saw this as a positive outcome. Phase 4—Evaluation Review of performance is an important step in adaptive management, but has not been widely attempted for forecasts systems. We distinguish performance here from the assessment of model skill, and refer to holistic evaluation to see if the system achieved the overall goal—improved decision-making and sustainable use or conservation of marine resources. Review of performance Despite demonstrating technical skill and delivery of forecasts, two of the seven examples presented here halted delivery after a trial period, due to a range of issues. The forecast system for dolphinfish in eastern Australia was successfully trialled for 1 year with stakeholders (Brodie et al., 2017), however, the project team decided not to proceed with ongoing delivery, as they did not want to offer a “fish-finding” service without management controls to limit overfishing. This ethical decision was consistent with their agency goal of supporting sustainable fisheries. The Gulf of Maine lobster team also confronted an ethical decision regarding continuation of forecast delivery (Example 4, Table 2). As scientists working from public funds, they felt an obligation to share what was learned, which argued for continuing the forecast. As stakeholders working in a complex community of harvesters, dealers, managers, and scientists, they also recognized obligations to be constructive and to listen to feedback. This would argue for stopping the forecast. At the time of writing (2018), the team has decided to stop issuing forecasts. This decision was reached after the experience during 2016 when the forecast led to undesirable industry impacts and outcomes. Three factors drove the decision: (i) harvesters found the statewide scale of the forecast difficult to apply to their local experiences, (ii) dealers absorbed a direct price impact upon release of the 2016 forecast, and (iii) other changes in the supply chain (e.g. enhanced processing capacity, Pershing et al., 2018) made the information in the forecast less valuable. The disconnect between the scale of the forecast and the harvesters’ scale of operation was particularly problematic. It led to the mistaken perception that the forecast was inaccurate, which risked undermining other forecasting efforts. The team continues to work towards an improved forecast product that addresses the local needs of harvesters, and are also adapting the forecast methods and analyses to biological questions relevant to management decisions. It is hoped that an ongoing dialogue with the industry will shape future forecast products and plans for their communication. A final ethical issue, while not explicitly a concern in any of the examples covered here, is if the forecast programme is successful and becomes operational, it risks a degree of scientific/regulatory capture by the subset of fisherman who participated in and helped shape the programme in a manner consistent with their own business interests, to the exclusion of others (see Equity for users). They may also fail to sufficiently challenge the forecast system if trust is overdeveloped (Lacey et al., 2018) and may miss other opportunities for enhancing performance or sustainability. Scientists should continue to work with stakeholders to ensure the support systems remain fit for purpose. Principles for ethical forecasting As a result of reviewing these case studies and our experiences, we suggest a set of principles that should be considered when scoping, developing, delivering, and evaluating ecological forecasts for marine resource users. Phase 1. Scoping the forecast system 1. Conflicts of interest: Principle 1: Be open and transparent. Work with diverse stakeholders to understand their needs and concerns. Address these concerns if possible, striving for “win–wins.” Tread carefully around zero-sum situations, where a forecast advantage for one group may be a disadvantage for another. 2. Ecosystem health: Principle 2: Do not deliver forecasts that would lead to unregulated impacts on the ocean (e.g. for fisheries without clear catch limits and/or enforcement). Phase 2. Developing the forecast system 3. Skill assessment: Principle 3: Undertake best practice skill assessment that tests the true skill of a model with out-of-sample testing. In forecasting science, this involves comparing a forecasted and a hindcasted fields once the climatology has been removed, using rigourous statistics. 4. Representation of uncertainty: Principle 4: Do not ignore uncertainty. Traditionally, uncertainty is computed through an ensemble or with permutations on the initial state and provided as a percent agreement between the trajectories of the simulations. While this mostly addresses the uncertainty in the forcing into the future, the uncertainty due to model construction is not easy to incorporate objectively, and needs additional work. Provide a discussion and metrics of uncertainty that include a perspective based on model performance, and the interpretation of probabilistic forecasts. Phase 3. Forecast delivery 5. Ongoing delivery: Principle 5: Plan for and manage stakeholder expectations regarding continued delivery. Planning for and enabling a mechanism for ongoing delivery after a project ends (if possible) and engaging stakeholder representatives early can be important for ensuring a smooth transition. Ultimately, a transition to operational forecasts as delivered by national weather services should be considered. 6. Engagement and education: Principle 6: Work to improve the literacy of all stakeholders around forecast use and interpretation, particularly on skill and uncertainty. 7. Delivery failures: Principle 7: Proactively explore the impact of loss of a predictor variable in a forecast system, and be able to explain what the loss of performance is when one variable is removed. Prepare stakeholders for potential breaks in delivery, and never compromise with delivery of substandard forecast products. 8. Equity for end users: Principle 8: Be vigilant for inequity in use of forecasts between users, and the creation of winners and losers arising from provision of information. Decide when open access is warranted, and when it is not. Include stakeholders in the formulation stage to understand these risks. If risks remain, work at a scale where benefits are clear. 9. Unintended consequences: Principle 9: Scope the system context widely, seek deep domain and system knowledge, and consider scenario testing, as happens for fishery management regulations now (e.g. management strategy evaluation). Seek feedback and learn from mistakes. Phase 4. Evaluation 10. Review of performance: Principle 10: Consider the holistic outcome of forecast system—if it is not achieving the overall goals, suspend delivery and work on improving the interaction of the forecast and the context in which it operates. In time, or based on other experiences, this list may change, expand or contract, however, it serves to stimulate thinking in each of the four phases of forecast development and delivery. We suggest project teams use this as a guide when working on forecast systems—particularly as research progress allows development on multi-year timescales (Salinger et al., 2016; Tommasi et al., 2017). These discussions should take place with forecast users too—in line with principles of co-production to maximize benefits (Cvitanovic et al., 2015). The future of marine ecological forecasts Development and delivery of forecast systems for marine resource managers and users has increased over the past decade (Payne et al., 2017), and this review has summarized some non-technical challenges for development teams around the world. A range of judgements and associated ethical issues occur in each phase of forecast development and delivery, including in the final evaluation stage. These issues arise in part because forecasts are developed to assist decision-making, and decision-making carries elements of risk. Most forecast systems we reviewed here were developed as scientist-led endeavours, in regions with strong marine management systems. Forecast teams could also benefit from including local and traditional knowledge about the biological system in question (Leite and Gasalla, 2013), particularly in areas where information on species–environmental relationships may be limited. Such engagement would also offer extra opportunity to identify unwanted consequences from forecast use. Overall, we contend that the benefits from developing forecasting systems outweigh the risks. For example, engagement with users can reveal novel information about the social and economic aspects of the industry (e.g. fishery, aquaculture business) and expose stakeholders to the nature of scientific processes. Regular forecasting offers the potential for continued learning in real time. Forecasting can be a beneficial and important contributor to ecosystem-based management systems. For example, spatial forecasts can improve economic efficiency (Eveson et al., 2015), which might reduce pressures on marine ecosystems, provided catch limits are in place. Precision fishing, an analogue to precision agriculture, is expected to increase in the future and should reduce collateral damage to ecosystem components, processes and services. The principles we propose can help other developers of forecast systems avoid pitfalls encountered to date. Many of these pitfalls arise from issues that extend beyond the technical aspects that are the primary focus of forecast scientists (Hobday et al., 2016). By considering the wider context, seeking input from a range of disciplines and stakeholders when constructing forecasts, supporting forecast users, and appreciating the different perspectives on the value of a forecast, unanticipated outcomes may be reduced and the primary goal of marine ecological forecasts—to enhance ocean sustainability—will come closer to reality. Acknowledgments AJH and JRH have been supported by research funding from FRDC, AFMA and industry associations. JPM has been supported by research funding provided by the Mid Atlantic Fisheries Management Council, the NOAA Fisheries and the Environment, Habitat, and Climate Programs, and the Northeast Fisheries Science Center Cooperative Research Program. MJO has been supported by the Lenfest Ocean Program, and the NASA Ecosystem Forecasting Program. KEM and AJP have been supported by a Saltonstall-Kennedy grant from NOAA (NA16NMF4270229). SAS has been supported by a grant from the NOAA OAP and NOAA MAPP programs. We appreciate the very enthusiastic comments from four anonymous reviewers which improved this paper and gave us much to consider for future work. References Anderson C. R. , Kudela R. M. , Kahru M. , Chao Y. , Rosenfeld L. K. , Bahr F. L. , Anderson D. M. , et al. . 2016 . Initial skill assessment of the California Harmful Algae Risk Mapping (C-HARM) system . Harmful Algae , 59 : 1 – 18 . Google Scholar Crossref Search ADS PubMed WorldCat Asseng S. , McIntosh P. C. , Wang G. , Khimashia N. 2012 . Optimal N fertiliser management based on a seasonal forecast . European Journal of Agronomy , 38 : 66 – 73 . Google Scholar Crossref Search ADS WorldCat Atlantic Sturgeon Status Review Team, ASSRT. 2007 . Status Review of Atlantic Sturgeon (Acipenser oxyrinchus oxyrinchus). Report to National Marine Fisheries Service, Northeast Regional Office. Barbier M. , Reitz A. , Pabortsava K. , Wölfl A.-C. , Hahn T. , Whoriskey F. 2018 . Ethical recommendations for ocean observation . Advances in Geosciences , 45 : 343 – 361 . Google Scholar Crossref Search ADS WorldCat Bell J. D. , Ganachaud A. , Gehrke P. C. , Griffiths S. P. , Hobday A. J. , Hoegh-Guldberg O. , Johnson J. E. et al. 2013 . Mixed responses of tropical Pacific fisheries and aquaculture to climate change . Nature Climate Change , 3 : 591 – 599 . Google Scholar Crossref Search ADS WorldCat Billard R. , Lecointre G. 2000 . Biology and conservation of sturgeon and paddlefish . Reviews in Fish Biology and Fisheries , 10 : 355 – 392 . Google Scholar Crossref Search ADS WorldCat Borodin N. 1925 . Biological observations on the Atlantic sturgeon (Acipenser sturio) . Transactions of the American Fisheries Society , 55 : 184 – 190 . Google Scholar Crossref Search ADS WorldCat Breece M. W. , Fox D. A. , Dunton K. J. , Frisk M. G. , Jordaan A. , Oliver M. J. 2017 . Dynamic seascapes predict the marine occurrence of an endangered species: Atlantic sturgeon Acipenser oxyrinchus oxyrinchus . Methods in Ecology and Evolution , 75 : 562 – 571 . WorldCat Brodie S. , Hobday A. J. , Smith J. A. , Spillman C. M. , Hartog J. R. , Everett J. D. , Taylor M. D. , et al. . 2017 . Seasonal forecasting of dolphinfish distribution in eastern Australia to aid recreational fishers and managers . Deep Sea Research Part II: Topical Studies in Oceanography , 140 : 229 – 239 . Google Scholar Crossref Search ADS WorldCat Brodie S. , Jacox M. G. , Bograd S. J. , Welch H. , Dewar H. , Scales K. L. , Maxwell S. M. , et al. . 2018 . Integrating dynamic subsurface habitat metrics into species distribution models . Frontiers in Marine Science , 5 : 219 . Google Scholar Crossref Search ADS WorldCat Brown C. W. , Hood R. R. , Long W. , Jacobs J. , Ramers D. L. , Wazniak C. , Wiggert J. D. , et al. . 2013 . Ecological forecasting in Chesapeake Bay: using a mechanistic-empirical modeling approach . Journal of Marine Systems , 125 : 113 – 125 . Google Scholar Crossref Search ADS WorldCat Champion C. , Hobday A. J. , Pecl G. T. , Tracey S. , Zhang X. 2019 . Changing windows of opportunity: past and future climate-driven shifts in temporal persistence of kingfish Seriola lalandi oceanographic habitat within southeast Australian bioregions . Marine and Freshwater Research , 70 : 33 – 42 . Google Scholar Crossref Search ADS WorldCat Cobb J. N. 1900 . The sturgeon fishery of Delaware River and Bay . Report of Commissioner of Fish and Fisheries , 25 : 369 – 380 . WorldCat Crowe M. 2015 . 2015–On Track for 2012 Molt Replay? Fishermen’s Voice 20: http://www.fishermensvoice.com/archives/2015022015OnTrackFor2012MoltReplay.html. Cvitanovic C. , Hobday A. J. , van Kerkhoff L. , Wilson S. K. , Dobbs K. , Marshall N. A. 2015 . Improving knowledge exchange among scientists and decision-makers to facilitate the adaptive governance of marine resources: a review of knowledge and research needs . Ocean and Coastal Management , 112 : 25 – 35 . Google Scholar Crossref Search ADS WorldCat Cvitanovic C. , McDonald J. , Hobday A. J. 2016 . From science to action: principles for undertaking environmental research that enables knowledge exchange and evidence-based decision-making . Journal of Environmental Management , 183 : 864 – 874 . Google Scholar Crossref Search ADS PubMed WorldCat Djenontin I. N. S. , Meadow A. M. 2018 . The art of co-production of knowledge in environmental sciences and management: lessons from international practice . Environmental Management , doi: 10.1007/s00267-00018-01028-00263. WorldCat Eveson J. P. , Hobday A. J. , Hartog J. R. , Spillman C. M. , Rough K. M. 2015 . Seasonal forecasting of tuna habitat in the Great Australian Bight . Fisheries Research , 170 : 39 – 49 . Google Scholar Crossref Search ADS WorldCat Gershwin L. , Condie S. A. , Mansbridge J. V. , Richardson A. J. 2014 . Dangerous jellyfish blooms are predictable . Journal of the Royal Society Interface , 11 : 20131168. Google Scholar Crossref Search ADS WorldCat Gigerenzer G. , Garcia-Retamero R. 2017 . Cassandra’s regret: the psychology of not wanting to know . Psychological Review , 124 : 179 – 196 . Google Scholar Crossref Search ADS PubMed WorldCat Hennessey T. , Healey M. 2000 . Ludwig’s Ratchet and the collapse of New England groundfish stocks . Coastal Management , 28 : 187 – 213 . Google Scholar Crossref Search ADS WorldCat Hewitson B. C. , Daron J. , Crane R. G. , Zermoglio M. F. , Jack C. 2014 . Interrogating empirical-statistical downscaling . Climatic Change , 122 : 539 – 554 . Google Scholar Crossref Search ADS WorldCat Hobday A. J. , Hartog J. , Spillman C. , Alves O. 2011 . Seasonal forecasting of tuna habitat for dynamic spatial management . Canadian Journal of Fisheries and Aquatic Sciences , 68 : 898 – 911 . Google Scholar Crossref Search ADS WorldCat Hobday A. J. , Hartog J. R. 2014 . Dynamic Ocean features for use in Ocean Management . Oceanography , 27 : 134 – 145 . Google Scholar Crossref Search ADS WorldCat Hobday A. J. , Spillman C. M. , Eveson J. P. , Hartog J. R. 2016 . Seasonal forecasting for decision support in marine fisheries and aquaculture . Fisheries Oceanography , 25 : 45 – 56 . Google Scholar Crossref Search ADS WorldCat Hodgkinson J. A. , Hobday A. J. , Pinkard E. A. 2014 . Climate adaptation in Australia's resource-extraction industries: ready or not? Regional Environmental Change , 14 : 1663 – 1678 . Google Scholar Crossref Search ADS WorldCat Kaplan I. C. , Williams G. D. , Bond N. A. , Hermann A. J. , Siedlecki S. 2016 . Cloudy with a chance of sardines: forecasting sardine distributions using regional climate models . Fisheries Oceanography , 25 : 15 – 27 . Google Scholar Crossref Search ADS WorldCat Keyl F. , Wolff M. 2008 . Environmental variability and fisheries: what can models do? Reviews in Fish Biology and Fisheries , 18 : 273 – 299 . Google Scholar Crossref Search ADS WorldCat Lacey J. , Howden S. M. , Cvitanovic C. , Dowd A. M. 2015 . Informed adaptation: ethical considerations for adaptation researchers and decision-makers . Global Environmental Change , 32 : 200 – 210 . Google Scholar Crossref Search ADS WorldCat Lacey J. , Howden M. , Cvitanovic C. , Colvin R. M. 2018 . Understanding and managing trust at the climate science–policy interface . Nature Climate Change , 8 : 22 – 28 Google Scholar Crossref Search ADS WorldCat Leite M. C. F. , Gasalla M. A. 2013 . A method for assessing fishers’ ecological knowledge as a practical tool for ecosystem-based fisheries management: seeking consensus in Southeastern Brazil . Fisheries Research , 145 : 43 – 53 . Google Scholar Crossref Search ADS WorldCat Liu G. , Eakin C. M. , Chen M. , Kumar A. , Cour J. L. D. L. , Heron S. F. , Geiger E. F. et al. 2018 . Predicting heat stress to inform reef management: NOAA Coral Reef Watch's 4-month coral bleaching outlook . Frontiers in Marine Science , 5 : 57 . Google Scholar Crossref Search ADS WorldCat Marshall A. G. , Hudson D. , Wheeler M. C. , Alves O. , Hendon H. H. , Pook M. J. , Risbey J. S. 2014 . Intra-seasonal drivers of extreme heat over Australia in observations and POAMA-2 . Climate Dynamics , 43 : 1915 – 1937 . Google Scholar Crossref Search ADS WorldCat Merrie A. , Dunn D. C. , Metian M. , Boustany A. M. , Takei Y. , Elferink A. O. , Ota Y. et al. 2014 . An ocean of surprises—trends in human use, unexpected dynamics and governance challenges in areas beyond national jurisdiction . Global Environmental Change , 27 : 19 – 31 . Google Scholar Crossref Search ADS WorldCat Mills K. E. , Pershing A. J. , Brown C. J. , Chen Y. , Chiang F.-S. , Holland D. S. , Lehuta S. et al. 2013 . Fisheries management in a changing climate: lessons from the 2012 ocean heat wave in the Northwest Atlantic . Oceanography , 26 : 191 – 195 . Google Scholar Crossref Search ADS WorldCat Mills K. E. , Pershing A. J. , Hernández C. M. 2017 . Forecasting the seasonal timing of Maine's lobster fishery . Frontiers in Marine Science , 4 : 337 . Google Scholar Crossref Search ADS WorldCat NEFSC. 2014 . 58th Northeast Regional Stock Assessment Workshop (58th SAW) Assessment Report. US Dept Commer, Northeast Fish Sci Cent. https://www.nefsc.noaa.gov/publications/crd/crd1404/crd1404.pdf NEFSC. 2018 . 64th Northeast Regional Stock Assessment Workshop (64th SAW) Assessment Summary Report. US Dept Commer, Northeast Fish Sci Cent Ref Doc. 18-03. 27 pp. https://www.nefsc.noaa.gov/publications/crd/crd1803/ Payne M. R. , Hobday A. J. , MacKenzie B. R. , Tommasi D. , Dempsey D. P. , Fässler S. M. M. , Haynie A. C. , et al. . 2017 . Lessons from the first generation of marine ecological forecasts . Frontiers in Marine Science , 4 : 289 . Google Scholar Crossref Search ADS WorldCat Pershing, A. J., Alexander, M. A., Hernandez, C. M., Kerr, L. A., Le Bris, A., Mills, K. E., Nye, J. A. et al. . 2015 . Slow adaptation in the face of rapid warming leads to collapse of the Gulf of Maine cod fishery . Science , 350 : 809 – 812 . Crossref Search ADS PubMed WorldCat Pershing A. J. , Mills K. E. , Dayton A. M. , Franklin B. S. , Kennedy B. T. 2018 . Evidence for adaptation from the 2016 marine heatwave in the Northwest Atlantic Ocean . Oceanography , 31 : 152 – 161 . Google Scholar Crossref Search ADS WorldCat Pinsky M. L. , Fogarty M. 2012 . Lagged social–ecological responses to climate and range shifts in fisheries . Climatic Change , 115 : 883 – 891 . Google Scholar Crossref Search ADS WorldCat Pinsky M. L. , Reygondeau G. , Caddell R. , Palacios-Abrantes J. , Spijkers J. , Cheung W. W. L. 2018 . Preparing ocean governance for species on the move . Science , 360 : 1189 – 1191 . Google Scholar Crossref Search ADS PubMed WorldCat Punt A. E. , A’mar T. , Bond N. A. , Butterworth D. S. , de Moor C. L. , De Oliveira J. A. A. , Haltuch M. A. , et al. . 2014 . Fisheries management under climate and environmental uncertainty: control rules and performance simulation . ICES Journal of Marine Science , 71 : 2208 – 2220 . Google Scholar Crossref Search ADS WorldCat Ritz D. , Hobday A. J. , Montgomery J. , Ward A. 2011 . Social aggregation in the pelagic zone with special reference to fish and invertebrates . Advances in Marine Biology , 60 : 161 – 227 . Google Scholar Crossref Search ADS PubMed WorldCat Salinger J. , Hobday A. J. , Matear R. J. , O'Kane T. J. , Risbey J. S. , Dunstan P. K. , Eveson J. P. , et al. . 2016 . Decadal-scale forecasting of climate drivers for marine applications . Advances in Marine Biology , 74 : 1 – 68 . Google Scholar Crossref Search ADS PubMed WorldCat Scales K. L. , Hazen E. L. , Jacox M. G. , Castruccio F. , Maxwell S. M. , Lewison R. L. , Bograd S. J. 2018 . Fisheries bycatch risk to marine megafauna is intensified in Lagrangian coherent structures. Proceedings of the National Academy of Sciences of the United States of America , 115 : 7362 – 7367 . Secor D. H. , Waldman J. R. 1999 . Historical abundance of Delaware Bay Atlantic sturgeon and potential rate of recovery . American Fisheries Society Symposium , 23 : 203 – 216 . WorldCat Siedlecki S. , Kaplan I. C. , Hermann A. J. , Nguyen T. T. , Bond N. A. , Newton J. A. , Williams G. D. , et al. . 2016 . Experiments with Seasonal Forecasts of ocean conditions for the northern region of the California Current upwelling system . Scientific Reports , 6 : 27203. Google Scholar Crossref Search ADS PubMed WorldCat Simpson P. C. , Fox D. A. 2009 . Contemporary understanding of the Delaware River Atlantic sturgeon: survival in a highly impacted aquatic ecosystem . American Fisheries Society Symposium , 69 : 867 – 870 . WorldCat Smith T. I. J. , Clugston J. P. 1997 . Status and management of Atlantic sturgeon, Acipenser oxyrinchus, in North America . Environmental Biology of Fishes , 48 : 61 – 72 . Google Scholar Crossref Search ADS WorldCat Spillman C. 2011 . Operational real-time seasonal forecasts for coral reef management . Journal of Operational Oceanography , 4 : 13 – 22 . Google Scholar Crossref Search ADS WorldCat Spillman C. M. , Hobday A. J. 2014 . Dynamical seasonal forecasts aid salmon farm management in an ocean warming hotspot . Climate Risk Management , 1 : 25 – 38 . Google Scholar Crossref Search ADS WorldCat Stein A. B. , Friedland K. D. , Sutherland M. 2004 . Atlantic sturgeon marine bycatch and mortality on the continental shelf of the northeast United States . North American Journal of Fisheries Management , 24 : 171 – 183 . Google Scholar Crossref Search ADS WorldCat Stock C. A. , Pegion K. , Vecchi G. A. , Alexander M. A. , Tommasi D. , Bond N. A. , Fratantoni P. S. , et al. . 2015 . Seasonal sea surface temperature anomaly prediction for coastal ecosystems . Progress in Oceanography , 137 : 219 – 236 . Google Scholar Crossref Search ADS WorldCat Testa J. M. , Clark J. B. , Dennison W. C. , Donovan E. C. , Fisher A. W. , Ni W. , Parker M. , et al. . 2017 . Ecological forecasting and the science of hypoxia in Chesapeake Bay . BioScience , 67 : 614 – 626 . Google Scholar Crossref Search ADS WorldCat Tommasi D. , Stock C. , Hobday A. J. , Methot R. , Kaplan I. , Eveson P. , Holsman K. , et al. . 2017 . Managing living marine resources in a dynamic environment: the role of seasonal to decadal climate forecasts . Progress in Oceanography , 152 : 15 – 49 . Google Scholar Crossref Search ADS WorldCat Turner S. M. , Hare J. A. , Manderson J. P. , Hoey J. J. , Richardson D. E. , Sarro C. L. , Silva R. 2017 . Cooperative research to evaluate an incidental catch distribution forecast . Frontiers in Marine Science , 4 : 116. Google Scholar Crossref Search ADS WorldCat United States Office of the Federal Registry . 2012 . Endangered and threatened wildlife and plants; threatened and endangered status for distinct population segments of Atlantic Sturgeon in the Northeast Region . United States Office of the Federal Registry , 77 : 5880 – 5912 . WorldCat van Putten E. I. , Cvitanovic C. , Fulton E. , Lacey J. , Kelly R. 2018 . The emergence of social licence necessitates reforms in environmental regulation . Ecology and Society , 23 : 24. Google Scholar Crossref Search ADS WorldCat Vladykov V. D. , Greeley J. R. 1963 . Fishes of the western North Atlantic . Sears Foundation for Marine Research, Yale University , 1 , pp. 24 – 60 . WorldCat Welch H. , Hazen E. L. , Bograd S. J. , Jacox M. G. , Brodie S. , Robinson D. , Scales K. L. , et al. . 2018 . Practical considerations for operationalizing dynamic management tools . Journal of Applied Ecology , doi: 10.1111/1365-2664.13281 WorldCat Published by International Council for the Exploration of the Sea 2019. This work is written by US Government employees and is in the public domain in the US. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
Interannual variability of Gulf Stream warm-core ring interactions with the outer continental shelf and potential broad scale relationships with longfin squid (Doryteuthis pealeii) relative abundance, 1981–2004Bisagni, James, J;Nichols, Owen, C;Pettipas,, Roger
doi: 10.1093/icesjms/fsz144pmid: N/A
Abstract Observations of WCR centers and radii for a 24-year (1981–2004) period displayed strong inter-annual variability (IAV) of encounters between WCRs and the outer continental shelf between 75° and 65° W, and decreasing encounter area east of 65° W. Temporal mean WCR/shelf encounters showed two significant maxima, i.e., “hotspots”, along the outer shelf between 72° and 71° W (south of Long Island) and 69° and 68° W (south of Nantucket Shoals). Higher temporal mean WCR/shelf encounter areas were observed during July–December (warm) months of up to twice those for the January–June (cold) months west of 65° W in agreement with earlier work. The correlation between time series of WCR/shelf encounter area and annual number of discrete WCRs from 75° and 65° W is small and insignificant. Longfin squid (Doryteuthis pealeii) displays large IAV in relative abundance between 75° and 65° W within the US Northeast Shelf Large Marine Ecosystem. Correlations of the spatially averaged annual mean WCR/shelf encounter area with D. pealeii autumn relative abundance between 75° and 65° W from 1981 and 2004 were small and insignificant. The weak correlation showed that no relationship exists between D. pealeii relative abundance and our WCR/shelf encounter metric over the broad-scale shelf region from Cape Hatteras to Georges Bank, suggesting finer scale studies are warranted. Introduction Gulf Stream (GS) warm-core rings (WCRs) are large clockwise-rotating mesoscale eddies that form north of the GS from northward-extending anticyclonic GS meanders located within the slope water (SLW) region between Cape Hatteras and the Tail of the Grand Banks (Saunders, 1971). WCRs are highly dynamic hydrographic features, 60–200 km in diameter and up to 1000 m deep that are distinguishable from the surrounding SLW through their positive temperature and salinity anomalies and maintain their identity for periods of weeks to several months at a time (Richardson, 1983). Much information on their number, sizes, translational velocities, trajectories, and lifetimes has been obtained from satellite-derived infrared sea surface temperature (SST) imagery, with each WCR’s central Sargasso Sea core and surrounding GS “ring” being generally warmer than the surrounding SLW, with SST fronts separating the different water masses (Figure 1). Time-series summaries and general statistical descriptions of WCRs have been presented and show several WCRs existing at any one time within the SLW region, generally forming east of the New England seamounts, with WCR translation speeds of 2–15 cm s−1 towards the west–southwest, where they eventually dissipate or are re-absorbed by the GS (Bisagni, 1976; Lai and Richardson, 1977; Halliwell and Mooers, 1979; Fitzgerald and Chamberlin, 1983; Joyce, 1984). Figure 1. Open in new tabDownload slide Spring 1997 satellite-derived sea surface temperature (SST) image of the western North Atlantic showing the locations of two Gulf Stream warm-core rings, the Gulf Stream and its North Wall, and the Shelf Slope Front. Also shown are multiple strong SST gradients between the warm-core rings and surrounding colder slope waters, but also including both warmer Gulf Stream and colder shelf water (SHW) entrainment “streamers” (modified after Monim, 2017). Figure 1. Open in new tabDownload slide Spring 1997 satellite-derived sea surface temperature (SST) image of the western North Atlantic showing the locations of two Gulf Stream warm-core rings, the Gulf Stream and its North Wall, and the Shelf Slope Front. Also shown are multiple strong SST gradients between the warm-core rings and surrounding colder slope waters, but also including both warmer Gulf Stream and colder shelf water (SHW) entrainment “streamers” (modified after Monim, 2017). In a recent study that examined seasonal and interannual variability (IAV) of WCR formation for the 2000–2016 period, satellite-derived SST oceanographic frontal charts were used to complete a WCR census between 75° and 55°W (Monim, 2017). This work documented the dates, positions, and sizes of WCRs during their formation and dissipation over the 17-year study period. Analysis showed a clear seasonal signal with more frequent WCR formation during the months of May, June, and July, agreeing with the timing of an eddy kinetic energy maximum for the region deduced from satellite radar altimeter data (Zhai et al., 2008). Over two-thirds of the 558 WCRs examined formed between 55° and 65°W, decreasing rapidly westward, with a maximum (minimum) of 42 (22) WCRs formed during 2003 (2016), thus displaying strong IAV for the total number of rings formed per year (Monim, 2017). In earlier work, Brown et al. (1986) discuss a 10-year (1974–1983) time-series of 87 WCRs that were also studied using satellite-derived SST oceanographic frontal charts and showed a strong bi-modal distribution of WCR lifetimes (54 and 229 days) with a mean lifetime of 130 days, with all long-lived WCRs forming east of 69°W longitude. WCR surface areas showed a reduction with time and were fitted using an exponential decay model that yielded e-folding times of 89 and 223 days for the short- and long-lived WCR modes, respectively. Comparison of WCR 82B (a WCR selected for study during the WCRs experiment) with the results reported by Brown et al. (1986) showed good agreement, indicating that the kinematics of WCR 82B reported by Evans et al. (1985) are representative of a typical long-lived WCR. Events experienced by WCR 82B included interactions with outer shelf bathymetry and submarine canyons, interaction with continental shelf waters (SHW) and the GS, resulting in entrainment and production of SHW and GS “streamers” around 82B from each water type, and air–sea interaction from passing storms. Similar to many WCRs examined by earlier workers, WCR 82B eventually became trapped between the shelf and the GS as it progressed southwest towards Cape Hatteras, with the GS modifying and finally absorbing 82B, thus ending its ∼200 day lifetime (Evans et al., 1985). One of the most prominent and easily identifiable interactions between WCRs and their surrounding waters are the offshore-directed colder SHW streamers located generally along the eastern margins of WCRs (Figure 1). Cold SHW streamers often become entrained in a WCR’s clockwise circulation when it is in proximity to SHW along the outer continental shelf (Saunders, 1971), resulting in a SHW streamer where: streamer length ≫ streamer width (Evans et al., 1985). Using the dynamic method and a reference depth of 300 m, Morgan and Bishop (1977) estimated a seaward transport of 8.9 mSv (1 mSv = 1 × 103 m3 s−1) within a WCR SHW streamer (for salinity <34 PSU) located near 38.5°N latitude, 72.5°W longitude, east of the Delmarva peninsula during August 1974. This WCR-induced SHW streamer transport could account for 208 km3 year−1 of SHW flowing off-shelf assuming three WCRs per year spending 90 days each (Morgan and Bishop, 1977), or ∼8.7% of the total 2400 km3 year−1 required loss to maintain the shelf’s salt balance (Wright, 1976), and suggested by the modelled decrease in the alongshelf transport towards Cape Hatteras (Lentz, 2008). This less-saline SHW streamer extended ∼93 km seaward of the ∼180-m isobath along the eastern margin of a WCR that exhibited 5-m drogue velocities of 30–34 cm s−1 whereas in contact waters located at the ∼180-m isobath. Lagrangian currents were measured directly within an SHW streamer located along the eastern margin of WCR 80A near 37.5°N latitude, 73.5°W longitude during late summer–fall 1980 using TIROS-N-equipped satellite-tracked drogued (10 m) drifters (Bisagni, 1983). Combining the measured Lagrangian mean streamer velocity of 25 cm s−1 with satellite measurement of streamer width and an estimated streamer depth of 50 m yielded a transport of 150 mSv for the streamer, or ∼40% of the crude estimate of ca. ∼380 mSv for a WCR streamer located near 39°N, 64°W during March 1976 (Smith, 1978). Again assuming three WCRs per year spending 90 days each in contact with SHW and entraining water only 50% of the time yielded a total yearly offshore transport of 703 km3 year−1 or ∼30% of the yearly of 2400 km3 year−1 computed for the region by Wright (1976; Bisagni, 1983). Salinity observations within the Middle Atlantic Bight (MAB) show that the 2400 km3 year−1 figure may be too small to account for the noted additional along-isobath increase in mean depth-averaged salinity towards Cape Hatteras (Lentz, 2010). Chaudhuri et al. (2009) show a 68% larger value of 4035 ± 700 km3 year−1 for their MAB sub-region determined using models as described below. The frequency of occurrence of WCR SHW streamer entrainment into SLW was examined for 49 WCRs that passed south of Georges Bank during the 7-year period from 1979 to 1985 (Garfield and Evans, 1987). This work used both National Oceanic and Atmospheric Administration (NOAA) Oceanographic Analysis SST charts and satellite-derived SST imagery (Garfield and Evans, 1987). Subjective classification was used for the presence of WCR SHW streamers, with a 7-year average of only 14 ± 13% of the observable days displaying no identifiable offshore transport, whereas 1984 (1980) showed the maximum (minimum) of no identifiable offshore transport south of Georges Bank or 35% (0%) of the observable days. SHW streamers were present 69% of the time south of Georges Bank, with more than one SHW streamer active 9 ± 6% of the time, with little seasonal variation but large IAV in the presence of SHW streamers. In the largest study to date Chaudhuri et al. (2009) reported model results of WCR SHW entrainment into the SLW region located seaward of the shelf between 75° and 50°W longitude over a 22-year period (1978–1999). This modelling work relied upon WCR centre positions, radii, and orientations determined from a Bedford Institute of Oceanography (BIO) satellite-derived dataset. The three parameters provided WCR swirl velocities that were used with a relative vorticity-based WCR entrainment model and a shelf proximity model to determine when SHW would be entrained, thus forming streamers. Spatial variation and IAV of the modelled WCR SHW entrainment were then examined. A mean of 21 WCRs per year affected the study region over the 22-year period, with a minimum (maximum) of 7 (31) WCRs during 1978 (1991), similar to the results of Auer (1987). Maximal WCR occurrence and SHW entrainment was observed within their Georges Bank longitude band (70°–65°W), decreasing both eastward and westward. Total yearly SHW entrainment volume transport between 75° and 50°W increased steadily from ∼1 × 104 km3 year−1 during the early 1980s to generally > ∼5 × 104 km3 year−1 during the early 1990s, decreasing by the late 1990s. Overall, the number of WCRs was positively correlated with the state of the North Atlantic Oscillation (NAO is the dominant mode of winter climate variability over the North Atlantic because of changes in atmospheric mass between the subtropical high and polar low, causing changes in winds, storm tracks, along with changes in temperature and precipitation, see: https://www.ldeo.columbia.edu/res/pi/NAO/) using the December–March NAO winter index (NAOWI) at a lag of zero years, with positive (+) NAOWI during the 1989–1995 period corresponding to the period of generally higher WCR SHW entrainment Chaudhuri et al. (2009). A positive NAOWI for a specific year implies that the mean winter pressure difference is above average, resulting in higher winds along with colder and drier conditions over the western North Atlantic and Mediterranean regions, whereas conditions are warmer and wetter than average in northern Europe, the eastern United States, and parts of Scandinavia. Conversely, a negative NAOWI implies lower than average winter pressure difference and reversal of these patterns (Hurrell, 1995). Summarizing, there is good evidence from both short-term focused process studies and longer term compilations that transport of SHW into SLW by the WCR entrainment “streamer” process can be a substantial fraction of that needed to maintain the shelf’s overall salt balance and that IAV of SHW transport by streamers is large. Coarse resolution (∼5°) longitudinal binning of WCR/SHW interactions shows strong variability in both space and time, with the Georges Bank shelf region exhibiting maximum WCR entrainment activity from both direct observations and modelling studies. Importantly, entrainment of SHW along the eastern boundary of WCRs into SLW has been shown to reduce recruitment of 16 groundfish stocks that inhabit the continental shelf located between the MAB and the Grand Banks using an egg-plus-larval susceptibility function for each stock (Myers and Drinkwater, 1989). The notable exception was cod recruitment on Georges Bank, although Flierl and Wroblewski (1985) found that cod and haddock recruitment appeared to vary with WCR activity off Georges Bank, displaying greater (lesser) recruitment during years with only a single (multiple) WCRs. Conversely, shoreward-directed transport along the western boundary of WCRs can bring warmer WCR water (Lee and Brink, 2010), or SHW entrained along the GS northern edge at Cape Hatteras into direct contact with the outer Northeast Shelf Large Marine Ecosystem (NESLME). This transport route is consistent with observed drifter transport and observed age distributions of larval bluefish and razorfish from the southeastern US continental shelf onto the NESLME (Hare et al., 2002), based upon earlier proposed models of Hare and Cowen (1991) and Cowen et al. (1993). Here we present results from an analysis of satellite chart-derived WCR/shelf encounter observations over a 24-year period (1981–2004) using a set of the same BIO WCR data examined previously (Chaudhuri et al., 2009). In addition, we examine the relationship between IAV of the WCR/shelf encounters and IAV of longfin squid (Doryteuthis pealeii) relative abundance that has been noted from NOAA/NMFS autumn bottom-trawl survey data from the NESLME. Doryteuthis pealeii is a short-lived (∼1 year) migratory species with ecological and economic value that has displayed large IAV in relative abundance over the NESLME between Cape Hatteras and the Gulf of Maine (Dawe et al., 2007). Although IAV of short-finned squid (Illex illecebrosus) has been linked to the latitudinal position of the shelf slope front (SSF) separating SHW from SLW (Dawe et al., 2000), here we examine the relationship between WCRs’ interactions with the shelf and autumn survey-derived D. pealeii relative abundance indices. Doryteuthis pealeii undertake diel vertical migration (Brodziak and Hendrickson, 1999; Jacobson et al., 2015) and may be exposed to environmental variability throughout the water column. Large-scale (cross-shelf) migratory movement of D. pealeii is strongly associated with bottom water temperature. A winter concentration of D. pealeii on the outer edges of the continental shelf shifts inshore when SHW increase in temperature in late spring, followed by a northerly range extension in summer and offshore movement to the warmer shelf edge waters in late fall (Summers, 1969; Serchuk and Rathjen, 1974; Vovk, 1978; Black et al., 1987; Hatfield and Cadrin, 2002). A proposed mechanism that may link D. pealeii to WCRs is that shoreward on-shelf incursions of warmer WCR water and seaward off-shelf flows of colder SHW entrainments (Zhang and Partida, 2018) may affect D. pealeii thermal habitat, thus causing changes in inshore relative abundance during summer and overwintering longfin squid aggregations near the SSF during winter. Alternatively, direct WCR SHW entrainment and loss of D. pealeii into SLW may also affect shelf-wide relative abundance of D. pealeii similar to reductions in recruitment of 16 groundfish stocks attributed to WCR SHW entrainments by Myers and Drinkwater (1989). Thus, despite a large body of literature on environmental effects on cephalopod population dynamics, including recruitment (Rodhouse et al., 2014), relatively little is known about such effects on D. pealeii. Dawe et al. (2007) used time-series analysis to examine relationships between atmospheric forcing, SSF latitudinal position, bottom temperature indices, and squid abundance calculated from surveys and landings. They hypothesized that interannual variation in D. pealeii abundance is directly related to local inshore conditions, particularly temperature variations as they relate to recruitment, development time, and the “critical period” when larvae and juveniles are exposed to predation. In this study we seek to only examine the broad-scale, shelf-wide relationship between IAV of WCRs interacting with the outer continental shelf and D. pealeii relative abundance between 75° and 65°W, with finer-scale spatial and temporal analyses planned for the future. Methods and data WCR data and WCR/shelf encounter analysis Prior to the general availability of digital infrared satellite-derived SST data during the mid-to-late 1980s, the US Naval Oceanographic Office (NAVOCEANO) and NOAA produced weekly Experimental Ocean Frontal Analysis and GS Analysis charts, respectively, showing the positions of SST fronts for the western North Atlantic Ocean. Chart coverage gradually expanded eastward from 1973 to 1980, but generally contained no data east of 55°W. Beginning during 1980, NOAA began publishing Oceanographic Features Analysis (OFA) charts three times per week and included coverage eastward to the Grand Banks and the Flemish Cap. BIO obtained all charts for the January 1973 through May 1980, period and NOAA OFA charts (produced by NOAA analyst Jenifer Clark) for the period from May1980 through 29 September 1995. Starting 26 April 1996, SST frontal charts were obtained by BIO from Jenifer Clark’s “Gulf Stream” (JCGS) commercially available analysis chart product (http://users.erols.com/gulfstrm/; Figure 2) that did not extend east of 55°W from 1996 to 2004. Beginning during May 2005, and because of budgetary constraints, BIO obtained a new no-cost NAVOCEANO analysis chart product. Moreover, each chart from 1973 to present, regardless of type, was analysed by differing technicians at BIO to produce the final database containing the locations of all WCRs, along with all GS cold-core rings, the SSF and Gulf Stream North Wall positions in gridded Mercator (latitude and longitude) coordinates using a digitizing tablet. Thus, all WCR locations produced at BIO using each type of chart were digitized by BIO workers from 75°W to as far east as 50°W over the 44-year (1973–2016) period. Although the complete 44-year dataset was initially examined for this study, temporal changes in the types of oceanographic analysis charts described above and obtained by BIO since 1973 appeared to account for at least some of the variability we are seeking to understand in this study. A year-by-year depiction of the total number of different WCRs present within each 1-degree longitude bin from 75° to 50°W over the entire 44-year (1973–2016) study period shows a clear difference in signal: fewer WCRs were detected prior to 1981 and again after 2004 across all longitudes (Figure 3), with these temporal changes occurring when the types of analysis charts changed during 1980 (to NOAA OFA) and again during 2005 (with the change to the no-cost NAVOCEANO product). Thus, given these long-term changes in the primary datasets, this study will limit its analysis to the 24-year (1981–2004) “core” period, during which, each chart was produced in a consistent manner (using a Mercator map projection) by a single analyst, Jenifer Clark. Figure 2. Open in new tabDownload slide Sample Jenifer Clark Gulfstream (JCGS) chart (4 May 2015) showing locations of Gulf Stream warm-core rings, the Gulf Stream and its North Wall, and the Shelf Slope Front (after Monim, 2017). Figure 2. Open in new tabDownload slide Sample Jenifer Clark Gulfstream (JCGS) chart (4 May 2015) showing locations of Gulf Stream warm-core rings, the Gulf Stream and its North Wall, and the Shelf Slope Front (after Monim, 2017). Figure 3. Open in new tabDownload slide Number of different WCRs observed in each longitudinal bin from 75° to 50°W from 1973 to 2016 showing decreased numbers both prior to 1981 and after 2004 (outside of black box). Satellite chart data is largely missing east of 55°W prior to 1980 and from 1996 to 2004. Figure 3. Open in new tabDownload slide Number of different WCRs observed in each longitudinal bin from 75° to 50°W from 1973 to 2016 showing decreased numbers both prior to 1981 and after 2004 (outside of black box). Satellite chart data is largely missing east of 55°W prior to 1980 and from 1996 to 2004. Each chart was analysed at BIO by digitizing closed outlines of all WCRs, with each WCR realization defined as a series of vertices, with each newly observed WCR being assigned a unique identifier. The raw digitized vertices from each WCR were used to compute each ring’s area, with each WCR’s centroid calculated from the locations of the vertices. Each WCR’s area was converted to an equivalent circular area using 100 vertices and used to calculate an equivalent WCR radius, with each circular ring’s position determined by its centroid and the 100 vertices. Ring circularity was assumed because the “true” edge of a WCR can only be defined by subsurface measurements of temperature and salinity using conductivity temperature depth or expendable bathythermograph stations. Ring circularity was assumed for all WCRs reported previously by Chaudhuri et al. (2009) to compute each WCR’s swirl velocity, with each WCR’s radius calculated as the square root of the product of the semi-major and semi-minor WCR axes using an automated ellipse-fitting WCR feature model (Glenn et al., 1990; Gangopadhyay et al., 1997). Initial circularity was also assumed for modelling of WCRs along the shelf edge (Cherian and Brink, 2016,, 2018). Chart digitization errors for our study are estimated to be at least 5 km while Brown et al. (1986) found an rms error of ±15 km between four WCR centre positions derived from analysis charts and digital satellite imagery. During periods of persistent cloud cover, positions of WCRs present in any given chart were estimated based upon the most recent valid prior position. Earlier work shows that SST fronts are usually detectable over the shelf and slope by a trained analyst throughout each year, despite weak horizontal gradients above the seasonal thermocline from late spring through early fall (Halliwell and Mooers, 1979). The large number of digitized WCR observation realizations (N = 16 108) in the 24-year BIO dataset used an automated procedure to determine the magnitudes and locations of interactions between all WCRs and the outer continental shelf. Each possible WCR/shelf encounter from each WCR observation was evaluated individually using a Matlab® image processing function (polyxpoly). The automated procedure computed the existence of any valid “overlap” area between each WCR observation and a 1-degree longitude resolution 200-m isobath (from a Mercator projection), defining the seaward edge of the shelf between 75° and 50°W. A valid WCR/shelf encounter was defined as the area (area >0 km2) of the closed polygon resulting from the overlap of a circular WCR’s vertices with the 1-degree resolution 200-m isobath, thus requiring exactly two points of intersection (Figure 4). WCR/shelf encounters resulting in one point or no points of intersection occurred when a portion of the interaction was located outside of the study domain, or there was no interaction, respectively. Results show 85.2% of the WCR observations (N = 13 731) were not valid because most WCRs did not intersect the shelf. Valid WCR/shelf encounters represented 14.8% (N = 2377) of the total (Figure 5). The longitude of the midpoint of each polygon was used to assign the computed area of each WCR/shelf encounter to one of 25 longitudinal bins between 75° and 50°W, allowing compilation of WCR/shelf encounter area magnitudes and locations over the 24-year period from 1981 through 2004. Annual mean WCR/shelf encounter areas across all longitude bins were computed by dividing the total area for each bin by the number of valid WCR encounters for each corresponding bin. Both quantities were then normalized by distance between longitudes, because distances vary between successive longitude lines along the 200-m isobath. Figure 4. Open in new tabDownload slide Schematic diagram showing the overlap region (grey shading) and the required two intersection points (arrows) for a valid encounter between a WCR (grey circle) and shelf located inshore of the one degree resolution 200-m isobath (heavy solid line) on southern Georges Bank. Also shown is the shelf slope front mean position (dashed line) reported by Bisagni et al. (2009). Figure 4. Open in new tabDownload slide Schematic diagram showing the overlap region (grey shading) and the required two intersection points (arrows) for a valid encounter between a WCR (grey circle) and shelf located inshore of the one degree resolution 200-m isobath (heavy solid line) on southern Georges Bank. Also shown is the shelf slope front mean position (dashed line) reported by Bisagni et al. (2009). Figure 5. Open in new tabDownload slide All WCRs with valid WCR/shelf encounters (N = 2377) from the BIO WCR dataset between 75° and 50°W over the 24-year period from 1981 to 2004 (grey circles). Also shown are the locations of the Middle Atlantic Bight (MAB), Georges Bank (GB), Scotian Shelf (SS), Laurentian Channel (LC), and Tail of the Grand Banks (TGB). Figure 5. Open in new tabDownload slide All WCRs with valid WCR/shelf encounters (N = 2377) from the BIO WCR dataset between 75° and 50°W over the 24-year period from 1981 to 2004 (grey circles). Also shown are the locations of the Middle Atlantic Bight (MAB), Georges Bank (GB), Scotian Shelf (SS), Laurentian Channel (LC), and Tail of the Grand Banks (TGB). NOAA/NMFS NEFSC D. pealeii autumn trawl survey data Since 1967, D. pealeii annual relative abundance has been estimated from the NOAA/NMFS Northeast Fisheries Science Centre’s (NEFSC) annual autumn bottom-trawl survey within the NESLME located between Cape Hatteras and the Gulf of Maine. The shelf-wide autumn bottom-trawl surveys provide the greatest overlap with D. pealeii, as the population is located inshore from spring until late autumn when it migrates offshore to overwinter in warmer waters along the edge of the continental shelf (Brodziak and Hendrickson, 1999; Jacobson, 2005). The rapid growth and short lifetime of D. pealeii (∼1 year) suggest that environmental conditions, including depth and temperature, are important factors for its distribution and abundance on the continental shelf (Brodziak and Hendrickson, 1999). IAV of surface and bottom temperature may provide an important source of changes within D. pealeii habitat causing the strong IAV of relative abundance noted for this species (Dawe et al., 2007). D. pealeii relative abundance as calculated from autumn NEFSC bottom-trawl surveys is affected by water temperature, individual trawl time-of-day, and depth (Serchuk and Rathjen, 1974), with varying effect based on longfin squid body size (Brodziak and Hendrickson, 1999; Cadrin and Hatfield, 1999; Jacobson et al., 2015). Therefore, NEFSC autumn D. pealeii bottom trawl relative abundance is usually adjusted for diel differences in catchability related to body size (Brodziak and Hendrickson, 1999). Specifically, annual values of two longfin squid autumn survey indices used by the NEFSC to assess IAV of D. pealeii relative abundance from 1975 to 2016: stratified mean number per tow and stratified mean kg per tow (Figure 6), were obtained and are the primary fisheries datasets to be analysed for this work. The two indices represent mean daytime-only trawl values at stations that could be sampled from the NOAA ship Henry B. Bigelow (FSV 225) computed over the entire autumn shelf survey for each year (L. Hendrickson, pers. comm., NOAA/NMFS, Woods Hole, MA), with surveys usually occurring from mid-September through mid-October of each year (Brodziak and Hendrickson, 1999). Figure 6. Open in new tabDownload slide Annual values of two autumn survey indices used by NOAA/NMFS to assess IAV of Doryteuthis pealeii relative abundance over the NESLME from 1975 to 2016. Shading indicates the 24-year (1981–2004) subset of the data used for this study. Figure 6. Open in new tabDownload slide Annual values of two autumn survey indices used by NOAA/NMFS to assess IAV of Doryteuthis pealeii relative abundance over the NESLME from 1975 to 2016. Shading indicates the 24-year (1981–2004) subset of the data used for this study. Results WCR/shelf encounters: annual mean binned data One degree-binned (longitude) annual mean WCR/shelf encounters were compiled and show a clear reduction of encounters east of 65°W longitude over the 24-year (1981–2004) study period, although both the annual number (number per km) and annual mean areas (km2 per km) show a limited number of higher values between 65° and 60°W longitude from 1984 to 2003 (Figure 7, top and middle panels). IAV of both the annual number of WCR/shelf encounters and annual mean WCR/shelf encounter area also appears to be high, with lower values for the early-1980s, late-1990s, and early-2000s. Overall, the temporal mean encounter areas show two maxima that are significant: located between 72° and 71°W (south of eastern Long Island to Rhode Island), and between 69° and 68°W (south of Nantucket Shoals to Georges Bank), along with a weak and insignificant maximum located between 62° and 61°W (south of Nova Scotia; Figure 7, bottom panel). WCR/shelf encounter time-series at each of the two western-most maxima were examined for autocorrelation using computed autocorrelation functions. No significant autocorrelation was detected at all non-zero lags based upon a 95% significance test (Chatfield, 1984) for both time-series. The maximum located between 56° and 57°W resulted from a single WCR that appeared to enter Laurentian Channel during 1983 (Figure 5). Partitioning of annual mean WCR/shelf encounter areas into two 6-month periods, i.e. “cold season” (January–June) and “warm season” (July–December), show strong seasonal differences with “cold” season annual means greatly reduced relative to those from the “warm” season (Figure 8), but with maxima located within the same longitudinal bands for each season as those for the full dataset (Figure 7, bottom panel). These maxima show significant differences between seasons and are again located from 72° to 71°W (south of eastern Long Island to Rhode Island), and from 69° to 68°W (south of Nantucket Shoals to Georges Bank) along with much-reduced but insignificant maxima from 62° to 61°W (south of Nova Scotia). Overall, the 24-year (1981–2004) temporal mean area for each longitudinal bin shows an ∼30–50% reduction in mean encounter area between 75° and 66°W during the cold season relative to the warm season (Figure 8). Figure 7. Open in new tabDownload slide Binned valid WCR/shelf encounters from 75° to 50°W for 1981–2004: Top panel: number of encounters (per km) for each year; Middle panel: annual mean encounter areas (km2 per km) for each year; Bottom panel: temporal mean encounter areas (km2 per km) for 1981–2004 from 75° to 50°W with standard errors. Figure 7. Open in new tabDownload slide Binned valid WCR/shelf encounters from 75° to 50°W for 1981–2004: Top panel: number of encounters (per km) for each year; Middle panel: annual mean encounter areas (km2 per km) for each year; Bottom panel: temporal mean encounter areas (km2 per km) for 1981–2004 from 75° to 50°W with standard errors. Figure 8. Open in new tabDownload slide Temporal mean encounter areas for 1981–2004 (km2 per km) from 75° to 50°W with standard errors: thin line is “cold” season (January–June), heavy line is “warm” season (July–December). Figure 8. Open in new tabDownload slide Temporal mean encounter areas for 1981–2004 (km2 per km) from 75° to 50°W with standard errors: thin line is “cold” season (January–June), heavy line is “warm” season (July–December). A spatially averaged, annual mean WCR/shelf encounter area time-series was computed for the continental shelf located between Cape Hatteras and the southwestern Scotian shelf by averaging the values from the ten spatial bins located between 75° and 65°W for each year over the 24-year (1981–2004) period (Figure 9). This region includes the outer shelf along the MAB, Nantucket Shoals, and Georges Bank, thus including the two regions strongly impacted by WCR/shelf encounters as shown above (Figures 7, bottom panel, and 8) and corresponds to the NOAA/NMFS autumn trawl survey region for D. pealeii. This region was also chosen because its length scale approximates the translation distance for a typical WCR (prior to dissipation) along the outer shelf, computed by multiplying the characteristic lifetime of a WCR within the SLW by its mean translational speed: given typical WCR values of a 90-day lifetime and 0.10 m s−1 translational speed to the west-southwest, the result is a WCR translation length scale of ∼800 km along the outer shelf, thus traversing nearly the entire SLW region for most WCRs that form near 65°W. Thus, the resulting WCR translation scale of nearly 800 km justifies the alongshelf “broad-scale” used for spatial averaging and production of the annual mean WCR/shelf encounter time-series presented for 1981–2004 (Figure 9). The WCR count noted for our study between 75° and 65°W shows IAV in the number of WCRs present each year, with a minimum (maximum) number of 1 (15) different WCRs occurring during 1992 (2003; Figure 9) and agrees with the timing of the 2003 maximum number of WCRs noted by Monim (2017) over the larger region located between 75° and 55°W. Figure 9. Open in new tabDownload slide WCR/shelf encounter area time-series and number of WCRs for the 24-year (1981–2004) period occurring between 75° and 65°W, i.e. from Cape Hatteras to the western Scotian Shelf. Figure 9. Open in new tabDownload slide WCR/shelf encounter area time-series and number of WCRs for the 24-year (1981–2004) period occurring between 75° and 65°W, i.e. from Cape Hatteras to the western Scotian Shelf. The overall mean (standard deviation) of the spatially averaged annual mean WCR/shelf encounter area time-series values for the 24-year period shown in Figure 9 is 7.43 (3.74) km2 per km, resulting in a large coefficient of variation of ∼50%. Thus, this time-series clearly displays large IAV of interactions between WCRs and the outer shelf as estimated by our methodology for our 24-year study period from 75° to 65°W. More specifically, IAV of the shelf-wide mean area values varies by over one order of magnitude, with a maximum of 17.83 km2 per km for 1988 to a minimum of 0.45 km2 for 1996 (Figure 9). Furthermore, these maximum and minimum encounter years do not appear to correspond to years with maximum and minimum numbers of observed WCRs, i.e. 2003 and 1992, respectively, as described above. However, no data were available between 1 January and 21 April 1996, possibly resulting in the minimum value for 1996. Nevertheless, the Pearson correlation coefficient computed between the spatially averaged annual mean WCR/shelf encounter area and number of observed WCRs time-series (Figure 9) is small and insignificant (r = −0.015, p = 0.945), thus indicating no relationship. Relationship between WCR/shelf encounter area and D. pealeii relative abundance In order to examine the broad-scale effects of WCRs on D. pealeii relative abundance, time-series of the two NEFSC autumn survey indices (Figure 6) were used to examine IAV of D. pealeii abundance from Cape Hatteras to the Gulf of Maine, and therefore correspond to our WCR/shelf encounter study region between 75° and 65°W. Each D. pealeii relative abundance index exhibits strong IAV, with year-to-year changes of 2–3 times or more, thus displaying extreme changes between consecutive years, similar to the year-to-year changes in magnitude of WCR/shelf encounter area described above over the same spatial domain. As a test of the overall broad-scale effect of WCRs on the commercially valuable D. pealeii, we computed correlations between 24-year subsets of autumn D. pealeii relative abundance index time-series (Figure 6) and our spatially averaged WCR/shelf encounter time-series for the same 24-year (1981–2004) period. Co-plots of D. pealeii relative abundance indices and WCR/shelf encounter time-series are shown (Figure 10), with results of our regression analyses provided in Tables 1 and 2. Correlations at a lag of 0 years were computed to examine the relationship between time-series using the versatile Matlab® built-in regression function (corr) that utilized both a parametric (Pearson) and non-parametric (Kendall) statistic between variables similar to earlier work by Myers and Drinkwater (1989). In the event that the assumptions for ordinary least squares regression are not met by our datasets, the non-parametric regression (Kendall rank correlation coefficient) was used to test whether the dependent variable, D. pealeii relative abundance is a monotonically decreasing function of the independent variable, WCR/shelf encounter area. Results show weak and insignificant correlation values for both the Pearson and Kendall statistics between year-to-year changes in D. pealeii relative abundance (number per tow) and our WCR/shelf encounter area over the 24-year period (Table 1). Similarly, results show weak and insignificant correlation values for both the Pearson and Kendall statistics between year-to-year changes in D. pealeii relative abundance (kg per tow) and our WCR/shelf encounter area over the same 24-year core period (Table 1). Eliminating two years (1995 and 1996) with significant satellite chart data outage periods did not result in improved correlations for both the number per tow and kg per tow relative abundance indices (Table 2). Overall, the weak and insignificant correlation values shown for both the parametric Pearson and non-parametric Kendall statistics show no relationship exists between both the number per tow and kg per tow D. pealeii relative abundance indices and our WCR/shelf encounter area time-series. We caution that the statistical significance for the present correlations shown in Tables 1 and 2 are calculated without accounting for autocorrelation in the time-series, which would only serve to likely lower the statistical significance even further. In summary, based upon our analysis, there is no relationship between our broad-scale WCR/shelf interaction metric and the spatially averaged NOAA/NMFS NEFSC D. pealeii autumn relative abundance during the 1981–2004-study period. Figure 10. Open in new tabDownload slide WCR/shelf encounter area time-series for the 24-year (1981–2004) period between 75° and 65°W plotted along with Doryteuthis pealeii stratified mean relative abundance. Top panel: number per tow. Bottom panel: kg per tow. Figure 10. Open in new tabDownload slide WCR/shelf encounter area time-series for the 24-year (1981–2004) period between 75° and 65°W plotted along with Doryteuthis pealeii stratified mean relative abundance. Top panel: number per tow. Bottom panel: kg per tow. Table 1. Cross-correlations (r) between WCR/shelf encounter areas and longfin squid (Doryteuthis pealeii) stratified mean relative abundance indices over the 24-year (1981–2004) period between 75° and 65°W. Stratified mean index N Pearson r (p-value) Kendall r (p-value) Abundance index (# per tow) 24 0.029 (0.893) 0.138 (0.363) Abundance index (kg tow−1) 24 0.290 (0.169) 0.280 (0.059) Stratified mean index N Pearson r (p-value) Kendall r (p-value) Abundance index (# per tow) 24 0.029 (0.893) 0.138 (0.363) Abundance index (kg tow−1) 24 0.290 (0.169) 0.280 (0.059) Open in new tab Table 1. Cross-correlations (r) between WCR/shelf encounter areas and longfin squid (Doryteuthis pealeii) stratified mean relative abundance indices over the 24-year (1981–2004) period between 75° and 65°W. Stratified mean index N Pearson r (p-value) Kendall r (p-value) Abundance index (# per tow) 24 0.029 (0.893) 0.138 (0.363) Abundance index (kg tow−1) 24 0.290 (0.169) 0.280 (0.059) Stratified mean index N Pearson r (p-value) Kendall r (p-value) Abundance index (# per tow) 24 0.029 (0.893) 0.138 (0.363) Abundance index (kg tow−1) 24 0.290 (0.169) 0.280 (0.059) Open in new tab Table 2. Cross-correlations (r) between WCR/shelf encounter areas and longfin squid (Doryteuthis pealeii) stratified mean relative abundance indices for the 24-year (1981–2004) period between 75° and 65°W excluding data from 1995 to 1996 owing to partial chart data outages. Stratified mean index N Pearson r (p-value) Kendall r (p-value) Abundance index (# per tow) 22 0.052 (0.817) 0.065 (0.696) Abundance index (kg tow−1) 22 0.207 (0.356) 0.234 (0.135) Stratified mean index N Pearson r (p-value) Kendall r (p-value) Abundance index (# per tow) 22 0.052 (0.817) 0.065 (0.696) Abundance index (kg tow−1) 22 0.207 (0.356) 0.234 (0.135) Open in new tab Table 2. Cross-correlations (r) between WCR/shelf encounter areas and longfin squid (Doryteuthis pealeii) stratified mean relative abundance indices for the 24-year (1981–2004) period between 75° and 65°W excluding data from 1995 to 1996 owing to partial chart data outages. Stratified mean index N Pearson r (p-value) Kendall r (p-value) Abundance index (# per tow) 22 0.052 (0.817) 0.065 (0.696) Abundance index (kg tow−1) 22 0.207 (0.356) 0.234 (0.135) Stratified mean index N Pearson r (p-value) Kendall r (p-value) Abundance index (# per tow) 22 0.052 (0.817) 0.065 (0.696) Abundance index (kg tow−1) 22 0.207 (0.356) 0.234 (0.135) Open in new tab Discussion WCR/shelf encounters: annual mean binned data Previous studies have not directly addressed WCR/shelf encounters along the entire outer continental shelf region located off the northeastern United States and Canada between Cape Hatteras and the Tail of the Grand Banks using observational data at 1-degree longitude resolution. A few observational studies over limited time periods have shown how WCRs may be locally important in limited regions because of their seaward SHW entrainment, i.e. Georges Bank (Garfield and Evans, 1987) and the Scotian shelf (Smith, 1978) and have provided useful statistics. In contrast, this study provides the first comprehensive analysis of a multi-decadal (24-year) observational dataset describing the interactions of WCRs with the entire outer continental shelf located between 75° and 50°W longitude. Furthermore, our input data are congruent with WCR data used by Chaudhuri et al. (2009) for their modelling analysis of WCR SHW entrainment “streamers” and thus will be compared with their spatial and temporal results. We have also carefully examined our data sources to exclude years with questionable chart inputs, resulting in the 24-year (1981–2004) study period that utilized a consistent type of satellite-derived ocean frontal analysis chart used to determine our digitized WCR locations and radii. Additional spatial averaging of our estimated annual mean WCR/shelf encounter area values from 75° to 65°W over the 24-year core period has allowed comparison with the noted strong IAV of autumn relative abundance of D. pealeii over the corresponding shelf region. The effect of WCRs on cephalopod abundances have generally not been examined, although the effect of WCRs on 16 species of groundfish stocks have shown a negative relationship between recruitment and a WCR entrainment index over a similar spatial domain (Myers and Drinkwater, 1989). Myers and Drinkwater (1989) used a shorter time-series (1973–1986) of satellite SST charts to determine a shelf water entrainment index based upon the distance of each WCR from the outer shelf, with maximum effect if the WCR was located at or inshore of the 200-m isobath. Our results show WCR/shelf encounters are clearly concentrated within the region west of 65°W, although not exclusively so, whether considering the raw valid WCR/shelf encounters (Figure 5) or annual mean encounter areas at each longitude (Figure 7). This result is in agreement with the WCR entrainment index results of Myers and Drinkwater (1989) that showed a strong decrease between their Eastern Georges Bank (Region 4) and Browns Bank (Region 5) sub-regions. It is particularly striking that temporal means of encounter areas show robust “hot spot” maxima near 72° to 71°W (south of eastern Long Island to Rhode Island), and from 69° to 68°W (south of Nantucket Shoals to Georges Bank) along with the much-reduced maximum from 62° to 61°W (south of Nova Scotia; Figure 7, bottom panel). Furthermore, the difference between the cold season temporal mean encounter area with that from the warm season (Figure 7) is large, with the later warm season displaying higher values by a factor of ∼2 from 75° to 65°W longitude. Monim (2017) has shown a clear seasonal signal with more frequent WCR formation during the months of May, June, and July that may result in greater numbers of WCRs available for interaction with the shelf, resulting in our higher warm season values. Spatially, the three maxima appear to correspond to changes in orientation of the generalized outer shelf bathymetry (200-m isobath) within the three longitudinal bands, with the largest maximum located farthest west where the outer shelf bathymetry changes from E–W to NE–SW near 72°W (Figures 5 and 7 bottom panel). The WCR/shelf encounter local maximum present in our data near Georges Bank is in agreement with the location of the maximum in SHW entrainment features noted by Chaudhuri et al. (2009) for their five-degree longitude “GB” (Georges Bank) region and the corresponding maximum in annual mean SHW entrainment transport of 7543 km3 year−1. Model values of SHW entrainment decrease eastward (westward) within their adjacent five-degree longitude “SS” Scotian shelf (“MAB”) regions, although our maximum corresponds to the region between 72° and 71°W within the MAB. Both quantitative analyses are also in agreement with the satellite-derived WCR analysis of Garfield and Evans (1987) showing that 86% of observable days corresponded to periods when SHW entrainment streamers were recorded south of Georges Bank from 1979 to 1985 for 49 WCRs, indicating that southern Georges Bank has long been recognized as a “hot spot” for WCR activity. Despite our spatial averaging of the binned annual mean WCR/shelf encounter areas located between 75° and 65°W (MAB through Georges Bank) over the 24-year study period, the large IAV of encounter area (by at least a factor of 2) is both surprising and an important result (Figure 9). Furthermore, given the large IAV of WCR/shelf encounter area shown, the lack of a relationship between the annual number of discrete WCRs and the annual mean encounter area values (r = −0.015, p = 0.945) is also somewhat surprising, but supports the lack of agreement between corresponding years of encounter area magnitude and WCR number extrema discussed earlier. We would have expected, a priori, that maximum and minimum WCR/shelf encounter years, should correspond to years with maximum and minimum numbers of WCRs, respectively, but clearly this is not the case. These results show that the magnitudes of interactions between WCRs and the outer shelf between 75° and 65°W do not depend upon the number of WCRs observed within this limited SLW region. An examination of all raw WCR realizations for 1981 and 1988 show clear differences in the magnitude of realizations that resulted in valid WCR/shelf encounters (Figure 11). The number, N, of discrete WCRs that were tabulated within the SLW region located between 75° and 65°W for each year is also shown (Figure 11) and are nearly identical. However, fewer WCR/shelf encounter “overlaps” are present for 1981 and produced a lowered spatially averaged area of only 2.05 km2 per km resulting from N = 2, discrete WCRs traversing the region (Figure 9). In contrast, more WCR/shelf encounter “overlap” regions are readily apparent during 1988, especially on southern Georges Bank, and produced the maximum spatially averaged area of 17.83 km2 per km resulting from N = 3 discrete WCRs. WCR radii also appear to be smaller for 1981 than the clearly larger WCR radii for 1988, especially for those WCRs located west of 70°W (Figure 11). Figure 11. Open in new tabDownload slide All raw WCR realizations for 1981 and 1988 used to determine WCR/shelf encounter areas for each of the 2 years from 75° to 50°W shown in Figure 7. Also shown is the Middle Atlantic Bight through Georges Bank region (black boxes) used for spatial averaging of the binned WCR/shelf encounter values from 75° to 65°W. A mean WCR/shelf encounter value of 2.05 km2 per km (17.83 km2 per km) was computed for 1981 (1988) within the boxed region. Also shown is the number N of discrete WCRs traversing the boxed region during each year. Figure 11. Open in new tabDownload slide All raw WCR realizations for 1981 and 1988 used to determine WCR/shelf encounter areas for each of the 2 years from 75° to 50°W shown in Figure 7. Also shown is the Middle Atlantic Bight through Georges Bank region (black boxes) used for spatial averaging of the binned WCR/shelf encounter values from 75° to 65°W. A mean WCR/shelf encounter value of 2.05 km2 per km (17.83 km2 per km) was computed for 1981 (1988) within the boxed region. Also shown is the number N of discrete WCRs traversing the boxed region during each year. An inter-year comparison of the specific differences between WCR/shelf encounters for 1981 and 1988 shows that mean latitudes of WCR centroids were nearly identical within the MAB to Georges Bank (75°W–65°W) longitudinal band (Figure 12 top panel). This longitudinal band contains the two maxima of the temporal mean encounter areas computed for the 24-year period (Figure 7, bottom panel). In contrast, annual mean WCR radii computed for the same longitudinal band for 1981 and 1988 show a clear difference in WCR size, with 1988 showing larger radii, with inter-year differences increasing westward to about a factor of two at 72.5°W (Figure 12 bottom panel). Overall, the extreme IAV of WCR/shelf encounters noted between 1981 and 1988 (Figure 9) appear controlled by WCR radius rather than WCR location. Figure 12. Open in new tabDownload slide Annual mean centroid latitude (top panel) and radius (bottom panel) for all WCRs from 1981 to 1988 for the ten longitude bands located between 75° and 65°W (corresponding to all WCRs within the black boxes from Figure 11). Figure 12. Open in new tabDownload slide Annual mean centroid latitude (top panel) and radius (bottom panel) for all WCRs from 1981 to 1988 for the ten longitude bands located between 75° and 65°W (corresponding to all WCRs within the black boxes from Figure 11). Although IAV in the number of WCRs formed each year and their seaward entrainment of SHW for 1978–1999 was positively correlated with the state of the NAO at 0-year lag (Chaudhuri et al., 2009), the physical mechanisms controlling WCR latitudes and sizes need additional studies. Monim (2017) showed that the annual number of WCRs formed from 2000 to 2016 was negatively correlated with the annual GS North Wall index of Taylor and Stephens (1998), i.e. a more southerly GS corresponded to larger numbers of WCRs. Aside from these two statistical relationships and the kinematical relationships between WCRs and the shelf along with its overall size after formation are still largely unknown at this time and beyond the scope of this work. WCR/shelf encounters area and D. pealeii abundance The extreme spatial and temporal variation in WCR/shelf encounters described above likely results in a very large fraction of the day-to-day physical variability (temperature, salinity, and current velocities) experienced along the outer continental shelf, from Cape Hatteras to Georges Bank. Moreover, we assumed a priori that such variation in outer continental shelf physical conditions may somehow affect D. pealeii abundance through changes in habitat temperature and/or advection of squid either onshore or offshore. However, although temporal changes in both the broad-scale WCR/shelf encounter values and D. pealeii relative abundance indices appear large, standard regression models indicate no relationship between these two strongly varying signals for our 24-year study period from 75° to 65°W. Thus, this lack of covariance does not agree with the results of Myers and Drinkwater (1989) who found negative effects of WCR-induced seaward entrainment of SHW on recruitment of 16 of 17 groundfish stocks (with the exception of Georges Bank cod) from the MAB to the Grand Banks. Interestingly, they found that seven pelagic stocks and one shellfish stock (sea scallops) did not show a consistent relationship, with only one pelagic species, argentine, that lives on the outer shelf showing decreased recruitment with increased WCR activity in agreement with the response of the groundfish. Given the pelagic nature of D. pealeii, especially during winter when the species is abundant along the outer shelf, the lack of correlation between the two relative abundance indices and WCR/shelf encounters is not surprising. Thus, we can speculate that two potential mechanisms affecting D. pealeii abundance along the outer shelf related to the presence of WCRs, i.e. changes in thermal habitat and direct offshore entrainment, may not affect the broad-scale, shelf-wide abundance of D. pealeii and its noted extreme year-to-year variability. This is somewhat surprising in that previous work has indicated that autumn survey catches of D. pealeii were correlated with both warmer surface and bottom temperatures (Brodziak and Hendrickson, 1999) whereas spatially-explicit habitat modelling efforts indicated that longfin squid relative abundance in the MAB was seasonally associated with areas of strong surface current divergence and frontal activity (Manderson et al., 2011). Future investigations of WCR effects on D. pealeii abundance should be conducted at finer spatio-temporal scales than that of this study. Incorporating squid size structure into future analyses of D. pealeii survey data may provide some insight as to potential mechanisms of WCR effects on D. pealeii relative abundance and the temporal scales at which they occur. Furthermore, the results of our broad-scale, shelf-wide analysis of the effects of WCRs on D. pealeii relative abundance indicate that more regionally-focused studies should be completed centring on the Georges Bank and MAB region maxima where WCR/shelf encounters are most likely to overlap with longfin squid distribution, particularly in winter. Fall survey-derived squid abundance data could be compared with the previous year’s summer WCR data to test for potential WCR effects on recruitment. Additional progress in this area will require more spatially and temporally explicit life-history stage data and modelling studies. Our WCR/shelf encounter metric may be applied using new data, and does not depend on any ring entrainment model, e.g. the index used by Myers and Drinkwater (1989) or the ring entrainment model of Chaudhuri et al. (2009), and is therefore a straightforward and easily measureable quantity given the raw WCR centres and areas that are available from future satellite-derived frontal charts. In addition, finer-scale spatio-temporal analyses of WCR-driven water mass dynamics (e.g. Gawarkiewicz et al., 2018; Zhang and Partida, 2018) may provide additional insight as to the mechanisms by which thermal habitats and distribution are affected. Outside of the WCR/shelf encounters described herein, a direct and strong interaction of the GS itself with the outer continental shelf off southern New England occurred during late 2011 when a very large shoreward meander of the GS produced high bottom temperatures on the shelf (Gawarkiewicz et al., 2012). Thus, although direct GS interactions with the outer shelf are not examined as part of this work, the 1500 km westward translation of the destabilization point for initiation of GS meandering that occurred between 1993 and 2014 using mapped satellite altimetry (Andres, 2016), may have increased the direct GS interactions with the outer shelf. Furthermore, we speculate that any continued future westward translation of the initiation of GS meandering may also increase both the WCR/shelf encounters reported here and an increase of direct GS impact on the outer shelf as well. Summary and conclusions Analysis of an extensive, satellite chart-derived, GS WCR dataset provided by BIO and covering the SLW region between 75° and 50°W shows strong spatial and IAV of WCR/shelf encounters inshore of the 200-m isobath. Careful consideration of the satellite-derived frontal charts digitized at BIO to produce the primary WCR dataset resulted in selection of the 24-year (1981–2004) study period when a consistent type of satellite chart was used. The large number of satellite chart WCR realizations contained in the BIO WCR dataset (N = 16 108) was analysed using an automated procedure to compute valid “overlap” areas between each WCR realization and a smoothed 200-m isobath that defined the seaward edge of the shelf. Results show that 14.8% of the WCR realizations (N = 2377) were valid WCR/shelf encounters, with the longitude of the midpoint of each valid encounter used to assign the computed area of each WCR/shelf encounter to one of 25 longitudinal bins between 75° and 50°W. This method allowed the number of WCR/shelf encounters and their area magnitudes to be compiled at 1-degree resolution from 1981 through 2004 between Cape Hatteras and the Tail of the Grand Banks. Annual mean WCR/shelf encounter areas for each longitudinal bin were computed by dividing the total annual area for each bin by the annual number of valid encounters for each corresponding bin. The 1981–2004 study period was also selected to be congruent in time with two relative abundance time-series indices (number per tow and kg per tow) of the commercially valuable longfin squid (D. pealeii) computed using NOAA/NMFS autumn surveys over the broad-scale region on the continental shelf from Cape Hatteras to Georges Bank. The short-lived (∼1-year) D. pealeii is an important species to evaluate possible broad-scale effects of WCR interactions, e.g. through direct WCR SHW entrainments and/or WCR-related temperature changes that may account for the extreme IAV of D. pealeii relative abundance on the continental shelf from Cape Hatteras to Georges Bank during autumn. Results show a large reduction of WCR/shelf encounters east of 65°W for the 24-year study period, although both the annual number and annual mean areas show a limited number of higher WCR/shelf encounter values between 65° and 60°W. Temporal mean WCR/shelf encounter areas show strong maxima: from 72° to 71°W (south of eastern Long Island to Rhode Island), and from 69° to 68°W (south of Nantucket Shoals to Georges Bank) along with a weak maximum from 62° to 61°W (south of Nova Scotia) for the 24-year study period. These maxima are indicative of “hot-spots” for WCR/shelf encounters in agreement with earlier studies and are located where changes in the orientation of outer shelf bathymetry are maximal. Higher temporal mean WCR/shelf encounter areas were observed for the July–December (warm) months of up to twice those for the January–June (cold) months for the 24-year study period west of 65°W in agreement with earlier work showing a strong seasonal cycle in WCR formation that is maximal from late-spring to early-summer. Spatial averaging of the ten longitudinal bins located between 75° and 65°W produced a time-series over the 24-year study period of annual mean WCR/shelf encounter area over a spatially broad-scale that is typical of a WCR translating through the SLW region. Limited data suggest that strong IAV of encounter area (year-to-year changes varying by a factor of at least two) is primarily because of IAV of WCR size (radius). The maximum (minimum) number of discrete WCRs present between 75° and 65°W over the 24-year study period is 15 (1), corresponding to 2003 (1992). The correlation between time-series of spatially averaged WCR/shelf encounter area and annual number of discrete WCRs is small and insignificant, indicating no relationship. Correlations of the spatially averaged annual mean WCR/shelf encounter area time-series with two D. pealeii autumn survey relative abundance indices from 75° to 65°W between 1981 and 2004 were weak and insignificant (p > 0.05) using both parametric and non-parametric methods. The weak and insignificant correlation values show no relationship between both the number per tow and kg per tow D. pealeii relative abundance indices and our WCR/shelf encounter metric over the broad-scale shelf region between Cape Hatteras and Georges Bank. Although a determination of specific mechanism(s) possibly connecting WCR/shelf encounter area with D. pealeii abundance is beyond the scope of this initial study, the lack of a relationship over the broad-scale shelf region used in this study suggests that finer scale regional studies may be warranted. Future work will examine the effects of WCR/shelf encounter “hot-spots” on D. pealeii populations similar to earlier work that examined the effect of direct seaward entrainment of SHW and loss of both eggs and larvae by WCRs using stock specific indices for recruitment of 16 groundfish stocks. Given that our WCR/shelf encounters, by definition, require a WCR to be located partly inshore of the 200-m isobath, suggest that changes in D. pealeii thermal habitat may also be occurring. Although annual censuses of GS WCR populations within the SLW are a useful measure of WCR activity within the SLW, estimates of WCR/shelf encounters and their impact on the outer shelf and its resident ecosystem are best made using an automated image analysis technique. Our automated technique can be easily applied to future WCR data when they become available. Acknowledgements The authors would like to thank Lisa Hendrickson, NOAA/NMFS Northeast Fisheries Science Centre, Woods Hole, MA, for providing the annual autumn D. pealeii bottom-trawl survey data and assisting with its use for this study. The authors would also like to especially thank Jenifer Clark, Dunkirk, MD, for providing the Gulf Stream satellite charts used in this study, and three reviewers for their efforts that provided constructive and worthwhile comments and suggestions that greatly improved the manuscript. References Andres M. 2016 . On the recent destabilization of the Gulf Stream path downstream of Cape Hatteras . Geophysical Research Letters , 43 : 9836. Google Scholar Crossref Search ADS WorldCat Auer S. J. 1987 . Five‐year climatological survey of the Gulf Stream system and its associated rings . Journal of Geophysical Research: Oceans , 92 : 11709 – 11726 . Google Scholar Crossref Search ADS WorldCat Bisagni J. J. 1976 . Passage of anticyclonic Gulf Stream eddies through deepwater dumpsite 106 during 1974 and 1975. NOAA Dumpsite Evaluation Report, 76-1. US Department of Commerce Pub. 39 pp. Bisagni J. J. 1983 . Lagrangian current measurements within the eastern margin of warm-core Gulf Stream ring . Journal of Physical Oceanography , 13 : 709 – 715 . Google Scholar Crossref Search ADS WorldCat Bisagni J. J. , Kim H. S. , Chaudhuri A. 2009 . Inter-annual variability of the shelf slope front position between 75° and 50°W . Journal of Marine Systems , 78 : 337 – 350 . Google Scholar Crossref Search ADS WorldCat Black G. A. P. , Rowell T. W. , Dawe E. G. 1987 . Atlas of the biology and distribution of the squids Illex illecebrosus and Loligo pealei in the northwest Atlantic . Canadian Special Publication of Fisheries and Aquatic Sciences , 100 : 62 . WorldCat Brodziak J. , Hendrickson L. 1999 . An analysis of environmental effects on survey catches of squids Loligo pealeii and Illex illecebrosus in the northwest Atlantic . Fishery Bulletin , 97 : 9 – 24 . WorldCat Brown O. B. , Cornillon P. C. , Emmerson S. R. , Carle H. M. 1986 . Gulf Stream warm rings: a statistical study of their behavior . Deep Sea Research Part A, Oceanographic Research Papers , 33 : 1459 – 1473 . Google Scholar Crossref Search ADS WorldCat Cadrin S. X. , Hatfield E. M. C. 1999 . Stock assessment of longfin inshore squid, Loligo pealeii. Northeast Fisheries Science Center Reference Document, 99-12. 107 pp. Chatfield C. 1984 . The Analysis of Time Series, an Introduction . University of Michigan Press , Ann Arbor, Michigan, USA . 286 pp. Google Preview WorldCat COPAC Chaudhuri A. H. , Bisagni J. J. , Gangopadhyay A. 2009 . Shelf water entrainment by Gulf Stream warm-core rings between 75°W and 50°W during 1978–1999 . Continental Shelf Research , 29 : 393 – 406 . Google Scholar Crossref Search ADS WorldCat Cherian D. A. , Brink K. H. 2016 . Offshore transport of shelf water by deep-ocean eddies . Journal of Physical Oceanography , 46 : 3599 – 3621 . Google Scholar Crossref Search ADS WorldCat Cherian D. A. , Brink K. H. 2018 . Shelf flows forced by deep-ocean anticyclonic eddies at the shelf break . Journal of Physical Oceanography , 48 : 1117 – 1138 . Google Scholar Crossref Search ADS WorldCat Cowen R. K. , Hare J. A. , Fahay M. P. 1993 . Beyond hydrography: can physical processes explain larval fish assemblages within the Middle Atlantic Bight? Bulletin of Marine Science , 53 : 567 – 587 . WorldCat Dawe E. G. , Colbourne E. B. , Drinkwater K. F. 2000 . Environmental effects on recruitment of short-finned squid (Illex illecebrosus) . ICES Journal of Marine Science , 57 : 1002 – 1013 . Google Scholar Crossref Search ADS WorldCat Dawe E. G. , Hendrickson L. C. , Colbourne E. B. , Drinkwater K. F. , Showell M. A. 2007 . Ocean climate effects on the relative abundance of short-finned (Illex illecebrosus) and long-finned (Loligo pealeii) squid in the northwest Atlantic Ocean . Fisheries Oceanography , 16 : 303 – 316 . Google Scholar Crossref Search ADS WorldCat Evans R. H. , Baker K. S. , Brown O. B. , Smith R. C. 1985 . Chronology of warm-core ring 82B . Journal of Geophysical Research , 90 : 8803 – 8811 . Google Scholar Crossref Search ADS WorldCat Fitzgerald J. L. , Chamberlin J. L. 1983 . Anti-cyclonic warm-core Gulf Stream eddies off the northeastern United States in 1980 . Annales Biologique , 37 : 41 – 47 . WorldCat Flierl G. R. , Wroblewski J. S. 1985 . The possible influence of warm core Gulf Stream rings upon shelf water larval fish distribution . Fisheries Bulletin , 83 : 313 – 330 . WorldCat Gangopadhyay A. , Robinson A. R. , Arango H. G. 1997 . Circulation and dynamics of the western North Atlantic, Part I: multiscale feature models . Journal of Atmosphere and Ocean Technology , 14 : 1314 – 1332 . Google Scholar Crossref Search ADS WorldCat Garfield N. , Evans D. L. 1987 . Shelf water entrainment by Gulf Stream warm-core rings . Journal of Geophysical Research , 95 : 13003 – 13012 . Google Scholar Crossref Search ADS WorldCat Gawarkiewicz G. , Todd R. E. , Plueddemann A. J. , Andres M. , Manning J. P. 2012 . Direct interaction between the Gulf Stream and the shelfbreak south of New England . Scientific Reports , 2 : 553. Google Scholar Crossref Search ADS PubMed WorldCat Gawarkiewicz G. , Todd R. E. , Zhang W. , Partida J. , Gangopadhyay A. , Monim M-U-H. , Fratantoni P. , et al. . 2018 . The changing nature of shelfbreak exchange revealed by the OOI pioneer array . Oceanography , 31 : 60 – 70 . Google Scholar Crossref Search ADS WorldCat Glenn S. M. , Forristall G. Z. , Cornillion P. , Milkowski G. 1990 . Observations of Gulf Stream ring 83-E and their interpretation using feature models . Journal of Geophysical Research , 92 : 13043 – 13064 . Google Scholar Crossref Search ADS WorldCat Halliwell G. R. , Mooers C. N. K. 1979 . The space-time structure and variability of the shelf water—slope water and Gulf Stream surface temperature fronts and associated warm-core rings . Journal of Geophysical Research , 84 : 7707 – 7725 . Google Scholar Crossref Search ADS WorldCat Hare J. A. , Cowen R. K. 1991 . Expatriation of Xyrichtys novacula (Pisces: Labridae) larvae: evidence of rapid cross-slope exchange . Journal of Marine Research , 49 : 801 – 823 . Google Scholar Crossref Search ADS WorldCat Hare J. A. , Churchill J. H. , Cowen R. K. , Berger T. J. , Cornillon P. C. , Dragos P. , Glenn S. M. , et al. . 2002 . Routes and rates of larval fish transport from the southeast to the northeast United States continental shelf . Limnology and Oceanography , 47 : 1774 – 1789 . Google Scholar Crossref Search ADS WorldCat Hatfield E. M. C. , Cadrin S. X. 2002 . Geographic and temporal patterns in size and maturity of the longfin inshore squid (Loligo pealeii) off the northeastern United States . Fishery Bulletin , 100 : 200 – 213 . WorldCat Hurrell J. W. 1995 . Decadal trends in the North Atlantic oscillation: regional temperatures and precipitation . Science , 269 : 676 – 679 . Jacobson L. D. 2005 . Longfin inshore squid, Loligo pealei, life history and habitat characteristics. NOAA Technical Memorandum NMFS-NE-193. 42 pp. Jacobson L. D. , Hendrickson L. C. , Tang J. 2015 . Solar zenith angles for biological research and an expected catch model for diel vertical migration patterns that affect stock size estimates for longfin inshore squid (Doryteuthis pealeii) . Canadian Journal of Fisheries and Aquatic Sciences , 72 : 1329 – 1338 . Google Scholar Crossref Search ADS WorldCat Joyce T. M. 1984 . Velocity and hydrographic structure of a Gulf Stream warm-core ring . Journal of Physical Oceanography , 14 : 936 – 947 . Google Scholar Crossref Search ADS WorldCat Lai D. Y. , Richardson P. L. 1977 . Distribution and movement of Gulf Stream rings . Journal of Physical Oceanography , 7 : 670 – 683 . Google Scholar Crossref Search ADS WorldCat Lee C. M. , Brink K. H. 2010 . Observations of storm-induced mixing and Gulf Stream Ring incursion over thesouthern flank of Georges Bank: Winter and summer 1997 . Journal of Geophysical Research , 115 : C08008 . Google Scholar Crossref Search ADS WorldCat Lentz S. J. 2008 . Observations and a model of the mean circulation over the Middle Atlantic Bight continental shelf . Journal of Physical Oceanography , 38 : 1203 – 1221 . Google Scholar Crossref Search ADS WorldCat Lentz S. J. 2010 . The mean along-isobath heat and salt balances over the Middle Atlantic Bight continental shelf . Journal of Physical Oceanography , 40 : 934 – 948 . Google Scholar Crossref Search ADS WorldCat Manderson J. , Palamara L. , Kohut J. , Oliver M. J. 2011 . Ocean observatory data are useful for regional habitat modeling of species with different vertical habitat preferences . Marine Ecology Progress Series , 438 : 1 – 17 . Google Scholar Crossref Search ADS WorldCat Monim M. 2017 . Seasonal and inter-annual variability of the Gulf Stream warm-core rings from 2000 to 2016. MS thesis, University of Massachusetts, Dartmouth. 113 pp. Morgan C. W. , Bishop J. M. 1977 . An example of Gulf Stream eddy-induced water exchange in the Mid-Atlantic Bight . Journal of Physical Oceanography , 7 : 472 – 479 . Google Scholar Crossref Search ADS WorldCat Myers R. , Drinkwater K. 1989 . The influence of Gulf Stream warm-core rings on recruitment of fish in the northwest Atlantic . Journal of Marine Research , 47 : 635 – 656 . Google Scholar Crossref Search ADS WorldCat Richardson P. L. 1983 . Gulf Stream rings. Chapter 2. In Eddies in Marine Science , pp. 19 – 45 . Ed. by Robinson A. R. Springer-Verlag , Berlin . Google Preview WorldCat COPAC Rodhouse P. G. K. , Pierce G. J. , Nichols O. C. , Sauer W. H. H. , Arkhipkin A. I. , Laptikhovsky V. V. , Lipiński M. R. , et al. . 2014 . Environmental effects on cephalopod population dynamics: implications for management of fisheries . Advances in Marine Biology , 67 : 99 – 233 . Google Scholar Crossref Search ADS PubMed WorldCat Saunders P. M. 1971 . Anticyclonic eddies formed on the shoreward meanders of the Gulf Stream . Deep-Sea Research Part , 18 : 1207 – 1219 . WorldCat Serchuk F. M. , Rathjen W. F. 1974 . Aspects of the distribution and abundance of long-finned squid, Loligo pealeii, between Cape Hatteras and Georges Bank . Marine Fisheries Review , 36 : 10 – 17 . WorldCat Smith P. C. 1978 . Low-frequency fluxes of momentum, heat, salt, and nutrients at the edge of the Scotian shelf . Journal of Geophysical Research , 83 : 4079 – 4096 . Google Scholar Crossref Search ADS WorldCat Summers W. C. 1969 . Winter population of Loligo pealei in the Mid-Atlantic Bight . Biological Bulletin , 137 : 202 – 216 . Google Scholar Crossref Search ADS WorldCat Taylor A. H. , Stephens J. A. 1998 . The North Atlantic oscillation and the latitude of the Gulf Stream . Tellus , 50A : 134 – 142 . Google Scholar Crossref Search ADS WorldCat Vovk A. N. 1978 . Peculiarities of the seasonal distribution of the North American squid, Loligo pealeii (Lesueur 1821) . Malacological Review , 11 : 130 . WorldCat Wright W. R. 1976 . The limits of shelf water south of Cape Cod, 1941–1972 . Journal of Marine Research , 34 : 1 – 14 . WorldCat Zhai X. , Greatbatch R. J. , Kohlmann J. D. 2008 . On the seasonal variability of eddy kinetic energy in the Gulf Stream region . Geophysical Research Letters , 35 : L24609 . Google Scholar Crossref Search ADS WorldCat Zhang W. G. , Partida J. 2018 . Frontal subduction of the Mid-Atlantic Bight shelf water at the offshore edge of a warm-core ring . Journal of Geophysical Research , 123 . WorldCat © International Council for the Exploration of the Sea 2019. All rights reserved. For permissions, please email: [email protected] This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
Observed and hindcasted subdecadal variability of the tropical Pacific climateMochizuki, Takashi; Watanabe, Masahiro
doi: 10.1093/icesjms/fsz026pmid: N/A
Abstract The tropical Pacific is highly responsible for controlling the tendencies of global climate. Here, we have found a subdecadal (i.e. 3-year-running means) variation that has been distinctively observed in the 2000s over the tropical Pacific. The warm water that originated from the positive ocean-heat-content (OHC) anomaly over the western Pacific in 2000 slowly extended eastward and stayed at the central equatorial Pacific between 2002 and 2005, rather than reaching the eastern edge. At the same time, an accompanying anticyclonic surface wind anomaly was observed in the off-equatorial North Pacific. This dynamical response of the upper ocean may have contributed to the subsequent warming in the western Pacific. In the preceding decades, in contrast, the observed OHC has usually represented a periodic fluctuation in the tropical Pacific, characterized by anomalous heat input/output in the meridional direction and slow eastward adjustment in the equatorial ocean thermocline. This subdecadal variation in the 2000s is quite distinct from our decadal hindcasts with initialization in Coupled Model Intercomparison Project Phase 5. The model predominantly simulates periodic-type fluctuations in any decade, and the resultant low predictability of the subdecadal variation in the 2000s can work to modulate the predictive skills at a lead time of several years. Introduction The first decade of the 21st century (i.e. the 2000s) is a challenging period for near-term (i.e. decadal) climate prediction, which constitutes a new and important aspect of Coupled Model Intercomparison Project Phase 5 (CMIP5) (Meehl et al., 2010, 2014). Decadal prediction will facilitate strategies that address socioeconomic problems arising from imminent climate change and serious impacts on marine ecosystems and fisheries. CMIP5 decadal hindcasts demonstrate that the surface air temperature (SAT) exhibits predictive capacities over decadal timescales as a global average in a statistical sense (Doblas-Reyes et al., 2011, 2013; Kim et al., 2012; Mochizuki et al., 2012) but has limited degrees of predictability for the slowdown of the global warming tendency in the 2000s (Easterling and Wehner, 2009), even when initialization is applied. Initialized decadal hindcasts validate predictive skills for specific climate modes observed mainly in the extratropics such as the Atlantic Multidecadal Oscillation (Smith et al., 2007; Keenlyside et al., 2008) and the Pacific Decadal Oscillation (Pohlmann et al., 2009; Mochizuki et al., 2010), but at this stage, decadal predictability has not been realized in the sea surface temperature (SST) changes over the tropical Pacific, which predominantly modulates the global warming tendency (Kosaka and Xie, 2013). A further understanding of physical processes, in addition to improving prediction systems (e.g. high quality climate models, initialization systems, and observations), should substantially contribute to improving our prediction skills (Eyring et al., 2015). While in early years, several studies pointed out the potential effects of external forcing on the observed slowdown of global warming tendency [such as reduction of water vapour in the stratosphere (Solomon et al., 2010) and weak solar activity (Kaufmann et al., 2011)], a number of studies has suggested the contribution of internally generated decadal variability to the climate system (Kosaka and Xie, 2013; Meehl et al., 2014,, 2016; Fyfe et al., 2016), thereby implying the potential effectiveness of the initialization in enhancing predictive skills. The observed SST changes display a spatial pattern similar to the negative phase of the Interdecadal Pacific Oscillation (IPO) (Trenberth and Hurrell, 1994) and suggest the primal contribution of the tropical Pacific, possibly in relation to enhanced ocean heat uptake (Meehl et al., 2011; Watanabe et al., 2013). In addition to the IPO, quasi-decadal-scale variability is dominantly observed in the tropical Pacific (Tourre et al., 2001), focusing on the timescales beyond the El Nino/Southern Oscillation (ENSO) fluctuation. The tropical Pacific SST represents a broad peak in power spectra on subdecadal to decadal timescales (Newman, 2007; Sun et al., 2017). It is characterized by an anticlockwise propagation of the tropical ocean heat content (OHC) anomaly (Hasegawa and Hanawa, 2003, 2007), showing an eastward evolution along the equator and a subsequent slow westward propagation in the off-equatorial North Pacific (Capotondi and Alexander, 2001; Tourre et al., 2005). Anomalous heat input and output in the meridional direction through wind-driven changes of the Pacific subtropical cells also contribute to low-frequency fluctuations of the equatorial Pacific OHC (Kleeman et al., 1999; Solomon et al., 2003; McPhaden and Zhang, 2004; Capotondi et al., 2005; Hasegawa et al., 2007). An advanced understanding of climate variability on shorter timescales than a decade, which can play an important role in controlling decadal climate predictability partly through a scale interaction, also contributes to improving our prediction skill, and the above challenge of decadal prediction sometimes targets the seasonal to decadal variability seamlessly. The climate prediction community has much experience on seasonal to interannual predictions over the tropical Pacific, such as the ENSO prediction, while we have an insufficient knowledge of subdecadal to decadal climate prediction. Here we examine the subdecadal climate predictability and variability, with a focus on 3-year-running mean states over the tropical Pacific as a key area controlling the global climate, with a goal towards filling a gap in seasonal to decadal predictions. We describe the characteristics of the subdecadal variation of the Pacific climate, that is distinctively observed in the 2000s, when the CMIP5 decadal hindcasts particularly show low predictability overall, and discuss its predictability in the initialized hindcasts. Observed and modelled datasets We examine gridded objective analysis and reanalysis datasets as good proxies of the observations; the ocean temperature and OHC are derived from an objective analysis (Ishii and Kimoto, 2009) used for the initialization in our decadal hindcast experiments as described below, and the SAT is taken from the HadCRUT3v dataset and the surface winds in the ERA-Interim reanalysis. We also analyse sets of decadal hindcasts with initialization performed in line with the CMIP5 protocol. As described in Mochizuki et al. (2012) and Tatebe et al. (2012), we perform decadal hindcasts using version 5 of the Model for Interdisciplinary Research on Climate (Watanabe et al., 2010). The initialization is based on an anomaly data assimilation for upper-ocean temperature and salinity of the above ocean objective analysis (Ishii and Kimoto, 2009). We perform two-member data assimilation runs with different initial conditions on 1 January 1945, that are derived from ensembles of the so-called historical simulation (e.g. global warming simulation without assimilating any ocean observation). The model temperature and salinity anomalies in the upper 3000 m depth are forced to approach the observations using an incremental analysis update method (Bloom et al., 1996). Using the results of data assimilation runs as initial conditions, we perform sets of 10-year-long, 6-member ensemble hindcast experiments for every year between 1 January 1961 and 31 December 2010 (Chikamoto et al., 2012; Mochizuki et al., 2012). The six initial conditions are derived from two sets of three snapshots of data assimilation runs, using a lagged average forecast method, with 3-month intervals: for example, when the ensemble hindcast experiment starts from 1 January 2001, we define the initial conditions using the snapshots of the assimilation results on 1 July 2000, 1 October 2000, and 1 January 2001. Note that 3-month intervals are practically effective to obtain ensembles of initial perturbations for decadal prediction (Mochizuki et al., 2012) and that three categories of lagged initial time and two data assimilation runs hardly show relative merits in hindcast skill. Hindcast skills on subdecadal timescales Before analysing the observations, we explore our hindcast skills on subdecadal timescales by using root-mean-squared error (RMSE) of 3-year-running mean data as a useful indicator (Figure 1). Even though we perform anomaly data assimilation, we have detected systematic errors (e.g. climate drift) that display a specific spatiotemporal structure in the hindcast data (Chikamoto et al., 2012). Therefore, we analyse anomalies relative to the climatology that is calculated as a function of lead time at each grid point during 1971–2010. As we performed decadal hindcasts with the initialization starting on 1 January of each year, we are able to discuss temporal fluctuation (i.e. modulation) of the hindcast skill for a specific lead time. The initial conditions well reproduce the observed climate states for both the internal and forced climate variability (Tatebe et al., 2012), but the information of internal variability given by the initialization is lost as lead time increases. The RMSE averaged over 1971–2010 at a lead time of 4 years reaches a saturation value (Figure 2) and consequently the hindcasts at a lead time over 5 years have similar skill to the historical simulation (Mochizuki et al., 2012). Figure 1. Open in new tabDownload slide (a) RMSEs in the global averages of the annual-mean SAT anomalies derived from the MIROC5 decadal hindcasts for specific lead time. (b) The same as in panel a, except for the averages over the NINO3 area (i.e. 5°S 5°N, 90°W 150°W). Figure 2. Open in new tabDownload slide (a) RMSEs in the global averages of the annual-mean SAT anomalies derived from the MIROC5 decadal hindcasts for specific lead time. Values for each year during 2001–2009 and the averages during 1971–2010 are plotted on the far right. (b) The same as in panel a, except for the averages over the NINO3 area. In the 2000s, in addition to the overall low quality of the SAT hindcasts, due to insufficiently reproduced slowdown of the global warming tendency (Mochizuki et al., 2012), a close examination suggests subdecadal-scale modulation of RMSE values (Figures 1 and 2). For example, the hindcasts at the lead time of 2–4 years simulate relatively large errors ∼2004 and small errors ∼ 2008. These hindcasted RMSEs in global mean SAT, including their subdecadal modulation, are tightly linked to the tropical Pacific climate states (Figures 1b and 2b) (Kosaka and Xie, 2013). At a lead time of 3 years, for example, the correlation coefficients of RMSEs between the global mean SAT and the tropical Pacific SAT are 0.59 in the 2000s and 0.70 during the period 1971–2010. Therefore, in the following sections, we attempt to clarify the characteristics of the observed and hindcasted subdecadal fluctuations in linearly detrended anomalies, with a focus on the tropical Pacific. Note that large RMSEs are also simulated in the mid-1970 (Figure 1). But they are modulated with lead time differently from those in the 2000s and primarily due to model errors found in our historical simulation which poorly reproduces a low SAT anomaly prior to the so-called late-1970s climate shift (Trenberth and Hurrell, 1994). In contrast, RMSEs are simulated to be smaller in the 1980s than in other decades (Figure 1), which may lead to obscurer relationship in RMSEs between the global mean SAT and the tropical Pacific SAT. At a lead time of several years, in particular, the small errors in the tropical Pacific can be closely related to high skill in hindcasting subdecadal OHC variability as suggested below. Observed and hindcasted subdecadal variations in the 2000s Even in the 2000s when the global warming tendency was suppressed as a decadal average, the observed upper 300 m OHC did not maintain the negative ENSO-like zonal contrast along the equator throughout the decade but showed a subdecadal fluctuation (Figure 3). At the beginning of the 21st century (i.e. in 2000), the 3-year means of OHC (and SST) along the equator were above normal over the western Pacific and are below normal over the central and eastern Pacific. The warm water in the western equatorial Pacific slowly moved eastward, and the OHC around the dateline warmed a couple of years later. This warm water no longer moves further eastward and stays in the central equatorial Pacific. Longitude-time sections of the OHCs along specific latitudes (Figure 4a) indicate that, at the same time, positive OHC anomalies are formed in the off-equatorial central Pacific areas (e.g. 15 degrees of the North and South Pacific). These high OHCs slowly move westward (c.f. Capotondi and Alexander, 2001; Tourre et al., 2005) and may contribute to the rapid decay of the anomalous equatorial warming in the central Pacific. The subsequent rising of the western Pacific SST can possibly contribute to the reemergence of a negative ENSO-like contrast between the eastern and western equatorial OHCs at the end of the 2000s. Figure 3. Open in new tabDownload slide Longitude-depth sections of 3-year mean ocean temperatures (contour) and anomalies (shade) observed over the equatorial Pacific. Figure 4. Open in new tabDownload slide (a) Temporal changes of 3-year mean OHC anomalies in the upper 300 m depth observed in the 2000s. Four panels indicate the longitude-time sections along 15°S 15°N and the equator and the latitude-time section averaged over 150°E 120°W. (b) The same as in panel a, except for the ensemble means of the initialized hindcasts starting on 1 January 2001. Hatched areas represent significant anomalies at the 90% confidence levels. (c) The same as in panel b, except for the initialized hindcasts starting on 1 January 2005. However, the initialized hindcasts starting on 1 January 2001 represent a different subdecadal variation than the observations (Figure 4b). While the initial high OHC in the western Pacific moves eastward as observed, it does not stay in the central Pacific. It reaches the eastern edge of the equatorial Pacific Ocean and may subsequently contribute to the high OHC moving westward particularly in the off-equatorial North Pacific by representing an anticlockwise propagation (c.f. Hasegawa and Hanawa, 2003, 2007). The anomalous heat input and output in the meridional direction work to generate these warming and cooling tendencies along the equator, respectively (c.f. Jin, 1997; Capotondi et al., 2005; Hasegawa et al., 2007). While the hindcasts show subdecadal variability in OHC, they fail to capture the spatial pattern in OHC anomalies, with warm anomalies being simulated to far east as compared to observation. The inability of the model to correctly evolve the 2001 initialization signal limits predictive skill to only a couple of years. The equatorial SAT hindcasts in 2004 actually show the largest RMSE value at a lead time of 3 years (Figure 2b). The termination of the eastward movement of the equatorial high OHC and the generation of the off-equatorial high OHC during 2002–2005 reveal key issues in improving our hindcast skills. When starting the initialized hindcasts on 1 January 2005 (Figure 4c), the initial conditions represent the high OHC in the equatorial and off-equatorial areas of the central Pacific, due to the direct assimilation of the ocean temperature, and the model reproduces the subsequent temporal evolutions in the observed OHCs over the tropical Pacific. The high OHC in the equator does not move further eastward in the hindcast calculation, in good agreement with the observations. The model also represents slow westward movement of the off-equatorial high OHCs, while the negative anomaly of the equatorial OHC at the end of the 2000s may not satisfy a statistically significant level in a one-side Student t-test. When the initial conditions represent the high OHCs observed in the tropical Pacific in the mid-2000s, we have the ability to predict the subsequent climate variations observed in the latter half of the 2000s and to demonstrate a relatively long predictive skill with a small RMSE value (Figures 1b and 2b). The equatorial SAT hindcasts in 2008 show the relatively small RMSE value at a lead time of 3 years (Figure 2b). Comparison between the 2000s and the preceding decades The initialized hindcasts over the preceding decades, for example, the decadal hindcasts starting on 1 January 1981 and 1991, also simulate periodic changes of the equatorial Pacific OHC on subdecadal timescales (Figure 5c and d), in a similar manner to those in the 2000s (Figure 4b). However, in these periods, the observed eastward adjustment of the equatorial OHC anomalies also reaches the eastern edge and the observed OHC in the tropical Pacific shows a periodic fluctuation (Figure 5a and b) in good agreement with the model. This highlights how the warm waters observed in the equatorial and off-equatorial areas during 2002–2005 show a quite distinct spatial evolution of subdecadal OHC variability as compared to the preceding decades. Figure 5. Open in new tabDownload slide (Upper panels) Longitude-time sections of 3-year mean OHC anomalies in the upper 300 m depth observed in (a) the 1980s and (b) the 1990s. (Lower panels) The same as in the upper panels, except for the ensemble means of the initialized hindcasts starting on (c) 1 January 1981 and (d) 1 January 1991. Hatched areas represent significant anomalies at the 90% confidence level. The slow eastward movement of OHC anomalies in the 1980s was apparent in the clear lagged relationships of the OHC anomalies timeseries averaged over the western, central, and eastern equatorial Pacific (Figure 6c), which was accompanied by zonal wind changes at the sea surface (Figure 6d). Over the off-equatorial North Pacific in the 1980s (Figure 6a and b), the zonally uniform changes of the surface winds and OHC suggest anomalous heat input/output between the tropics and the extratropics (right panel of Figure 4b) (c.f. Capotondi et al., 2005; Hasegawa et al., 2007). In contrast, in the 2000s, the OHC and surface winds in the above three areas of the equatorial Pacific represent simultaneous phase reversals (Figure 6c and d). In addition, the western and central off-equatorial areas clearly show an out-of-phase relationship in both the OHC and surface wind curl anomalies (Figure 6a and b). Figure 6. Open in new tabDownload slide (a) The observed OHC anomalies in the upper 300 m depth along 15°N, averaged over the western (140°E 160°E) and central (130°W 170°W) Pacific, respectively. (b) The same as in panel a, except for the surface wind curl anomalies taken from the ERA-I reanalysis. Positive and negative values correspond to anticyclonic and cyclonic wind anomalies, respectively. (c) The observed OHC anomalies in the upper 300 m depth along the equator, averaged over the western (140°E 160°E), central (130°W 170°W), and eastern (90°W 150°W) Pacific, respectively. (d) The zonal wind anomalies at the sea surface along the equator taken from the ERA-I reanalysis, averaged over the western (150°E 180°), central (130°W 170°W), and eastern (110°W 140°W) Pacific, respectively. Composite maps based on the OHC anomalies in the western equatorial Pacific (black line in Figure 6c) also illustrate clear differences among the periods (Figure 7), albeit these maps should be regarded as schematics due to the small number of samples. When the negative OHC anomaly is observed in the western equatorial Pacific, for example, the high OHC anomalies in the 2000s are observed to be strongest in the central Pacific, where the warm water remained during 2002–2005, as described above (Figure 7c). However, in the preceding decade, the eastern OHC anomalies were strongest due to the eastern reach of the warm water (Figure 7a). Figure 7. Open in new tabDownload slide (a) Composited map of the observed OHC (shades) and SST (contours). Plotted values are the differences between the averages during 1981–1983, 1985–1987, and 1991–1993 and those during 1984–1986 and 1988–1990. (b) The same as in panel a, except for the surface wind curl (shades) and surface zonal wind (contour). Positive and negative values of the surface wind curl correspond to anticyclonic and cyclonic anomalies, respectively. (c and d) The same as in panels a and b, except for the differences between the averages during 2002–2004 and those during 1998–2000 and 2007–2009. Corresponding to the differences in the location of the warm water, the zonal winds along the equator in the central Pacific indicate an anomalous convergence at the sea surface in the 2000s (Figure 7d), which was different from the broadly enhanced trade wind in the preceding decades (Figure 7b). The equatorial anomalies in Figure 7c are similar to those of the so-called central Pacific ENSO (Ashok et al., 2007; Kug et al., 2009), while the noticeably high OHC (and SST) is also observed in the off-equatorial area, together with strong surface wind curl aloft on subdecadal timescales (Figure 7c and d). These composites suggest that deepening of the ocean thermocline works to realize the high OHC in the off-equatorial area (Figure 6a) as a dynamical response to the strong surface wind curl (Figure 6b). The surface wind curl anomaly which coincides with the zonal wind anomaly along the equator can play an important role in controlling the subdecadal variations of the off-equatorial OHC. Detailed physical processes in relation to the OHC and the surface wind curl in the off-equatorial area require further investigations. Concluding remarks We have found subdecadal climate variability that is distinctive in the 2000s over the tropical Pacific and rarely observed in other decades. It is characterized by the termination of the eastward movement of the equatorial high OHC anomaly and the generation of the off-equatorial high OHC anomaly during 2002–2005. It is additionally characterized by accompanying surface wind changes, such as the strong Pacific trade wind along the equator and the anticyclonic anomaly probably working to deepen the ocean thermocline in the off-equatorial areas. Our CMIP5 decadal hindcasts indicate that these phenomena are difficult to predict even several years in advance, since the model usually simulates periodic fluctuation in the tropical OHC, as observed in the preceding decades. Consequently, it can contribute to modulating the hindcast skill of the tropical and global climates on subdecadal timescales. Further studies to clarify the origin of these subdecadal variations in the equatorial and off-equatorial Pacific climate in the 2000s are now underway by both observational and modelling approaches. In particular, as well as a possible feedback in the Pacific climate system, possible remote influence from other oceans may contribute to surface wind changes in the tropical Pacific. For example, recent studies have indicated that the tropical Pacific winds, SST and OHC are partly affected by the Atlantic climate on decadal to interdecadal (England et al., 2014; McGregor et al., 2014; Chikamoto et al., 2015, 2016; Kucharski et al., 2016; Li et al., 2016) and seasonal to interannual timescales (Ham et al., 2013; Wang et al., 2017). Several studies have also indicated that the Indian Ocean contributes to controlling the Pacific trade winds (Luo et al., 2012; Hamlington et al., 2014; Han et al., 2014; Lee et al., 2015; Mochizuki et al., 2016). An understanding of the processes governing subdecadal variability with noticeable dependency on the periods, as part of the seasonal, interannual to decadal variability, should greatly help us predict the tropical Pacific climate and tightly linked global climate changes. Acknowledgements The authors wish to thank M. Kimoto and Y. Chikamoto, T. Nishimura, H. Tatebe, and M. Ishii, and H. Kanai for their support in performing the experiments. Funding This work is supported by the Integrated Research Program for Advancing Climate Models and by the Grant-in-Aid for Scientific Research program (C-17K05661). References Ashok K. , Behera S. K., Rao S. A., Weng H., Yamagata T. 2007 . El Nino Modoki and its possible teleconnection . Journal of Geophysical Research , 112 : C11007. Google Scholar Crossref Search ADS WorldCat Bloom S. C. , Takacs L., da Silva A. M., Ledvina D. 1996 . Data assimilation using incremental analysis updates . Monthly Weather Review , 124 : 1256 – 1271 . Google Scholar Crossref Search ADS WorldCat Capotondi A. , Alexander M. A. 2001 . Rossby waves in the tropical North Pacific and their role in decadal thermocline variability . Journal of Physical Oceanography , 31 : 3496 – 3515 . Google Scholar Crossref Search ADS WorldCat Capotondi A. , Alexander M. A., Deser C., McPhaden M. 2005 . Anatomy and decadal evolution of the Pacific subtropical-tropical cells (STCs) . Journal of Climate , 18 : 3739 – 3758 . Google Scholar Crossref Search ADS WorldCat Chikamoto Y. , Kimoto M., Ishii M., Watanabe M., Mochizuki T., Tatebe H., Sakamoto T. T., et al. . 2012 . Predictability of a stepwise shift in Pacific climate during the late 1990s in hindcast experiments using MIROC . Journal of the Meteorological Society of Japan , 90A : 1 – 21 . Google Scholar Crossref Search ADS WorldCat Chikamoto Y. , Mochizuki T., Timmermann A., Kimoto M., Watanabe M. 2016 . Potential tropical Atlantic impacts on Pacific decadal climate trends . Geophysical Research Letters , 43 : 7143 – 7151 . Google Scholar Crossref Search ADS WorldCat Chikamoto Y. , Timmerman A., Luo J.-J., Mochizuki T., Kimoto M., Watanabe M., Ishii M., et al. . 2015 . Skilful multi-year predictions of tropical trans-basin climate variability . Nature Communications , 6 : 6869. Google Scholar Crossref Search ADS PubMed WorldCat Doblas-Reyes F. J. , Andreu-Burillo I., Chikamoto Y., Garcia-Serrano J., Guemas V., Kimoto M., Mochizuki T., et al. . 2013 . Initialized near-term regional climate change prediction . Nature Communications , 4 : 1715. Google Scholar Crossref Search ADS PubMed WorldCat Doblas-Reyes F. J. , Balmaseda M. A., Weisheimer A., Palmer T. N. 2011 . Decadal climate prediction with the ECMWF coupled forecast system: impact of ocean observations . Journal of Geophysical Research , 116 : D19111. Google Scholar Crossref Search ADS WorldCat Easterling D. R. , Wehner M. F. 2009 . Is the climate warming or cooling? Geophysical Research Letters , 36 : L08706. Google Scholar Crossref Search ADS WorldCat England M. H. , McGregor S., Spence P., Meehl G. A., Timmermann A., Cai W., Gupta A. S., et al. . 2014 . Recent intensification of wind-driven circulation in the Pacific and the ongoing warming hiatus . Nature Climate Change , 4 : 222 – 227 . Google Scholar Crossref Search ADS WorldCat Eyring V. , Bony S., Meehl G. A., Senior C., Stevens B., Stouffer R. J., Taylor K. E. 2015 . Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organisation . Geoscientific Model Development Discussions , 8 : 10539 – 10583 . Google Scholar Crossref Search ADS WorldCat Fyfe J. C. , Meehl G. A., England M. H., Mann M. E., Santer B. D., Flato G. M., Hawkins E., et al. . 2016 . Making sense of the early-2000s warming slowdown . Nature Climate Change , 6 : 224 – 228 . Google Scholar Crossref Search ADS WorldCat Ham Y.-G. , Kug J.-S., Park J.-Y., Jin F.-F. 2013 . Sea surface temperature in the north tropical Atlantic as a trigger for El Nino/Southern Oscillation events . Nature Geoscience , 6 : 112 – 116 . Google Scholar Crossref Search ADS WorldCat Hamlington B. D. , Strassburg M. W., Leben R. R., Han W., Nerem R. S., Kim K.-Y. 2014 . Uncovering an anthropogenic sea-level rise signal in the Pacific Ocean . Nature Climate Change , 4 : 782 – 785 . Google Scholar Crossref Search ADS WorldCat Han W. , Meehl G. A., Hu A., Alexander M. A., Yamagata T., Yuan D., Ishii M., et al. . 2014 . Intensification of decadal and multi-decadal sea level variability in the western tropical Pacific during recent decades . Climate Dynamics , 43 : 1357 – 1379 . Google Scholar Crossref Search ADS WorldCat Hasegawa T. , Hanawa K. 2003 . Decadal-scale variability of upper ocean heat content in the tropical Pacific . Geophysical Research Letters , 30 : 1272. Google Scholar OpenURL Placeholder Text WorldCat Hasegawa T. , Hanawa K. 2007 . Upper ocean heat content and atmospheric anomaly fields in the off-equatorial North Pacific related to ENSO . Journal of Oceanography , 63 : 561 – 572 . Google Scholar Crossref Search ADS WorldCat Hasegawa T. , Yasuda T., Hanawa K. 2007 . Generation mechanism of quasidecadal variability of upper ocean heat content in the equatorial Pacific Ocean . Journal of Geophysical Research , 112 : C08012. Google Scholar OpenURL Placeholder Text WorldCat Ishii M. , Kimoto M. 2009 . Reevaluation of historical ocean heat content variations with time-varying XBT and MBT depth bias corrections . Journal of Oceanography , 65 : 287 – 299 . Google Scholar Crossref Search ADS WorldCat Jin F.-F. 1997 . An equatorial ocean recharge paradigm for ENSO Part I: conceptual model . Journal of the Atmospheric Sciences , 54 : 811 – 829 . Google Scholar Crossref Search ADS WorldCat Kaufmann R. K. , Kauppi H., Mann M. L., Stock J. H. 2011 . Reconciling anthropogenic climate change with observed temperature 1998–2008 . Proceedings of the National Academy of Sciences of the United States of America , 108 : 11790 – 11793 . Google Scholar Crossref Search ADS PubMed WorldCat Keenlyside N. S. , Latif M., Jungclaus J., Kornblueh L., Roeckner E. 2008 . Advancing decadal-scale climate prediction in the North Atlantic sector . Nature , 453 : 84 – 88 . Google Scholar Crossref Search ADS PubMed WorldCat Kim H. , Webster P., Curry J. 2012 . Evaluation of short-term climate change prediction in multi-model CMIP5 decadal hindcasts . Geophysical Research Letters , 39 : L10701. Google Scholar OpenURL Placeholder Text WorldCat Kleeman R. , McCreary J. P., Klinger B. A. 1999 . A mechanism for generating ENSO decadal variability . Geophysical Research Letters , 26 : 1743 – 1746 . Google Scholar Crossref Search ADS WorldCat Kosaka Y. , Xie S.-P. 2013 . Recent global-warming hiatus tied to equatorial Pacific surface cooling . Nature , 501 : 403 – 407 . Google Scholar Crossref Search ADS PubMed WorldCat Kucharski F. , Ikram F., Molteni F., Farneti R., Kang I.-S., No H.-H., King M. P., et al. . 2016 . Atlantic forcing of Pacific decadal variability . Climate Dynamics , 46 : 2337 – 2351 . Google Scholar Crossref Search ADS WorldCat Kug J.-S. , Jin F.-F., An S.-I. 2009 . Two types of El Nino events: cold tongue El Nino and warm pool El Nino . Journal of Climate , 22 : 1499 – 1515 . Google Scholar Crossref Search ADS WorldCat Lee S.-K. , Park W., Baringer M. O., Gordon A. L., Huber B., Liu Y. 2015 . Pacific origin of the abrupt increase in Indian Ocean heat content during the warming hiatus . Nature Geoscience , 8 : 445 – 449 . Google Scholar Crossref Search ADS WorldCat Li X. , Xie S.-P., Gille S. T., Yoo C. 2016 . Atlantic-induced pan-tropical climate change over the past three decades . Nature Climate Change , 6 : 275 – 279 . Google Scholar Crossref Search ADS WorldCat Luo J.-J. , Sasaki W., Masumoto Y. 2012 . Indian Ocean warming modulates Pacific climate changes . Proceedings of the National Academy of Sciences of the United States of America , 109 : 18701 – 18706 . Google Scholar Crossref Search ADS PubMed WorldCat McGregor S. , Timmerman A., Stuecker M. F., England M. H., Merrifield M., Jin F.-F., Chikamoto Y. 2014 . Recent Walker circulation strengthening and Pacific cooling amplified by Atlantic warming . Nature Climate Change , 4 : 888 – 892 . Google Scholar Crossref Search ADS WorldCat McPhaden M. J. , Zhang D. 2004 . Pacific Ocean circulation rebounds . Geophysical Research Letters , 31 : L18301. Google Scholar OpenURL Placeholder Text WorldCat Meehl G. A. , Goddard L., Boer G., Burgman R., Branstator G., Cassou C., Corti S., et al. . 2014 . Decadal climate prediction: an update from the trenches . Bulletin of the American Meteorological Society , 94 : 243 – 267 . Google Scholar Crossref Search ADS WorldCat Meehl G. A. , Arblaster J. M., Fasullo J. T., Hu A., Trenberth K. E. 2011 . Model-based evidence of deep-ocean heat uptake during surface-temperature hiatus periods . Nature Climate Change , 1 : 360 – 364 . Google Scholar Crossref Search ADS WorldCat Meehl G. A. , Hu A., Arblaster J. M., Fasullo J., Trenberth K. E. 2013 . Externally forced and internally generated decadal climate variability associated with the Interdecadal Pacific Oscillation . Journal of Climate , 26 : 7298 – 7310 . Google Scholar Crossref Search ADS WorldCat Meehl G. A. , Hu A., Santer G. D., Xie S.-P. 2016 . Contribution of the Interdecadal Pacific Oscillation to twentieth-century global surface temperature trends . Nature Climate Change , 6 : 1005 – 1008 . Google Scholar Crossref Search ADS WorldCat Meehl G. A. , Hu A., Tebaldi C. 2010 . Decadal prediction in the Pacific region . Journal of Climate , 23 : 2959 – 2973 . Google Scholar Crossref Search ADS WorldCat Mochizuki T. , Chikamoto Y., Kimoto M., Ishii M., Tatebe H., Komuro Y., Sakamoto T. T., et al. . 2012 . Decadal prediction using a recent series of MIROC global climate models . Journal of the Meteorological Society of Japan , 90A : 373 – 383 . Google Scholar Crossref Search ADS WorldCat Mochizuki T. , Ishii M., Kimoto M., Chikamoto Y., Watanabe M., Nozawa T., Sakamoto T. T., et al. . 2010 . Pacific decadal oscillation hindcasts relevant to near-term climate prediction . Proceedings of the National Academy of Sciences of the United States of America , 107 : 1833 – 1837 . Google Scholar Crossref Search ADS PubMed WorldCat Mochizuki T. , Kimoto M., Watanabe M., Chikamoto Y., Ishii M. 2016 . Interbasin effects of the Indian Ocean on Pacific decadal climate change . Geophysical Research Letters , 43 : 7168 – 7175 . Google Scholar Crossref Search ADS WorldCat Newman M. 2007 . Interannual to decadal predictability of tropical and North Pacific sea surface temperatures . Journal of Climate , 20 : 2333 – 2356 . Google Scholar Crossref Search ADS WorldCat Pohlmann H. , Jungclaus J. H., Köhl A., Stammer D., Marotzke J. 2009 . Initializing decadal climate predictions with the GECCO oceanic synthesis: effect on the North Atlantic . Journal of Climate , 22 : 3926 – 3938 . Google Scholar Crossref Search ADS WorldCat Smith D. M. , Cusack S., Colman A. W., Folland C. K., Harris G. R., Murphy J. M. 2007 . Improved surface temperature prediction for the coming decade from a global climate model . Science , 317 : 796 – 799 . Google Scholar Crossref Search ADS PubMed WorldCat Solomon A. , McCreary J. P., Kleeman R., Klinger B. A. 2003 . Interannual and decadal variability in an intermediate coupled model of the Pacific region . Journal of Climate , 16 : 383 – 405 . Google Scholar Crossref Search ADS WorldCat Solomon S. , Rosenlof K. H., Portmann R. W., Daniel J. S., Davis S. M., Sanford T. J., Plattner G. K. 2010 . Contributions of stratospheric water vapor to decadal changes in the rate of global warming . Science , 327 : 1219 – 1223 . Google Scholar Crossref Search ADS PubMed WorldCat Sun C. , Kucharski F., Li J., Jin F.-F., Kang I.-S., Ding R. 2017 . Western tropical Pacific multidecadal variability forced by the Atlantic multidecadal oscillation . Nature Communications , 8 : 15998. Google Scholar Crossref Search ADS PubMed WorldCat Tatebe H. , Ishii M., Mochizuki T., Chikamoto Y., Sakamoto T. T., Komuro Y., Mori M., et al. . 2012 . Initialization of the climate model MIROC for decadal predictions by assimilating ocean hydrography . Journal of the Meteorological Society of Japan , 90A : 275 – 294 . Google Scholar Crossref Search ADS WorldCat Tourre Y. M. , Cibot C., Terray L., White W. B., Dewitte B. 2005 . Quasi-decadal and inter-decadal climate fluctuations in the Pacific Ocean from a CGCM . Geophysical Research Letters , 32 : L07710. Google Scholar OpenURL Placeholder Text WorldCat Tourre Y. M. , Rajagopalan B., Kushnir Y., Barlow M., White W. B. 2001 . Patterns of coherent ocean decadal and interdecadal climate signals in the Pacific basin during the 20th century . Geophysical Research Letters , 28 : 2069 – 2072 . Google Scholar Crossref Search ADS WorldCat Trenberth K. E. , Hurrell J. W. 1994 . Decadal atmosphere-ocean variations in the Pacific . Climate Dynamics , 9 : 303 – 319 . Google Scholar Crossref Search ADS WorldCat Wang L. , Yu J.-Y., Paek H. 2017 . Enhanced biennial variability in the Pacific due to Atlantic capacitor effect . Nature Communications , 8 : 14887. Google Scholar OpenURL Placeholder Text WorldCat Watanabe M. , Kamae Y., Yoshimori M., Oka A., Sato M., Ishii M., Mochizuki T., et al. . 2013 . Strengthening of ocean heat uptake efficiency associated with the recent climate hiatus . Geophysical Research Letters , 40 : 3175 – 3179 . Google Scholar Crossref Search ADS WorldCat Watanabe M. , Suzuki T., O’ishi R., Komuro Y., Watanabe S., Emori S., Takemura T., et al. . 2010 . Improved climate simulation by MIROC5: mean states, variability, and climate sensitivity . Journal of Climate , 23 : 6312 – 6335 . Google Scholar Crossref Search ADS WorldCat © International Council for the Exploration of the Sea 2019. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. © International Council for the Exploration of the Sea 2019.
Projected biophysical conditions of the Bering Sea to 2100 under multiple emission scenariosHermann, Albert J; Gibson, Georgina A; Cheng, Wei; Ortiz, Ivonne; Aydin, Kerim; Wang, Muyin; Hollowed, Anne B; Holsman, Kirstin K
doi: 10.1093/icesjms/fsz043pmid: N/A
Abstract A regional biophysical model is used to relate projected large-scale changes in atmospheric and oceanic conditions from CMIP5 to the finer-scale changes in the physical and biological structure of the Bering Sea, from the present through the end of the twenty-first century. A multivariate statistical method is used to analyse the results of a small (eight-member) dynamically downscaled ensemble to characterize and quantify dominant modes of variability and covariability among a broad set of biophysical features. This characterization provides a statistical method to rapidly estimate the likely response of the regional system to a much larger (63-member) ensemble of possible future forcing conditions. Under a high-emission [Representative Concentration Pathway 8.5 (RCP8.5)] scenario, results indicate that decadally averaged Bering Sea shelf bottom temperatures may warm by as much as 5°C by 2100, with associated loss of large crustacean zooplankton on the southern shelf. Under a lower emission scenario (RCP4.5), these effects are predicted to be approximately half their calculated change under the high emission scenario. Introduction Widespread change is anticipated for the Bering Sea (AK) under climate change, including substantial oceanographic warming that scales with future carbon mitigation scenarios (IPCC, 2013, 2014). Climate-driven changes to oceanographic conditions have the potential to propagate through the food web and impact fish and fisheries in the region (Holsman et al., 2018). The Bering Sea is a highly productive system that supports a wide diversity of species, some critically endangered, as well as multiple small coastal fishing communities that depend on subsistence harvest (Haynie and Huntington, 2016) and large-scale commercial fisheries that annually represent more than 40% of the U.S. commercial catch (Fissel et al., 2017). In this article, we report estimates of anticipated change to the physical and lower trophic level dynamics of the Bering Sea, derived both through application of dynamical model downscaling and through statistical projections based on those results. Overview of the Bering Sea ecosystem Prominent physical features of the Bering Sea include seasonal ice cover, strong advection of ice, and tidally generated biophysical domains. Ice formed each winter in the northern Bering Sea is advected to the southeast, where it gradually melts as it encounters warmer water and air temperatures. This southward advection contributes to the latitudinal salinity gradient of the Bering Sea and its interannual variability. A cross-shelf gradient in the vertical penetration of tidal mixing sets up distinct biophysical regimes with associated biota. Classically, the southeastern shelf is classified as having three biophysical domains: a vertically well-mixed inner shelf domain (ocean depth is between ∼0 and 50 m), a middle shelf domain (with ocean depth ∼50–100 m) which is well-mixed in the winter and has two distinct layers separated by a sharp thermocline in the summer, and an outer shelf domain (ocean depth ∼100–200 m) which is more gradually stratified (Kinder and Schumacher, 1981; Coachman, 1986; Kachel et al., 2002). A map of the region with these features identified is provided in Figure 1. Figure 1. Open in new tabDownload slide Bathymetry (m) with biophysical domains of the Bering Sea. Shown are inner shelf domain (0–50 m), middle shelf domain (50–100 m), outer shelf domain (100–200 m), shelf break (200–1000 m), and deep basin (>1000 m). Colour version of this figure is available online at ICESJM online. Distinct biological features of the Bering Sea ecosystem include ice algae as a potential food source to secondary producers, and strong benthic-pelagic coupling. Within the different biophysical regimes, the relative magnitude of pelagic vs. benthic pathways of carbon flux varies interannually, and is believed to be strongly influenced by the extent of seasonal ice through its effects on stratification (Hunt et al., 2002, 2011). The relative importance of pelagic vs. benthic pathways is likely to shift under the influence of global warming, partially through its impact on seasonal ice extent in the Bering Sea. Field data suggest that cold temperatures in the Bering Sea lead to an increase in large crustacean zooplankton, favoured as food items by juvenile pollock in the fall season (Coyle et al., 2011). The present hydrography and seasonal ice-driven climatology of the Bering Sea result in a highly productive ecosystem, with plankton biomass ultimately supporting large populations of shellfish and finfish (and major fisheries), marine birds, and marine mammals (Sigler et al., 2016). Such intense production derives, in part, from a broad shelf with strong tidally induced mixing, a plentiful supply of the micro-nutrient iron, and seasonal stratification which maintains the phytoplankton in the euphotic zone, adjacent to a deep, macronutrient-rich basin. Interannual variation in winter ice extent over the Bering Sea modulates annual variability in productivity in the system. A cold period in the Bering Sea from 2006 to 2011 (Stabeno et al., 2012) was followed by a return to warmer conditions, with reduced ice (Stabeno et al., 2016, 2017) and attendant changes in primary and secondary productivity (Sigler et al., 2016). In previous studies, a model-based multivariate analysis was used to help explore the relationships between physical and biological factors on the Bering Sea shelf (Hermann et al., 2013). The analysis suggested that the Bering Sea shelf may not respond uniformly to changes in climate forcing. For example, large crustacean zooplankton (lcz) are negatively correlated with temperature on the outer, southwestern shelf, and positively correlated to temperature on the inner, northeastern shelf. Areas of positive correlation tended to correspond with those areas with greatest change in ice cover. As in the revised Oscillating Control Hypothesis of Hunt et al. (2011), the ratio of large to total zooplankton was enhanced at lower temperatures. On the outer shelf, higher temperatures may be leading to reduced lcz production either through effects on stratification (and hence nutrient limitation), or through direct effects of temperature on growth, respiration, predation, and vertical migration. Changes on the northern shelf may involve a complex interplay of light and nutrient limitation effects, as modulated by a reduction in the duration of seasonal ice cover. Ice dynamics of the Bering Sea have been explored in both observational and modelling studies (Stabeno et al., 2010; Danielson et al., 2011; Cheng et al., 2014; Li et al., 2014a, b; Sullivan et al., 2014). Ice is formed seasonally in the northern Bering Sea and is advected southward, resulting in a net transfer of freshwater from north to south. Heat budgets from these studies have underscored the importance of sensible heat flux between the atmosphere and the ice in the northern Bering, and between the ocean surface and the ice in the southern Bering, where the ice edge retreats each spring. In previous publications (Hermann et al., 2013, 2016a), we projected future Bering Sea conditions out to 2040 under an intermediate carbon mitigation scenario (i.e. A1B) from phase 3 of the Coupled Model Intercomparison (CMIP3; Meehl et al., 2007). Here, we extend this work to include a larger ensemble of global models under unmitigated (RCP 8.5) and moderate (RCP 4.5) mitigation scenarios from phase 5 of the CMIP (CMIP5) out to 2100 (Taylor et al., 2012). The goals of this contribution are: (i) to summarize the results of dynamical biophysical downscaling of a limited ensemble of CMIP5 models to the Bering Sea; (ii) to describe a multivariate method which allows extrapolation of those dynamically obtained results to a much larger ensemble of CMIP5 output; (iii) to apply that method in order to obtain a more robust estimate of the regional response of the Bering Sea to global change. Output from these simulations are being used in fisheries models, both to project stocks and for management-strategy evaluation, as part of NOAA’s Alaska Climate Integrated Modeling Project (ACLIM; Hollowed et al., in prep.). Methods Dynamical downscaling regional model The regional model used for these analyses (Bering10K) is identical to the one described in Hermann et al. (2016a). In that earlier publication, the regional model was driven by results from three global simulations from CMIP3 under emission scenario A1B. Here we take a similar approach, using global simulations from CMIP5 under emission scenarios RCP4.5 and RCP8.5. Collectively these downscaling runs span a wide range of model types and assumed future behaviours of mankind (e.g. RCP 4.5 vs. RCP 8.5 emission scenarios). Major features of the regional model are as follows. The model is based on the Regional Ocean Modeling System version 3.2. ROMS is a sigma-coordinate model with curvilinear horizontal coordinates; a description of basic features and implementation can be found in Haidvogel et al. (2008) and Shchepetkin and McWilliams (2005). The Bering10K regional grid has approximately 10 km horizontal resolution, with ten vertical levels. Fine-scale bathymetry is based on ETOPO5 and supplementary data sets as described in Danielson et al. (2011); smoothing of that bathymetry was utilized for numerical stability. Any oceanic regions shallower than 10 m were set to be 10 m deep. Mixing is based on the algorithms of Large et al. (1994). Both ice (Budgell, 2005) and tidal dynamics are included in this model; the explicit inclusion of tidal flows allows tidally generated mixing and tidal residual flows to develop. Freshwater runoff was applied by freshening of the surface salinity field within a few grid points of the coastline, using climatological monthly runoff values based on Dai et al. (2009). Bulk forcing, based on algorithms of Large and Yeager (2008), were used to relate winds, air temperature, relative humidity, and downward shortwave and longwave radiation to surface stress and the net transfers of sensible heat, latent heat, net shortwave, and net longwave radiation through the sea surface. Further detail of model tuning, implementation, and biases are available in Hermann et al. (2016a). The lower trophic level dynamics (Nutrient-Phytoplankton-Zooplankton, NPZ) model is described in detail in Gibson and Spitz (2011). Briefly, this model includes two size categories of phytoplankton and ice plankton, and distinguishes among microzooplankton, copepods, neocalanus, and euphausiids, as well as jellyfish, benthic detritus, and epibenthos. Limiting nutrients are nitrate, ammonium, and dissolved iron. Metabolic and grazing rates are temperature dependent, which leads to substantially different food web structure under cold vs. warm conditions. Many of the state variables from this model were utilized in our analyses here. Results from hindcasts and forecasts with this model are described in Hermann et al. (2013, 2016a) and Ortiz et al. (2016). Global models used for downscaling The three CMIP5 models used as the basis for this study are: GFDL-ESM2M (Dunne et al., 2012), CESM (Kay et al., 2015), and MIROC (Watanabe et al., 2011). These were chosen based on: (i) performance in the Bering Sea under present conditions; (ii) the desire for a representative subset of CMIP5 members, spanning their range of variability; (iii) the ready availability of both physical and biogeochemical output. Spatial and temporal resolution of the output from the three chosen global models is shown in Table 1. Knutti et al. (2013) performed a dendogram analysis of the reference state of CMIP3/CMIP5 models; the GFDL-ESM2M, MIROC, and CESM models in fact span a broad range of global patterns for precipitation and SST. A regional summary of CMIP5 output was obtained through the NOAA climate change web portal (https://www.esrl.noaa.gov/psd/ipcc/cmip5/). Table 1. Spatial and temporal resolution of the three IPCC models. Model . GFDL-ESM2M . CESM . MICOC . Ocean 0.33–1.0° latitude 1.0° longitude 50 levels vertically Monthly nominally 1° latitude nominally 1° longitude 60 levels vertically Monthly 0.56° near equator, 1.71° latitude at the poles 1.4 o longitude 44 levels vertically Monthly Atmosphere 2.0° latitude 2.5° longitude 6 hourly 0.94° latitude 1.25° longitude Daily 2.79° latitude 2.81° longitude Daily Model . GFDL-ESM2M . CESM . MICOC . Ocean 0.33–1.0° latitude 1.0° longitude 50 levels vertically Monthly nominally 1° latitude nominally 1° longitude 60 levels vertically Monthly 0.56° near equator, 1.71° latitude at the poles 1.4 o longitude 44 levels vertically Monthly Atmosphere 2.0° latitude 2.5° longitude 6 hourly 0.94° latitude 1.25° longitude Daily 2.79° latitude 2.81° longitude Daily Open in new tab Table 1. Spatial and temporal resolution of the three IPCC models. Model . GFDL-ESM2M . CESM . MICOC . Ocean 0.33–1.0° latitude 1.0° longitude 50 levels vertically Monthly nominally 1° latitude nominally 1° longitude 60 levels vertically Monthly 0.56° near equator, 1.71° latitude at the poles 1.4 o longitude 44 levels vertically Monthly Atmosphere 2.0° latitude 2.5° longitude 6 hourly 0.94° latitude 1.25° longitude Daily 2.79° latitude 2.81° longitude Daily Model . GFDL-ESM2M . CESM . MICOC . Ocean 0.33–1.0° latitude 1.0° longitude 50 levels vertically Monthly nominally 1° latitude nominally 1° longitude 60 levels vertically Monthly 0.56° near equator, 1.71° latitude at the poles 1.4 o longitude 44 levels vertically Monthly Atmosphere 2.0° latitude 2.5° longitude 6 hourly 0.94° latitude 1.25° longitude Daily 2.79° latitude 2.81° longitude Daily Open in new tab Air temperatures over the eastern Bering Sea from these three models span ∼80% of the range of CMIP5 results, with the CESM model being closest to the ensemble average (Figure 2). These models likewise span a wide range of ice climatologies for the Bering Sea, where that climatology was calculated using observations and model output for the period 1980–1999; GFDL has more ice than observed, MIROC has less ice than observed, and CESM has close to observed climatological ice. The models diverge considerably in their estimates of Bering Sea air temperature change after 2040. For the CMIP5 downscaling simulations, we continuously span years 2006–2100. Oceanic physical boundary conditions for the Bering10K model were derived from the CMIP5 models’ monthly averages, while atmospheric forcing is either daily (MIROC, CESM) or 6 hourly (GFDL). For the primary downscaling runs of the CMIP5 models (GFDL, MIROC, and CESM), biological ICs and BCs were derived from present monthly climatologies (or assumed near-zero) as described in Hermann et al. (2013, 2016a). Yearly (GFDL) and monthly (CESM) nitrate and dissolved iron values were obtained from the global biogeochemical projections. These were interpolated as the ICs and BCs for two additional realizations of Bering10K. These two additional downscaling runs (GFDL_RCP8.5_BIO, CESM_RCP8.5_BIO) utilized the same physical ICs, BCs and forcing as the runs with purely climatological biological conditions (GFDL_RCP8.5, CESM_RCP8.5). These additional runs sample additional structural uncertainty of the projections, as they utilize projected changes in the large-scale biological conditions (e.g. those described in Rykaczewski and Dunne, 2010). Variables analysed As a summary of model behaviour, we examined a broad spectrum of physical and biological variables. A full listing of the variables chosen, with their units, are shown in Table 2. The variables chosen are similar to those used in the multivariate analysis of Hermann et al. (2013); a major new element here is the inclusion of atmospheric forcing variables. Ocean variables analysed include relevant surface and bottom values, as well as vertical averages. A derived variable used to summarize zooplankton biomass, here termed “large crustacean zooplankton,” is as follows: lcz = cope + nca + eup where lcz, cope, nca, and eup refer to large crustacean zooplankton, copepods, neocalanus, and euphausiids, respectively (Table 2). Table 2. Variables used in the multivariate analysis of the simulations, with short names and units. Surface temperature sst oC Bottom temperature sbt oC Surface salinity sss psu Ice cover iceco fractional area Mixed layer depth mld m (positive up coordinates; hence, negative change denotes deepening MLD) Vertical mixing (depth avg.) akts m2 s−1 Nitrate + ammonium (depth avg.) nut mgN m−3 Ice phytoplankton iceph mgC m−2 Small plus large phytoplankton (depth avg.) phyt mgC m−3 Microzooplankton (depth avg.) mzoo mgC m−3 Small copepods (depth avg.) cope mgC m−3 Neocalanus (depth avg.) nca mgC m−3 Euphausiids (depth avg.) eup mgC m−3 Benthic detritus detbe mgC m−2 Benthic infauna benth mgC m−2 Sea surface height ssh m Sea surface cross-shelf velocity utop m s−1 Sea surface along-shelf velocity vtop m s−1 Air temperature tair oC Air pressure pair Pa Specific humidity qair kg kg−1 Zonal wind uwind m s−1 Meridional wind vwind m s−1 Downward longwave radiation lwrad W m−2 Downward shortwave radiation swrad W m−2 Surface temperature sst oC Bottom temperature sbt oC Surface salinity sss psu Ice cover iceco fractional area Mixed layer depth mld m (positive up coordinates; hence, negative change denotes deepening MLD) Vertical mixing (depth avg.) akts m2 s−1 Nitrate + ammonium (depth avg.) nut mgN m−3 Ice phytoplankton iceph mgC m−2 Small plus large phytoplankton (depth avg.) phyt mgC m−3 Microzooplankton (depth avg.) mzoo mgC m−3 Small copepods (depth avg.) cope mgC m−3 Neocalanus (depth avg.) nca mgC m−3 Euphausiids (depth avg.) eup mgC m−3 Benthic detritus detbe mgC m−2 Benthic infauna benth mgC m−2 Sea surface height ssh m Sea surface cross-shelf velocity utop m s−1 Sea surface along-shelf velocity vtop m s−1 Air temperature tair oC Air pressure pair Pa Specific humidity qair kg kg−1 Zonal wind uwind m s−1 Meridional wind vwind m s−1 Downward longwave radiation lwrad W m−2 Downward shortwave radiation swrad W m−2 Variables shown in bold were log-transformed prior to use in the statistical analyses. Open in new tab Table 2. Variables used in the multivariate analysis of the simulations, with short names and units. Surface temperature sst oC Bottom temperature sbt oC Surface salinity sss psu Ice cover iceco fractional area Mixed layer depth mld m (positive up coordinates; hence, negative change denotes deepening MLD) Vertical mixing (depth avg.) akts m2 s−1 Nitrate + ammonium (depth avg.) nut mgN m−3 Ice phytoplankton iceph mgC m−2 Small plus large phytoplankton (depth avg.) phyt mgC m−3 Microzooplankton (depth avg.) mzoo mgC m−3 Small copepods (depth avg.) cope mgC m−3 Neocalanus (depth avg.) nca mgC m−3 Euphausiids (depth avg.) eup mgC m−3 Benthic detritus detbe mgC m−2 Benthic infauna benth mgC m−2 Sea surface height ssh m Sea surface cross-shelf velocity utop m s−1 Sea surface along-shelf velocity vtop m s−1 Air temperature tair oC Air pressure pair Pa Specific humidity qair kg kg−1 Zonal wind uwind m s−1 Meridional wind vwind m s−1 Downward longwave radiation lwrad W m−2 Downward shortwave radiation swrad W m−2 Surface temperature sst oC Bottom temperature sbt oC Surface salinity sss psu Ice cover iceco fractional area Mixed layer depth mld m (positive up coordinates; hence, negative change denotes deepening MLD) Vertical mixing (depth avg.) akts m2 s−1 Nitrate + ammonium (depth avg.) nut mgN m−3 Ice phytoplankton iceph mgC m−2 Small plus large phytoplankton (depth avg.) phyt mgC m−3 Microzooplankton (depth avg.) mzoo mgC m−3 Small copepods (depth avg.) cope mgC m−3 Neocalanus (depth avg.) nca mgC m−3 Euphausiids (depth avg.) eup mgC m−3 Benthic detritus detbe mgC m−2 Benthic infauna benth mgC m−2 Sea surface height ssh m Sea surface cross-shelf velocity utop m s−1 Sea surface along-shelf velocity vtop m s−1 Air temperature tair oC Air pressure pair Pa Specific humidity qair kg kg−1 Zonal wind uwind m s−1 Meridional wind vwind m s−1 Downward longwave radiation lwrad W m−2 Downward shortwave radiation swrad W m−2 Variables shown in bold were log-transformed prior to use in the statistical analyses. Open in new tab Analysis methods Here we describe both univariate and multivariate analyses of the dynamically downscaled results. A summary of the CMIP5 realizations used in the various calculations is provided in Table 3. Table 3. Summary of the regional downscaling realizations and their use in dynamical-statistical results. Regional model run acronym . CMIP5 model driver . Emission scenario . Bio-BCs . EOFs . RCP85 dynamical results . RCP45 dynamical results . Hybrid method results . GFDL_RCP85 GFDL RCP85 clim x x x GFDL_RCP85_BIO GFDL RCP85 interp x x CESM_RCP85 CESM RCP85 clim x x x CESM_RCP85_BIO CESM RCP85 interp x x MIROC_RCP85 MIROC RCP85 clim x x x GFDL_RCP45 GFDL RCP45 clim x x x CESM_RCP45 CESM RCP45 clim x x x MIROC_RCP45 MIROC RCP45 clim x x x Regional model run acronym . CMIP5 model driver . Emission scenario . Bio-BCs . EOFs . RCP85 dynamical results . RCP45 dynamical results . Hybrid method results . GFDL_RCP85 GFDL RCP85 clim x x x GFDL_RCP85_BIO GFDL RCP85 interp x x CESM_RCP85 CESM RCP85 clim x x x CESM_RCP85_BIO CESM RCP85 interp x x MIROC_RCP85 MIROC RCP85 clim x x x GFDL_RCP45 GFDL RCP45 clim x x x CESM_RCP45 CESM RCP45 clim x x x MIROC_RCP45 MIROC RCP45 clim x x x CMIP5 model driver, CMIP5 global model output used as forcing and boundary conditions; Emission scenario, emission scenario of the driver; Bio-BCs, biological boundary conditions used for regional model (clim, present-day climatology; interp, interpolate from biogeochemical output of the global driver); EOFs, models used in derivation of univariate/multivariate EOFs; dynamical results, models used in calculation of dynamical method results; hybrid results, models used in calculation of hybrid (dynamical-statistical) method results. Open in new tab Table 3. Summary of the regional downscaling realizations and their use in dynamical-statistical results. Regional model run acronym . CMIP5 model driver . Emission scenario . Bio-BCs . EOFs . RCP85 dynamical results . RCP45 dynamical results . Hybrid method results . GFDL_RCP85 GFDL RCP85 clim x x x GFDL_RCP85_BIO GFDL RCP85 interp x x CESM_RCP85 CESM RCP85 clim x x x CESM_RCP85_BIO CESM RCP85 interp x x MIROC_RCP85 MIROC RCP85 clim x x x GFDL_RCP45 GFDL RCP45 clim x x x CESM_RCP45 CESM RCP45 clim x x x MIROC_RCP45 MIROC RCP45 clim x x x Regional model run acronym . CMIP5 model driver . Emission scenario . Bio-BCs . EOFs . RCP85 dynamical results . RCP45 dynamical results . Hybrid method results . GFDL_RCP85 GFDL RCP85 clim x x x GFDL_RCP85_BIO GFDL RCP85 interp x x CESM_RCP85 CESM RCP85 clim x x x CESM_RCP85_BIO CESM RCP85 interp x x MIROC_RCP85 MIROC RCP85 clim x x x GFDL_RCP45 GFDL RCP45 clim x x x CESM_RCP45 CESM RCP45 clim x x x MIROC_RCP45 MIROC RCP45 clim x x x CMIP5 model driver, CMIP5 global model output used as forcing and boundary conditions; Emission scenario, emission scenario of the driver; Bio-BCs, biological boundary conditions used for regional model (clim, present-day climatology; interp, interpolate from biogeochemical output of the global driver); EOFs, models used in derivation of univariate/multivariate EOFs; dynamical results, models used in calculation of dynamical method results; hybrid results, models used in calculation of hybrid (dynamical-statistical) method results. Open in new tab Univariate analyses To summarize behaviours of the model through time, we calculate yearly averages of the variables shown in Table 2 within 1.0 degree longitude × 0.5 degree latitude bins spanning the area between 180W–150W and 52N–66N (hence each bin contains approximately 5 × 5 of the native gridpoints of the Bering10K model). This averaging highlights the interannual/multidecadal variability of the output at ∼50 km spatial resolution. These binned values are used to calculate annual/decadal averages and internal vs. model variability, and for derivation of the multivariate modes used in the statistical projections. For the dynamically downscaled results, mean change for each variable under RCP8.5 was summarized by two methods: (i) averaging over all the RCP8.5 ensemble members (GFDL_RCP8.5, CESM_RCP8.5, MIROC_RCP8.5, GFDL_RCP8.5_BIO, and CESM_RCP8.5_BIO) and over all spatial bins to obtain a single yearly time series; (ii) averaging over all RCP8.5 ensemble members and over individual decades to obtain ensemble mean maps of change between decades. Estimates of uncertainty Our calculation of internal vs. model variability (also known as “intrinsic” vs. “structural” uncertainty) is similar to methods used elsewhere (e.g. Hawkins and Sutton, 2009). While it is recognized that oceanic and atmospheric time series have significant energy over a broad range of frequencies (red spectra), it is useful for our present purposes to divide these into sub-decadal and intra-decadal frequencies. Here, internal variability within a particular decade was calculated using the variance of each model’s yearly averages during that decade, followed by averaging that statistic over all members of the ensemble under a particular emission scenario—that is, our working definition of “internal variability” refers to fluctuations of the annual averages over periods shorter than a decade (intra-decadal frequencies). Model variability was calculated as the variance of the decadal average change across ensemble members at each horizontal location—hence it is a measure of differences in decadal means (sub-decadal frequencies) due to model structure, where the “internal” variability has been filtered out. Using a similar distinction, Figure 2 displayed the model variability of spatially averaged, low-pass-filtered air temperature from the CMIP5 models. Here we will calculate internal and model variability for all of the regional and forcing variables at each spatial bin during the period 2090–2099. These definitions and methods for internal vs. model variability are separately applied to the dynamically and statistically derived ensembles of this study, and are subsequently combined into weighted estimates of uncertainty (see “Weighted estimates of change and uncertainty” section). Figure 2. Open in new tabDownload slide Characterization of the chosen ensemble members for the Bering Sea, relative to other CMIP5 models. Upper left panel: low-passed (10-year running mean), spatially averaged air temperature for the eastern Bering Sea from CMIP5 members under RCP8.5, from 1976 through 2080, obtained from the NOAA climate change web portal (https://www.esrl.noaa.gov/psd/ipcc/cmip5/). Ensemble mean of all CMIP5 models (ENSMN) is shown along with individual trajectories of CESM, MIROC, and GFDL models. Light grey, medium grey, and dark grey illustrate the range of: 100% of CMIP5 members, 80% of CMIP5 members nearest to ensemble mean, and 50% of CMIP5 members nearest to ensemble mean, respectively. Upper right panel illustrates temperature change relative to individual model climatologies during 1976–2005. Lower panel: seasonal climatology of sea ice extent among different CMIP5 models as compared to observations during 1980–1999. Solid thick black line shows observed climatology (OBS); grey shading indicates ± 1 SD of all observations. Thin solid and light grey lines represent other CMIP5 model climatologies. Multivariate analyses Multivariate statistical methods can be useful in summarizing the behaviour of oceanographic models and how they respond to large-scale atmospheric forcing. Here we employ a variant of the canonical correlation analysis (CCA) technique to identify the dominant spatial patterns of multiple variables which rise and fall together through time. Similar analyses (Hermann et al., 2013, 2016b) utilized coupled principal component analysis (CPCA) to achieve this aim. Related techniques have been described in Preisendorfer (1988) and Bretherton et al. (1992). As an example of CCA, the time amplitude of a dominant regional spatial pattern of sea surface temperature, identified through univariate EOF analysis, may well be correlated with the time amplitudes of dominant spatial patterns of air temperature and primary production. These spatial patterns need not be similar across the variables; indeed, the primary production response, due to advection and bathymetry, will typically look very different than any wind, air temperature or SST pattern with which it is correlated through time. Details of the statistical procedure are provided in the Appendix; here we summarize the basic steps. To find the most significant modes of multivariate behaviour—those which efficiently reproduce most of the covariance in the original time series of N variables—we begin with the univariate EOFs, which decompose the original fields into pairs of space and time functions. To obtain the most universal patterns, encompassing the widest range of scenario, model, and internal variabilities in a balanced fashion, we include the following six dynamically downscaled ensemble members: GFDL_RCP4.5, GFDL_RCP8.5, CESM_RCP4.5, CESM_RCP8.5, MIROC_RCP4.5, and MIROC_RCP8.5. We demean the binned annual time series for each realization, then concatenate all members together along an abstract time axis. Note that the EOF analysis calculates the eigenvectors of the covariance matrix, and this matrix is unaffected by the values of time per se or their ordering; here we are preparing a collection of independent yearly realizations of each field, which happens to include both multiple years and multiple ensemble members. The top few (say M) EOFs and their modulating “time series” for each of the N variables provide a new, dimensionally reduced, set of functions summarizing the patterns in the original data. A subsequent principal component analysis of the set of N × M multivariate time series will then identify how the spatial modes of these different variables tend to covary in our yearly realizations. This yields a final set of modulating series, with associated spatial patterns for each variable, which efficiently describes much of the correlated variability across both space and variables: Vilt=∑jXjilTjit=∑kCkilΓkt(1) where X and T are the traditional EOF space and time functions, whereas C and Γ are the new modes emphasizing covariance among the different variables. The indices i, j, k, l, t represent variable type i, spatial location l, univariate mode number j, multivariate mode number k, and time index t. We map the spatial structures of the most significant univariate and multivariate modes, and note the spatially averaged variance explained by each. The value of this decomposition lies not only in its illumination of multivariate modes of variability, but also in its potential use in predicting any subset of “response” variables (here, the regional quantities not yet generated through dynamical downscaling) from a collection of “forcing” variables (readily available from coarse-scale CMIP5 output). This is achieved by projecting those forcing variables onto the previously derived multivariate modes—that is, by estimating how much of each multivariate mode C (which include the response variables) is contained in those readily available fields, for each year where downscaling has not been performed, to obtain the corresponding Γvalue for that year. The result is an estimate of likely regional biophysical response, given only the large-scale or coarse-scale atmospheric forcing from some new global model (e.g. other CMIP5 or CMIP6 models), not yet dynamically downscaled to the Bering Sea: Vestilt=∑kCkilΓestkt(2) Effectively this bootstraps the results from a few, computationally intensive, dynamically downscaling runs to an estimate of what downscaling results would have been obtained from a much larger ensemble. The success of this approach depends on finding multivariate modes which explain non-trivial variance in both the forcing variables (for which we have a large ensemble of realizations) and the regional response (for which we have a more limited ensemble). In the present case, we choose air temperature, air pressure, zonal winds, and meridional winds as our “forcing” variables, since they are readily available online for most of the CMIP5 models, subjected to RCP4.5 (35 different models) and RCP8.5 (28 different models) emission scenarios (http://apdrc.soest.hawaii.edu/datadoc/cmip5.php). We compare the estimates from statistical projection of our six-member ensemble with the corresponding dynamically downscaled output, to quantify how much of the signal in each variable is captured by the technique. Specifically, we calculate the mean square difference between the original and estimated fields, normalized by the variance of the original field (Rfit, henceforth termed “fractional variance”): Rfitil=1.0-∑tVestilt-Vilt2/∑tVilt2(3) Hence a value of 1 indicates a perfect fit between original and estimated fields (and values less than 0 are possible for an exceptionally bad fit, though in practice all values fell between 0 and 1). Finally, we use the technique to generate a 63-member ensemble of regional projections under RCP4.5 and RCP8.5, by estimating the Γkt for each of those members in each projected year, and summing the modes to obtain Vest [Equation (2)] Details regarding the derivation of Γest (essentially, projection of four atmospheric forcing variables onto the multivariate modes) are shown in the Appendix. Weighted estimates of change and uncertainty The goal of our hybrid dynamical-statistical downscaling method is an improved estimate of the expected change, and the uncertainty of that estimate, due to the internal variability of the real system across a broad range of time scales and the structural uncertainty (many possible formulations of equations and parameters) of both the global and regional models. Dynamical downscaling methods are more likely to capture large, nonlinear changes in the regional system under global climate change, compared to simple statistical downscaling methods based on correlations between present-day global and regional data. However, there is no guarantee that the (affordably) small ensemble used for dynamical downscaling will faithfully capture the mean and variability which would have been obtained had the larger ensemble of CMIP5 global results all been used. The application of multivariate statistical rules, derived from the smaller ensemble, to the larger ensemble of global results, serves to correct for these discrepancies. This provides improved estimates of mean expected change and its uncertainty, when a full dynamical downscaling of all CMIP5 members is not feasible. Note, however, that the statistical method is still just an approximation, and will fail to capture some of the changes observed under full dynamical downscaling. Its efficacy as a replacement will be both space and variable dependent. Hence, a final “best” estimate of change and uncertainty combines both the dynamical and statistical results. A reasonable approach here is to utilize the fractional variance (goodness of fit) metric to derive a weighted average of the two: Vwgtilt=Rfitil<Vestilt>+ 1-Rfitil<Vilt>(4) where <.> denotes the ensemble average over all available dynamical (V) or statistical (Vest) members. That is, for each variable, in areas where the statistical method is least successful, we preferentially utilize our small ensemble, dynamically downscaled results, while in areas where the statistical method is most successful, we preferentially utilize our large ensemble, statistically derived results. An analogous weighted mean is derived from the internal and model variability estimates, derived separately from the dynamical and statistical members. Results We begin by comparing the seasonal means for 10-year periods of the “present” (2010–2019), the “near-future” (2050–2059), and the “far-future” (2090–2099) under RCP8.5, as derived from direct dynamical downscaling. We then present results from the multivariate analysis, and statistically expand our original ensemble to include a much larger set of CMIP5 forcings. Finally we combine the dynamical and statistical estimates, present the anticipated changes by 2100 under RCP4.5 and RCP8.5, and present the magnitude of internal and model variability for these combined estimates. Mean spatial patterns under present forcing Decadal ensemble mean spatial patterns of “present” (2010–2019) conditions from the five-member RCP8.5 ensemble are displayed in Figure 3 using the raw (untransformed) variable values (for all subsequent plots and analyses, some of the variables were log transformed as noted in Table 2). Prominent spatial features include the following: (i) warmer sea surface temperatures in the Gulf of Alaska and cooler temperatures to the northwest; (ii) lowest sea bottom temperatures at mid shelf (the “cold pool”); (iii) lowest salinity near the coast; (iv) elevated phytoplankton on the middle shelf and near the coast; (v) depleted nitrate plus ammonium on the shelf; (vi) elevated microzooplankton, copepods, and euphausiids on the southeastern and middle shelf; (vii) elevated neocalanus on the southeastern shelf; (viii) elevated mixing on the northwestern shelf; (ix) elevated benthic detritus and epi-benthos along the 50 m isobath, with maxima to the north and south; (x) elevated ice cover and ice phytoplankton on the northwestern shelf; (xi) deepest mixed layer depth at the outer shelf; (xii) highest sea surface height in the Gulf of Alaska and on the southeastern shelf; (xiii) highest air temperature and absolute humidity in the southeast; (xiv) northeastward winds in the southeast and southeastward winds in the northwest; (xv) strongest downward longwave and shortwave radiation in the southeast. Figure 3. Open in new tabDownload slide Decadal average spatial patterns for the years 2010–2019, downscaled from the five-member ensemble of RCP8.5 runs. Descriptions of each variable and its units are listed in Table 2. In this subsequent areal plots, isobaths are shown at depths of 50, 100, 200, and 1000 m. In these plots, the raw (untransformed) values are displayed; for all subsequent plots and analyses, some of the variables are log transformed as shown in Table 2. Colour version of this figure is available online at ICESJM online. Many of these model-generated features correspond to observed properties of the present-day Bering Sea. The mid-shelf “cold pool” of low bottom temperatures is a prominent feature of the region, with strong impacts on fish production. Correspondence of the modelled cold pool and ice cover with observations under past and present-day forcing, as well as extensive comparisons of temperature and salinity throughout the water column, have been described in Hermann et al. (2013, 2016a) and Ortiz et al. (2016). Similar patterns were observed here under “present” (2010–2019) CMIP5 forcing. The spatial pattern of yearly averaged phytoplankton—higher in the southeastern Bering (e.g. Bristol Bay and along the Aleutian chain), the northwestern Bering (e.g. the Bering Strait), and along the middle shelf domain—resembles the satellite-derived annual chlorophyll climatology reported in Brown et al. (2011). The southeasterly, mid-shelf maxima of euphausiids resemble the acoustically derived euphausiid biomass patterns reported in Ressler et al. (2012, 2014). A similar (southeasterly and mid-shelf) distribution of copepod biomass generally corresponds to the (spatially limited) copepod abundance estimates (Coyle et al., 2008; Hunt et al., 2011; D. Kimmel, pers. comm.). The model is less successful at replicating observed patterns of neocalanus, where the available data suggest maximal biomass near the shelf break (D. Kimmel, pers. comm.). Some of these pelagic biomass comparisons are complicated by the fact that we are using vertical averages from the model, as compared to near-surface vertical integrals or surface values reported in the observational studies. Spatial patterns of modelled benthic detritus and infaunal biomass in the northern Bering (e.g. local maxima near Bering Strait) roughly correspond to the infaunal patterns reported for the northern shelf in Grebmeier (2012) and Grebmeier et al. (2015). Limited information is available for total benthic infaunal biomass in the southeastern Bering; however, bottom trawl surveys typically indicate maxima of red king crabs along the Aleutian peninsula and the inner shelf domain near Bristol Bay, and maxima of snow crabs in the northern portion of the middle shelf domain (e.g. Siddon and Zador, 2017). Our model indicates relative maxima of total benthic infaunal biomass in these regions. Spatially averaged trends under RCP8.5 As a simplest index of projected change, we now examine the time series of the spatial average of each quantity, averaged over the five-member RCP8.5 ensemble (Figure 4; see Table 3 for models used). Quantities exhibiting consistent upward trends include sea surface temperature, sea bottom temperature, sea surface height, cross-shelf transport, air temperature, absolute humidity, and downward longwave radiation. Quantities exhibiting consistent downward trends include sea surface salinity, phytoplankton, nutrients, copepods, euphausiids, benthic detritus and epibenthos, ice cover, mixed layer depth (more negative values indicate mixed layer deepening), and ice phytoplankton. For most of these variables, the largest changes appear after 2040. Note how even after averaging across space and ensemble members, there is still substantial interannual as well as interdecadal variability, reflecting the intrinsic variability within each simulation. Figure 4. Open in new tabDownload slide Time series of spatial averaged properties, averaged over the five-member RCP8.5 ensemble. Projected change in spatial patterns under RCP8.5 To examine the spatial patterns of these changes in more detail, and to correct for systematic bias in each of the models, we plot the average decadal change between the 2010s and the 2050s for the five-member RCP8.5 ensemble in Figure 5 (same models used as in Figure 4). These patterns differ substantially from those of the mean state shown in Figure 3. Sea surface temperature (sst) increases uniformly across the model domain. Changes to sea bottom temperature (sbt) are focused on the inner and middle shelf, with maximum increase to the northwest. Sea surface salinity (sss) exhibits greatest decrease near the coast. Depth-averaged phytoplankton, copepods, and euphausiids (phyt, cope, eup) exhibit greatest losses along the outer shelf, while neocalanus (nca) exhibits a slight increase on the northwestern shelf, and microzooplankton (mzoo) exhibits a substantial increase all across the shelf. Ice cover (iceco) and an ice phytoplankton (iceph) decrease in the northwest. Sea surface height (ssh) increases along the coast, consistent with enhanced shoreward flow across the shelf (utop). Air temperature (tair) and absolute humidity (qair) increase everywhere, especially in the northwest, while air pressure (pair) decreases, especially in the northwest. Changes to winds (uwind, vwind) are geostrophically consistent with the changes in air pressure (pair), i.e. enhanced northeastward winds along the enhanced atmospheric pressure gradient. Increased downward longwave radiation (lwrad) is especially prominent in the northwest. Figure 5. Open in new tabDownload slide Decadal average change between 2010–2019 and 2050–2059 from downscaled projections based on GFDL-ESM2M, CESM, and MIROC RCP 8.5 global projections. Colour version of this figure is available online at ICESJM online. These broad spatial trends under RCP8.5 continue through the remainder of the twenty-first century (Figure 6). In addition, a substantial increase in vertical mixing (akt) and deepening of the mixed layer (mld) are produced along the shelf break. Sea surface and sea bottom temperatures increase by as much as 5°C, while copepods, euphausiids, and benthos (all log10 transformed in Figures 5 and 6) decrease by as much as half their initial values (log10 decreased by 0.3 or greater). Changes to shortwave radiation lack a clear trend or spatial pattern by mid-century, but are reduced by ∼10% of their present-day mean by the end of the century. Figure 6. Open in new tabDownload slide Decadal average change between 2010–2019 and 2090–2099 from downscaled projections based on GFDL-ESM2M, CESM, and MIROC RCP 8.5 global projections. Colour version of this figure is available online at ICESJM online. To summarize: the CMIP5 projections of the small ensemble (based on GFDL, CESM, and MIROC models) anticipate a shift to warmer air temperatures (especially in the northern Bering Sea) and a shift to more northward winds under the RCP8.5 scenario. When dynamically downscaled through the Bering10K model, these and related changes in the forcing lead to substantially warmer surface and bottom temperatures (hence a smaller “cold pool”), reduced ice cover, enhanced cross-shelf surface flux, enhanced growth of small zooplankton and neocalanus in the (increasingly ice-poor) northern Bering, and reduced biomass of phytoplankton, copepods, and euphausiids on the outer shelf of the (increasingly warm) southern Bering. Multivariate analysis Univariate EOFs These analyses use the six-member ensemble (see Table 3), which includes both RCP4.5 and RCP8.5 members. We begin by presenting the univariate EOFs of each analysed variable, based on the annually averaged fields (Figure 7). For many but not for all cases, these spatial patterns are similar to those of the long-term decadal trends under RCP8.5 in Figures 5 and 6. Note that this EOF analysis includes interannual through interdecadal scales of variability, as well as the considerable structural and scenario variability across the CMIP5 models driving the Bering10K realizations. This suggests a consistent, spatially dependent response in those realizations to recurring patterns in the forcing. In some cases (e.g. sst, tair, qair), the first univariate mode captures nearly all of the variance in the original time series, whereas in others (e.g. surface velocities utop and vtop) less than 25% is explained by the first univariate mode. Figure 7. Open in new tabDownload slide First-mode univariate EOFs for forcing and regional response variables, calculated using a six-member ensemble which includes both RCP4.5 and RCP8.5. Fractional variance explained by the univariate EOF is listed for each variable. Colour version of this figure is available online at ICESJM online. Multivariate modes When the multivariate modes are calculated (Figure 8), we find many spatial patterns similar to those of the first univariate EOFs (compare Figures 7 and 8, bearing in mind the sign of each pattern in Figure 7 is arbitrary). The spatial patterns of the first multivariate mode clearly indicate many of the patterns rise and fall in synchrony across variables; for example, note how different spatial patterns of sea bottom temperature and air temperature co-occur, and how these co-occurring patterns explain over 70% of each variable’s original signal. Variables most strongly explained by the multivariate mode are: sea surface and bottom temperature, ice cover, air temperature, and absolute humidity. Variables most weakly explained by this mode are: air pressure, surface velocities, winds, and shortwave radiation. A significant fraction of the biological signals are explained. In many locations, this “warm mode” is associated with reduced biomass of phytoplankton and large crustacean zooplankton, especially along the outer shelf. In some locations, the higher temperatures are associated with increased biomass, for example for microzooplankton and neocalanus. Figure 8. Open in new tabDownload slide First-mode multivariate EOFs for the collection of forcing and regional response variables, with fractional variance of the original time series explained by that mode. Colour version of this figure is available online at ICESJM online. The second multivariate mode (Figure 9) explains substantially less of the original signals; however, averaged over the entire domain, the following variables have at least 10% of their original variance explained: air pressure, zonal and meridional winds, along-shelf surface velocity, downward longwave radiation, shortwave radiation, vertical mixing (greater on the outer shelf), and benthic infauna. None of the biological variables have more than 10% of their spatially averaged variance explained by this mode, although in some cases there are spatially localized effects where much of the original signal is captured (e.g. note the near-shore reduction in many of the biological groups, and the near-shore increase in nutrients). Figure 9. Open in new tabDownload slide Second-mode multivariate EOFs for the collection of forcing and regional response variables, with fractional variance of the original time series explained by that mode. Colour version of this figure is available online at ICESJM online. The differences between the two modes are further illustrated by plotting the variance explained for each variable (shown on Figures 8 and 9) on a 2D scatter plot (Figure 10). The two multivariate modes appear as fundamentally different modes of variability (a “heat” mode and a “wind” mode), with nearly all variables falling into one or the other “factor” group. The notable exceptions are benthic detritus and downward longwave radiation; modes 1 and 2 both explain a significant percentage of the original signal for those variables. Note also that both modes include increased vertical mixing and deepening of the mixed layer on the outer shelf (Figures 8 and 9). Figure 10. Open in new tabDownload slide Variable loadings on the first 2 multivariate modes; a scatter plot of total variance explained for each variable by multivariate mode 1 (x-axis) vs. multivariate mode 2 (y-axis). Variances plotted here are also listed in Figures 8 and 9, respectively We now calculate how much of the variance in our “training” set V can be replicated by Vest, with multivariate mode amplitudes calculated exclusively by projecting only four of atmospheric forcing variables used in our dynamical downscaling: air temperature, air pressure, zonal winds, and meridional winds (Figure 11). In general, greater variance is explained within those areas where large change was calculated between 2010s and 2090s (Figure 6); these areas naturally dominate the univariate and multivariate modes as well (Figures 7–9). In many areas over 50% of the variance is replicated by the method, that is, we capture over 50% of the dynamically downscaled results using our statistical estimator alone (Figure 11). As described in “Weighted estimates of change and uncertainty” section, this fractional variance at each location is used in our final weighted estimates of change and uncertainty for each variable. Figure 11. Open in new tabDownload slide Fractional variance of the dynamically downscaled output reproduced by projecting four atmospheric forcing variables (tair, pair, uwind, and vwind) onto the top three multivariate modes. Colour version of this figure is available online at ICESJM online. Statistical expansion of the ensemble using the multivariate modes We now estimate the regional response to an expanded set of CMIP5 models under emission scenarios RCP4.5 and RCP8.5. As described in the Methods, this proceeds by projecting each member’s available atmospheric “forcing” variables (here, air temperature, air pressure, zonal winds, and meridional winds) onto the multivariate modes calculated from the dynamical downscaling results, and subsequently taking a weighted average of the statistically and dynamically downscaled results [Equation (4)] The statistically calculated change in decadal sea bottom temperature (Figure 12) shows considerable variability among the different members, reflecting substantial structural variability among CMIP5 models. An ensemble mean is calculated from those members; as with the dynamically downscaled results, it exhibits an increase of up to 5°C on the northern Bering Sea shelf. The same procedure applied to total large crustacean zooplankton (lcz), using the logarithm of those values, predicts a decrease of −0.2 log units on the southeastern shelf by 2100—that is, a reduction nearly by half from the values earlier in the century (Figure 13). Conversely, there is mean increase in lcz over the northern shelf. Figure 12. Open in new tabDownload slide Ensemble results for sea bottom temperature (sbt), obtained by projecting atmospheric forcing variables (from 28 different CMIP5 models under emission scenario RCP8.5) onto the multivariate modes. Upper panel: calculated change in 10-year average sbt between 2010–2019 and 2090–2099 for each CMIP5 realization. The forcing CMIP5 model is listed at the bottom of each panel. Isobaths are shown at depths of 50, 100, 200, and 1000 m. Lower left panel: yearly areal average for each CMIP5 realization, relative to the 2010–2019 mean. Dark black line shows ensemble mean; light black lines indicate ± 1 SD for that year. Lower right panel: ensemble mean change based on the 28 CMIP5 models. Colour version of this figure is available online at ICESJM online. Figure 13. Open in new tabDownload slide Ensemble results as in Figure 12, for log10 (large crustacean zooplankton). Colour version of this figure is available online at ICESJM online. The weighted means of the (5-member) dynamical ensemble and (28-member) statistical ensemble result for decadal change under RCP8.5 for all variables (Figure 14) bear a very strong resemblance to the means calculated from the dynamical results alone under that emission scenario (compare Figures 6 and 14). Relative to the small ensemble results, the large ensemble appears to have greater northward winds (vwind), leading to greater onshore flow (utop) and slightly reduced ice cover (iceco). Other differences include slightly reduced changes for air temperature (tair), sea surface temperature (sst), and sea bottom temperatures (sbt). Some of these differences between Figures 6 and 14 are hard to discern, as the spatial patterns are so similar, but emerge in difference maps (not shown). The strong resemblance of the spatial patterns by the two methods results from two factors: (i) the mean change in air temperature of the five-member dynamical ensemble is very similar to the mean change of the full CMIP5 ensemble (e.g. see Figure 2), and that particular forcing variable strongly correlates with the regional responses (the “heat” mode revealed by the analysis); (ii) not all areas were well reproduced by the statistical modes (low fractional variance explained, see Figure 11), and hence the original, dynamical estimate was preferentially used in those areas [Equation (4)]. Figure 14. Open in new tabDownload slide Predicted decadal average change between 2010–2019 and 2090–2099 under RCP8.5, using statistical extrapolation to include 28 ensemble members (weighted average of dynamical and statistical results). Colour version of this figure is available online at ICESJM online. The analogous weighted maps based on the (3-member) dynamical and (35-member) statistical ensembles under RCP4.5 (Figure 15) exhibit marked differences from the results under RCP8.5. Of particular note, they predict only half the increase in surface and bottom temperatures anticipated under RCP8.5. The general sense of the spatial patterns is retained, but at much smaller amplitude as compared to RCP8.5. Figure 15. Open in new tabDownload slide Predicted decadal averages changes between 2010–2019 and 2090–2099 under RCP4.5, using statistical extrapolation to include 35 ensemble members (weighted average of dynamical and statistical results). Colour version of this figure is available online at ICESJM online. Internal vs. model variability of the projections Variabilities (“uncertainties”) of projections computed from the dynamical method alone (not shown here) were in fact very different from those computed by the statistical method, a reflection of the small sample size of the dynamical set. The much larger statistical set presumably gives us a superior estimate of the true uncertainty among projections, at least in those areas where the statistical method is able to capture a substantial fraction of the dynamical variance (see Figure 11). The weighted dynamical-statistical means for internal and model variability among the RCP8.5 projections are shown in Figures 16 and 17, respectively. The model variability is larger than the internal variability for nearly all variables and locations; this is especially true for sea surface height. The exceptions are the atmospheric pressure and associated winds, which exhibit far greater internal than model variability. For many of the variables, model variability is greatest in areas where the largest mean interdecadal change was calculated (Figure 6). In particular, sea bottom temperature exhibits greatest such uncertainty (greater than 2°C) in the northern Bering Sea. Figure 16. Open in new tabDownload slide Internal (interannual) variability of each variable during 2090–2099, using statistical extrapolation to include 28 ensemble members from RCP 8.5 global projections (weighted average of dynamical and statistical results). Levels and colour key for each variable correspond to those used in plots of decadal change. Colour version of this figure is available online at ICESJM online. Figure 17. Open in new tabDownload slide Cross-model variability of each variable during 2090–2099, using statistical extrapolation to include 28 ensemble members from RCP 8.5 global projections (weighted average of dynamical and statistical results). Levels and colour key for each variable correspond to those used in plots of decadal change. Colour version of this figure is available online at ICESJM online. For a majority of the variables, both the model and internal variability are smaller than the calculated mean change in most areas by 2100. Winds and air pressure exhibit some of the greatest internal variability relative to mean change, whereas sea surface height, longwave radiation, and shortwave radiation exhibit some of the greatest model variability relative to mean change. Discussion Causal pathways and limits of the dynamical model The basic patterns of change from the dynamically downscaled results suggest large changes to the future Bering Sea, with considerable spatial detail—not all areas warm to the same extent, and large zooplankton are reduced most severely on the outer shelf. These results are consistent with our previous downscaling runs of the model through 2040 (Hermann et al., 2016a), as well as recent observations of warm vs. cold years in the Bering Sea (Hunt et al., 2011). The warmer surface and bottom temperatures are largely a response to warmer air temperatures and reduced ice (hence greater penetration of shortwave radiation). Reduced southward advection of ice (as winds become more northward) may contribute secondarily to warming in the south. Reduced formation of ice also manifests as reduced salinity in Norton Sound, as brine rejection is reduced there. One probable causal pathway reducing the large zooplankton is heightened grazing pressure on those groups (represented as a temperature-dependent quadratic loss term in the model). The northern increase in neocalanus is difficult to interpret, especially as the present-day pattern produced by the model may not reflect the true observed distribution. The broad increase in microzooplankton is likely driven in part by an increase in temperature-dependent growth rates. The decreased phytoplankton on the outer shelf may be partly due to the deepened mixed layer, for example through light limitation. Further analysis of the model output may help to clarify these mechanisms. Known biases and imperfections of the regional dynamical model used here include a limited number (10) of vertical layers, a fixed Bering Sea through flow, and a simple monthly climatology (no interannual variability) for coastal runoff. Several of these issues have been addressed in more recent simulations, which include more (30) layers and improved algorithms for light extinction, primary production, and zooplankton diapause (K. Kearney, pers. comm.). Previous work has indicated that the use of only 10 layers does not strongly limit our ability to reproduce the fundamental seasonal and interannual temperature patterns, although it may contribute to a persistent shallow bias in mixed layer depth (Hermann et al., 2016a). An observed bias toward late melting of ice in the hindcasts (Ortiz et al., 2016), while undesirable, should not strongly affect the annual averages employed for our analyses. Recent improvements to the ice dynamics and thermodynamics code (K. Hedstrom, pers. comm.) have significantly reduced this bias. More generally, it is recognized that for the biological elements in particular: (i) there is considerable uncertainty about the present structure of the food web and appropriate rates for each variable, and (ii) we are implicitly assuming that the present major groups (state variables) and their rates will endure over the twenty-first century. As in global models, we have no guarantee that this will be the case; new species types, with different rates and strategies, may ultimately come to dominate the Bering Sea as it warms. Caution is of course warranted in the use of such projections. Limits of the hybrid dynamical-statistical methodology The differences in temporal change calculated from the smaller (dynamically downscaled) and larger (statistically downscaled) ensembles reflect the added model (structural) uncertainty of CMIP5 encompassed by the latter. The five-member RCP8.5 ensemble (actually only three different models, with two additional runs using biological boundary conditions from GFDL and CESM models), while deliberately chosen to span a range of “warmer” and “colder” models from CMIP5, is still a small subsample of that set. Indeed, our initial choice of three models (GFDL, CESM, and MIROC) is a small sample from a population with large variance (e.g. see Figure 2), hence subject to significant error as an estimate of the true mean and variance (uncertainty) of that population. This fact—and the high cost of dynamical downscaling—is the primary motivator for seeking an efficient hybrid (dynamical-statistical) method to downscale a larger population of CMIP5 output. The fact that the statistically generated ensemble results are so similar in pattern to the dynamically generated results is to some extent a consequence of so much of the variance and covariance being contained in the first, “heat,” multivariate mode. The forcing data from the larger set of CMIP5 output projects strongly onto this mode, which generates a final pattern similar to that of the dynamically generated results. However, we now have the added benefit of the much larger ensemble, to inform us regarding the expected mean amplitudes of those patterns in any future year, and their uncertainty. We achieve this benefit without dynamically downscaling all of the CMIP5 members. Obviously we will not capture all of the signal which would emerge from a full dynamical downscaling through this hybrid method. To compensate, we use an objective measure of our statistical method (fractional variance explained, Figure 11) in our final weighted sums [Equation (4)], to extract the greatest value from both methods. Simpler statistical methods (e.g. linear regression between observed air temperature and regional ocean temperature, applied to predict the future ocean temperature using projected air temperature from CMIP5) have been used in other studies to quickly predict the regional response to a broad range of possible futures. The fact that our multivariate technique emphasizes covariance across space and across variables (e.g. air temperature is correlated with absolute humidity, and the sea level pressure field is correlated with the winds) reduces the possibility of spurious correlations being obtained for any single location and/or single pair of predictor/predictand variables. Instead, we have assimilated the covariance structure of the entire set of multivariate fields into our unified statistical rule. Further testing and alternate methods In the present work, we used the same set of data both for: (i) generation of the multivariate modes and (ii) testing how well a subset of the variables could replicate, through projection onto those multivariate modes, the original set. More rigorous tests of the method will include sequentially withholding individual ensemble members from the original training set, and calculating how well the methods can replicate the withheld member. We will also explore the relative performance of simpler multivariate regression methods at each spatial location. Finally, this and alternate methods will be tested for application to seasonal patterns. Seasonal evolution entails strong autocorrelation of signals, and hence alternate methods such as linear inverse modelling (LIM) (Newman et al., 2003; Alexander et al., 2008) may be more appropriate for that case. Such LIM methods entail the calculation of a space-time correlation matrix, sometimes using a dimensionally reduced dataset (e.g. Capatondi and Sardeshmukh, 2015), as was used in our analysis. Conclusions As part of an interdisciplinary project to explore management strategy in a future Bering Sea (ACLIM), an eight-member ensemble of global CMIP5 output was dynamically downscaled using a 10-km resolution regional biophysical model. These downscaled projections indicate that shelf bottom temperatures may increase by as much as 5°C by the end of the twenty-first century, given a continuation of present global trends in the emission of greenhouse gases (i.e. IPCC scenario RCP8.5). These changes may be accompanied by a significant reduction in large crustacean zooplankton over the outer shelf of the southeastern Bering Sea. Such impacts are substantially reduced (approximately by half) by 2100 under a moderate emission mitigation scenario (RCP4.5). These results suggests that future anthropogenic atmospheric carbon emissions will have a strong impact on Bering Sea physics and lower trophic level biology by the end of the twenty-first century. These are in turn expected to have strong consequences on the fisheries of the region; such impacts are being explored in fishery and socioeconomic models under ACLIM. A hybrid multivariate method, based on a six-member dynamically downscaled ensemble, was used to statistically expand our ensemble to over 60 members (35 for RCP4.5 and 28 for RCP8.5). The multivariate analysis suggests two primary independent modes of variability, the primary one involving temperature and zooplankton productivity (among other covariates), and a secondary one involving air pressure and winds with fewer associated biological effects. The expanded results from the hybrid dynamical-statistical method are similar in spatial pattern to those from dynamical downscaling alone, but provide an estimate of the magnitude of change we would have obtained if all 63 members were to be dynamically downscaled, and the associated uncertainty of those estimates due to both internal and cross-model (CMIP5) variability. As such it yields improved estimates of future conditions in the Bering Sea, which are useful in management strategy evaluation for fisheries of the region under high vs. low emission scenarios. Results thus far suggest that the projected mean changes of many biophysical attributes, from the present through 2100, exceed the internal variability or model-generated uncertainty of those estimates under RCP8.5. Acknowledgements This study is part of NOAAs Alaska Climate Integrated Modeling project (ACLIM) and FATE project 14-05. We thank the entire ACLIM team for feedback and discussions regarding projection modelling and application. ACLIM was supported by Fisheries and the Environment (FATE), Stock Assessment Analytical Methods (SAAM), North Pacific Climate Regimes & Ecosystem Productivity (NPCREP), Economics and Human Dimensions Program, NOAA Integrated Ecosystem Assessment Program (IEA), NOAA Research Transition Acceleration Program (RTAP), the Alaska Fisheries Science Center (ASFC), and the NOAA Office of Oceanic and Atmospheric Research (OAR) and National Marine Fisheries Service (NMFS). We are especially grateful to the Asia-Pacific Data Resource Centre (APDRC) for hosting CMIP5 output in a readily accessed manner. This is PMEL contribution 4846 and Eco-FOCI contribution 0921. This publication was partially funded by the Joint Institute for the Study of the Atmosphere and Ocean (JISAO) under NOAA Cooperative Agreement NA15OAR4320063. Author contributions A.J.H. and W.C. performed the dynamical downscaling, which includes NPZ model elements developed by G.A.G. A.J.H. developed and applied the hybrid statistical method and drafted the manuscript. All of the co-authors were involved in selecting the appropriate CMIP5 models for dynamical downscaling, and the interpretation of results. A.B.H. and K.K.H. are co-leads of the ACLIM project, and as such provided guidance on the necessary scope of the climate projections and their application to the suite of biological models. References Alexander M. A. , Matrosova L., Penland C., Scott J. D., Chang P. 2008 . Forecasting pacific SSTs: linear inverse model predictions of the PDO . Journal of Climate , 21 : 385 – 402 . Google Scholar Crossref Search ADS WorldCat Budgell W. P. 2005 . Numerical simulation of ice-ocean variability in the Barents Sea region: towards dynamical downscaling . Ocean Dynamics , 55 : 370 – 387 . Google Scholar Crossref Search ADS WorldCat Bretherton C. S. , Smith C., Wallace J. M. 1992 . An intercomparison of methods for finding coupled patterns in climate data . Journal of Climate , 5 : 541 – 560 . Google Scholar Crossref Search ADS WorldCat Brown Z. W. v. , Dijken G. L., Arrigo K. R. 2011 . A reassessment of primary production and environmental change in the Bering Sea . Journal of Geophysical Research , 116 : C08014 . Google Scholar OpenURL Placeholder Text WorldCat Capotondi A. , Sardeshmukh P. D. 2015 . Optimal precursors of different types of ENSO events . Geophysical Research Letters , 42 : 9952 – 9960 . Google Scholar Crossref Search ADS WorldCat Cheng W. , Curchitser E., Ladd C., Stabeno P. J., Wang M. 2014 . Influences of sea ice on the eastern Bering Sea: NCAR CESM simulations and comparison with observations . Deep Sea Research Part II: Topical Studies in Oceanography , 109 : 27 – 38 . Google Scholar Crossref Search ADS WorldCat Coachman L. K. 1986 . Circulation, water masses, and fluxes on the southeastern Bering Sea shelf . Continental Shelf Research , 5 : 23 – 108 . Google Scholar Crossref Search ADS WorldCat Coyle K. O. , Eisner L. B., Mueter F. J., Pinchuk A. I., Janout M. A., Cieciel K. D., Farley E. V., et al. . 2011 . Climate change in the southeastern Bering Sea: impacts on pollock stocks and implications for the Oscillating Control Hypothesis . Fisheries Oceanography , 20 : 139 – 156 . Google Scholar Crossref Search ADS WorldCat Coyle K. O. , Pinchuk A. I., Eisner L. B., Napp J. M. 2008 . Zooplankton species composition, abundance and biomass on the eastern Bering Sea shelf during summer: the potential role of water column stability and nutrients in structuring the zooplankton community . Deep Sea Research Part II: Topical Studies in Oceanography , 55 : 1775 – 1791 . Google Scholar Crossref Search ADS WorldCat Dai A. , Qian T., Trenberth K. E., Milliman J. D. 2009 . Changes in continental freshwater discharge from 1948–2004 . Journal of Climate , 22 : 2773 – 2791 . Google Scholar Crossref Search ADS WorldCat Danielson S. , Curchitser E., Hedstrom K., Weingartner T., Stabeno P. 2011 . On ocean and sea ice modes of variability in the Bering Sea . Journal of Geophysical Research , 116 : C12034 . Google Scholar Crossref Search ADS WorldCat Dunne J. P. , John J. G., Adcroft A. J., Griffies S. M., Hallberg R. W., Shevliakova E., Stouffer R. J., et al. . 2012 . GFDL’s ESM2 global coupled climate-carbon earth system models. Part I: physical formulation and baseline simulation characteristics . Journal of Climate , 25 : 6646 – 6665 . Google Scholar Crossref Search ADS WorldCat Fissel B. , Dalton M., Garber-Yonts B., Haynie A., Kasperski S., Lee J., Lew D., et al. . 2017 . Stock assessment and fishery evaluation report for the groundfish fisheries of the Gulf of Alaska and Bering Sea/Aleutian Islands area: Economic status of the groundfish fisheries of Alaska, 2016. NPFMC Bering Sea, Aleutian Islands and Gulf of Alaska SAFE, Seattle, WA. Retrieved from http://www.afsc.noaa.gov/refm/docs/2017/economic.pdf (last accessed 9 January 2019). Gibson G. A. , Spitz Y. H. 2011 . Impacts of biological parameterisation, initial conditions, and environmental forcing on parameter sensitivity and uncertainty in a marine ecosystem model for the Bering Sea . Journal of Marine Systems , 88 : 214 – 231 . Google Scholar Crossref Search ADS WorldCat Grebmeier J. M. 2012 . Shifting patterns of life in the Pacific Arctic and sub-Arctic seas . Annual Review of Marine Science , 4 : 63 – 78 . Google Scholar Crossref Search ADS PubMed WorldCat Grebmeier J. M. , Bluhm B. A., Cooper L. W., Danielson S. L., Arrigo K. R., Blanchard A. L., Clarke J. T., et al. . 2015 . Ecosystem characteristics and processes facilitating persistent macrobenthic biomass hotspots and associated benthivory in the Pacific Arctic . Progress in Oceanography , 136 : 92 – 114 . Google Scholar Crossref Search ADS WorldCat Haidvogel D. B. , Arango H., Budgell W. P., Cornuelle B. D., Curchitser E., Di Lorenzo E., Fennel K., et al. . 2008 . Regional ocean forecasting in terrain-following coordinates: model formulation and skill assessment . Journal of Computational Physics , 227 : 3595 – 3624 . Google Scholar Crossref Search ADS WorldCat Hawkins E. , Sutton R. 2009 . The potential to narrow uncertainty in regional climate predictions . Bulletin of the American Meteorological Society , 90 : 1095 – 1108 . Google Scholar Crossref Search ADS WorldCat Haynie A. C. , Huntington H. P. 2016 . Strong connections, loose coupling: the influence of the Bering Sea ecosystem on commercial fisheries and subsistence harvests in Alaska . Ecology and Society , 21 : 6 . Google Scholar Crossref Search ADS WorldCat Hermann A. J. , Gibson G. A., Bond N. A., Curchitser E. N., Hedstrom K., Cheng W., Wang M., et al. . 2013 . A multivariate analysis of observed and modeled biophysical variability on the Bering Sea shelf: multidecadal hindcasts (1970–2009) and forecasts (2010–2040) . Deep Sea Research Part II: Topical Studies in Oceanography , 94 : 121 – 139 . Google Scholar Crossref Search ADS WorldCat Hermann A. J. , Gibson G. A., Bond N. A., Curchitser E. N., Hedstrom K., Cheng W., Wang M., et al. . 2016a . Projected future biophysical states of the Bering Sea . Deep-Sea Research Part II: Topical Studies in Oceanography , 134 : 30 – 47 . Google Scholar Crossref Search ADS WorldCat Hermann A. J. , Ladd C., Cheng W., Curchitser E. N., Hedstrom K. 2016b . A model-based examination of multivariate physical modes in the eastern and western Gulf of Alaska . Deep Sea Research Part II: Topical Studies in Oceanography , 132 : 68 – 89 . Google Scholar Crossref Search ADS WorldCat Holsman K. , Hollowed A., Ito S.-I., Bograd S., Hazen E., King J., Mueter F., et al. . 2018 . Climate change impacts, vulnerabilities and adaptations: North Pacific and Pacific Arctic marine fisheries. In Impacts of Climate Change on Fisheries and Aquaculture: Synthesis of Current Knowledge, Adaptation and Mitigation Options, Fisheries and Aquaculture Technical Paper 627 , pp. 113 – 138 . Ed. by Barange M., Bahri T., Beveridge M. C. M., Cochrane K. L., Funge-Smith S., Poulain F. FAO , Rome . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Hunt G. L. Jr , Coyle K. O., Eisner L., Farley E. V., Heintz R., Mueter F., Napp J. M., et al. . 2011 . Climate impacts on eastern Bering Sea foodwebs: a synthesis of new data and an assessment of the Oscillating Control Hypothesis . ICES Journal of Marine Science , 68 : 1230 – 1243 . Google Scholar Crossref Search ADS WorldCat Hunt G. L. Jr , Stabeno P., Walters G., Sinclair E., Brodeur R. D., Napp J. M., Bond N. A. 2002 . Climate change and control of the southeastern Bering Sea pelagic ecosystem . Deep Sea Research Part II: Topical Studies in Oceanography , 49 : 5821 – 5853 . Google Scholar Crossref Search ADS WorldCat IPCC. 2013 . Climate change 2013: the physical science basis. In Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change . Ed. by Stocker T. F., Qin D., Plattner G.-K., Tignor M. M. B., Allen S. K., Boschung J., Nauels A., et al. . Cambridge University Press , Cambridge, UK and New York . 1535 pp. https://doi.org/10.1017/CBO9781107415324 (last accessed 9 January 2019). Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC IPCC. 2014 . Climate change 2014: impacts, adaptation, and vulnerability, Part. B: regional aspects. In Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change . Ed. by Barros V. R., Field C. B., Dokken D. J., Mastrandrea M. D., Mach K. J., Bilir T. E., Chatterjee M., et al. . Cambridge University Press , Cambridge, UK and New York . 688 pp. Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Kachel N. B. , Hunt G. L. Jr, Salo S. A., Schumacher J. D., Stabeno P. J., Whitledge T. E. 2002 . Characteristics and variability of the inner front of the southeastern Bering Sea . Deep Sea Research Part II: Topical Studies in Oceanography , 49 : 5889 – 5909 . Google Scholar Crossref Search ADS WorldCat Kay J. E. , Deser C., Phillips A., Mai A., Hannay C., Strand G., Arblaster J. M., et al. . 2015 . The Community Earth System Model (CESM) large ensemble project: a community resource for studying climate change in the presence of internal climate variability . Bulletin of the American Meteorological Society , 96 : 1333 – 1349 . Google Scholar Crossref Search ADS WorldCat Kinder T. H. , Schumacher J. D. 1981 . Hydrographic structure over the continental shelf of the southeastern Bering Sea. In Eastern Bering Sea Shelf: Oceanography and Resources , Vol. 1 , pp. 31 – 51 . Ed. by Hood D.W., Calder J. A. USDOC/NOAA/OMPA , Washington, D.C . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Knutti R. , Masson D., Gettelman A. 2013 . Climate model genealogy: generation CMIP5 and how we got there . Geophysical Research Letters , 40 : 1194 – 1199 . Google Scholar Crossref Search ADS WorldCat Large W. G. , McWilliams J. C., Doney S. C. 1994 . Oceanic vertical mixing: a review and a model with a nonlocal boundary layer parameterization . Reviews of Geophysics , 32 : 363 – 403 . Google Scholar Crossref Search ADS WorldCat Large W. G. , Yeager S. G. 2009 . The global climatology of an interannually varying air-sea flux data set . Climate Dynamics , 33 : 341 – 364 . Google Scholar Crossref Search ADS WorldCat Li L. , McClean J., Miller A., Eisenman I., Hendershott M., Papadopoulos C. 2014 . Processes driving sea ice variability in the Bering Sea in an eddying ocean/sea ice model: mean seasonal cycle . Ocean Modelling , 84 : 51 – 66 . Google Scholar Crossref Search ADS WorldCat Li L. , Miller A., McClean J., Eisenman I., Hendershott M. 2014 . Processes driving sea ice variability in the Bering Sea in an eddying ocean/sea ice model: anomalies from the mean seasonal cycle . Ocean Dynamics , 64 : 1693 – 1717 . Google Scholar Crossref Search ADS WorldCat Meehl G. A. , Covey C., Delworth T., Latif M., McAvaney B., Mitchell J. F. B., Stouffer R. J., et al. . 2007 . The WCRP CMIP3 multimodel dataset: a new era in climate change research . Bulletin of the American Meteorological Society , 88 : 1383 – 1394 . Google Scholar Crossref Search ADS WorldCat Newman M. , Sardeshmukh P. D., Winkler C. R., Whitaker J. S. 2003 . A study of subseasonal predictability . Monthly Weather Review , 131 : 1715. Google Scholar Crossref Search ADS WorldCat Ortiz I. , Aydin K., Hermann A. J., Gibson G., Punt A. E., Wiese F., Eisner L. B., et al. . 2016 . Climate to fish: synthesizing field work, data and models in a 39-year retrospective analysis of seasonal processes on the eastern Bering Sea shelf and slope . Deep Sea Research Part II: Topical Studies in Oceanography , 134 : 390 – 412 . Google Scholar Crossref Search ADS WorldCat Preisendorfer R. W. 1988 . Principal Component Analysis in Meteorology and Oceanography , Elsevier , Amsterdam . 425 pp. Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Ressler P. H. , De Robertis A., Kotwicki S. 2014 . The spatial distribution of euphausiids and walleye pollock in the eastern Bering Sea does not imply top-down control by predation . Marine Ecology Progress Series , 503 : 111 – 122 . Google Scholar Crossref Search ADS WorldCat Ressler P. H. , De Robertis A., Warren J. D., Smith J. N., Kotwicki S. 2012 . Developing an acoustic index of euphausiid abundance to understand trophic interactions in the Bering Sea ecosystem . Deep Sea Research Part II: Topical Studies in Oceanography , 65–70 : 184 – 195 . Google Scholar Crossref Search ADS WorldCat Rykaczewski R. R. , Dunne J. P. 2010 . Enhanced nutrient supply to the California Current Ecosystem with global warming and increased stratification in an earth system model . Geophysical Research Letters , 37 : L21606 . Google Scholar Crossref Search ADS WorldCat Shchepetkin A. F. , McWilliams J. C. 2005 . The regional oceanic modeling system (ROMS): a split-explicit, free-surface, topography-following-coordinate oceanic model . Ocean Modelling , 9 : 347 – 404 . Google Scholar Crossref Search ADS WorldCat Siddon E. , Zador S. 2017 . Ecosystem Considerations 2017: Status of the Eastern Bering Sea Marine Ecosystem, Stock Assessment and Fishery Evaluation Report. North Pacific Fishery Management Council, Anchorage, AK. Sigler M. F. , Napp J. M., Stabeno P. J., Heintz R. A., Lomas M. W., Hunt G. L. 2016 . Variation in annual production of copepods, euphausiids, and juvenile walleye pollock in the southeastern Bering Sea . Deep Sea Research Part II: Topical Studies in Oceanography , 134 : 223 – 234 . Google Scholar Crossref Search ADS WorldCat Stabeno P. J. , Danielson S., Kachel D., Kachel N. B., Mordy C. W. 2016 . Currents and transport on the eastern Bering Sea shelf: an integration of over 20 years of data . Deep Sea Research Part II: Topical Studies in Oceanography , 134 : 13 – 29 . Google Scholar Crossref Search ADS WorldCat Stabeno P. J. , Duffy-Anderson J. T., Eisner L. B., Farley E. V., Heintz R. A., Mordy C. W. 2017 . Return of warm conditions in the southeastern Bering Sea: physics to fluorescence . PLoS One , 12 : e0185464 . Google Scholar Crossref Search ADS PubMed WorldCat Stabeno P. J. , Kachel N. B., Moore S. E., Napp J. M., Sigler M., Yamaguchi A., Zerbini A. N. 2012 . Comparison of warm and cold years on the southeastern Bering Sea shelf and some implications for the ecosystem . Deep Sea Research Part II: Topical Studies in Oceanography , 65–70 : 14 – 30 . Google Scholar Crossref Search ADS WorldCat Stabeno P. J. , Napp J., Mordy C., Whitledge T. 2010 . Factors influencing physical structure and lower trophic levels of the eastern Bering Sea shelf in 2005: sea ice, tides and winds . Progress in Oceanography , 85 : 180 – 196 . Google Scholar Crossref Search ADS WorldCat Sullivan M. E. , Kachel N. B., Mordy C. W., Salo S. A., Stabeno P. J. 2014 . Sea ice and water column structure on the eastern Bering sea shelf . Deep Sea Research Part II: Topical Studies in Oceanography , 109 : 39 – 56 . Google Scholar Crossref Search ADS WorldCat Taylor K. E. , Stouffer R. J., Meehl G. A. 2012 . An overview of CMIP5 and the experiment design . Bulletin of the American Meteorological Society , 93 : 485 – 498 . Google Scholar Crossref Search ADS WorldCat Watanabe S. , Hajima T., Sudo K., Nagashima T., Takemura T., Okajima H., Nozawa T., et al. . 2011 . MIROC-ESM 2010: model description and basic results of CMIP5-20c3m experiments . Geoscientific Model Development , 4 : 845 – 872 . Google Scholar Crossref Search ADS WorldCat Appendix The steps of our hybrid statistical procedure, based on dynamically downscaled output, may be summarized as follows: 1. Calculate individual spatial modes (EOFs) of each variable with associated time series (PCs). For this analysis, we use the series of annual averages at each bin prepared as described in “Univariate analyses” section, with each of the ensemble members concatenated together after removal of the time mean. Vilt=∑jXjilTjit where V represents the demeaned and concatenated data, X represents the spatial EOFs, T represents the modulating time series, and the indices i, j, l, t represent variable type i, spatial location l, mode number j, and time index t. In this decomposition, T has zero mean and unit variance. 2. Calculate “multivariate PCs” using the univariate Ts as the input set of time series. (In practice the best results are obtained when Ts are each scaled by the percent variance of V explained by that univariate mode). We are now calculating EOFs across variables, rather than across space, decomposing the univariate Ts into a single multivariate set of time functions Γand variable loadings M: Tjit=∑kMkjiΓkt where k represents the multivariate mode number. This ultimately yields a new set of multivariate PCs (multivariate time series) with associated loadings (multivariate spatial modes) for each variable: Vilt=∑jXjil∑kMkjiΓkt=∑kΓkt∑jMkjiXjil=∑kCkilΓkt Note how the new multivariate spatial modes C are simply a linear combination of the univariate spatial modes X, weighted by M. This new basis set retains the orthogonality property of the original EOF basis set for each variable, as it entails the multiplication of one orthogonal matrix by another. Hence it can in theory be used to fully reconstruct the original data, in a manner which compactly emphasizes the covariance among the dominant spatial patterns of different variables. 3. For convenience, we convolve each of the multivariate time series with the original time series for each variable at each spatial bin to obtain the new multivariate spatial modes: Ckil=∑tΓktVilt In general, these will look different than the original univariate EOFs, as they represent the dominant spatial patterns of variables rising and falling together through time. 4. Now use the multivariate modes derived from the training set as predictors of the regional response, given only a new set of coarse, large-scale atmospheric/oceanic forcing. We project the new set of available forcing variables (in our application these are atmospheric variables from other CMIP5 members) onto the multivariate basis set at each time step, to obtain the amplitude of each mode through time. Ideally, this allows us to estimate the covarying regional variables not contained in that new set, as a proxy for what would have appeared if we had conducted a full dynamical downscaling of that forcing through the regional model. For each forcing variable i at future time t we construct the estimate as follows: Γikt*=∑lViltCkil/∑lCkilCkil Consider each forcing variable as an independent estimate of the modal amplitude. Some of these estimates are likely to be more accurate, as some of the variables are captured more fully by that mode. Hence for our final estimate of the modal amplitudes at each future time, we weight the individual estimates by the percent variance of that variable explained by that multivariate mode: Rik=∑ltViltCkil/∑ltViltVilt∑ltCkilCkil1/2 Γkt**=∑iΓikt*Rik/∑iRik So our estimate of the “unobserved” (regional biophysical response) variables at some future time, based on the projection of “observed” (large-scale atmospheric forcing) variables onto the multivariate modes, becomes: Vestilt=∑kCkilΓkt** This estimate can be further tuned by linearly regressing each Vestilt against the original Vilt over the original time sequence t; the slopes bilof those regressions may be used in our final estimates: Vfinalilt=bilVestilt In practice, it was found that this tuning made little difference to the results. © International Council for the Exploration of the Sea 2019. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. © International Council for the Exploration of the Sea 2019.
Event scale and persistent drivers of fish and macroinvertebrate distributions on the Northeast US ShelfFriedland, Kevin, D;McManus, M, Conor;Morse, Ryan, E;Link, Jason, S
doi: 10.1093/icesjms/fsy167pmid: N/A
Abstract The Northeast US Continental Shelf Large Marine Ecosystem is experiencing warming at a rate exceeding that of many other large marine ecosystems and has undergone significant climate-related changes. We examined the effect of thermal events and shifting patterns of primary and secondary productivity on the distribution of fish and macroinvertebrate species during the period 1968–2016. Though subject to inter-annual change, the along-shelf centre of gravity of chlorophyll concentration lacked a trend. Similarly, zooplankton bio-volume and total abundance along-shelf distance were also found to be without trend. However, the trend in the centre of gravity of copepod taxa diverged from the trends in bio-volume and non-copepod zooplankton abundance, suggesting most of these taxa had shifted in distribution to the northeast. The centres of gravity of fish and macroinvertebrate species have trended significantly to the northeast, suggesting copepods may play a key role in the distribution of higher trophic levels. Analysis of thermal events suggest that abrupt change in temperature can actuate persistent change in the distribution of fish and macroinvertebrate species. In aggregate, these broad trophic level patterns imply that distributional changes affecting upper trophic levels were dominated by thermal mechanisms, whereas lower trophic productivity, although subject to the same thermal conditions, exhibited less of a response. We hypothesize this lack of distributional response at lower trophic levels is due to their higher rates of production and turnover, and hence reflect a capacity to better integrate seasonal thermal changes. Furthermore, distributional changes of upper trophic levels may also be significantly impacted by feeding interactions at specific life history stages, where temperature affects both predator and prey. Introduction In recent decades, the thermal conditions for the Northeast US Continental Shelf Large Marine Ecosystem (hereafter the Northeast US Shelf, or NEUS) have changed, with the ecosystem experiencing a rate of warming among the fastest worldwide (Belkin, 2009; Mills et al., 2013; Pershing et al., 2015). Such warming can have profound impacts on the ecophysiology of marine organisms (Neill et al., 1994) through the biological processes influenced by temperature (Brett, 1979). Hence, changes in temperature can alter the habitats of marine fish and macroinvertebrates with significant impacts on growth and mortality (Anderson, 1988; Pepin, 1991), with these impacts often focused on the essential life history stages of a species (Anderson et al., 2013). Contrasting regions of changing habitat quality of a species will likely cause local abundances to vary with concomitant changes in species distribution. For many NEUS marine fish and macroinvertebrate taxa, thermal habitat preferences are well understood (Kleisner et al., 2017), which when coupled with the expected warming of this ecosystem under future climate scenarios (Saba et al., 2016), suggest that the distribution of many species may shift beyond current ranges. This creates an expectation that the ecosystem will be redefined and the manner in which humans interact with it will change as well (Hare et al., 2016). In the NEUS, the influence of climate change on marine fish populations has often been represented through changes in latitude and depth distribution (Nye et al., 2009; Lucey and Nye, 2010; Pinsky and Fogarty, 2012; Pinsky et al., 2013a; Kleisner et al., 2015, 2016). More nuanced work has addressed changes in the population abundance over species-specific life stages (Walsh et al., 2015) and how distribution changes may be related to changes in habitat suitability for a given species (Hare et al., 2012; Lynch et al., 2015; McManus et al., 2018). Several methods have been used to quantify spatial distribution shifts for marine populations; each varying based on the abundance and distance units of measure and desired outcome products (Pinsky et al., 2013b; Thorson et al., 2016; Adams, 2017). Centre of gravity estimates have been one of the more commonly used indicators used to assess changes in spatial distribution (Woillez et al., 2007), which provide insight into the average movement over time. While spatial distribution indicators and thermal relationships have been described for many NEUS marine fish and macroinvertebrates, similar evaluations for lower trophic levels in the region are fewer. The NEUS has higher trophic connectivity than many other marine systems (Link et al., 2010), making it imperative to understand changes in distribution across trophic levels. Furthermore, this high level of connectivity suggests that adaptability rather than specificity of predators may play a larger role in shaping marine distributions than thermal and thermally related stressors. As a significant prey source for pelagic fish and the early life history stages of many fish species, zooplankton, particularly copepods, have been used as an indicator of marine ecosystem regime shifts (Sherman et al., 1998). Long-term changes in zooplankton biomass and species composition have also been linked to large shifts in the biomass and recruitment success of upper trophic levels in the NEUS and other marine systems (Beaugrand et al., 2003; Alheit and Niquen, 2004; Mackas et al., 2007; Hipfner, 2008; Bi et al., 2011; Hunt et al., 2011; Tanasichuk and Routledge, 2011; Friedland et al., 2013). In turn, zooplankton have a trophic dependence on phytoplankton, which defines the length of the food chain and the pathways from primary productivity to higher trophic levels (Ryther, 1969; Canales et al., 2016). While phytoplankton constitute the base of the NEUS food web, evaluating both phytoplankton and zooplankton are important when assessing trophic impacts on fish and macroinvertebrate populations. For example, on global scales, the amount of primary productivity channeled to mesozooplankton is more highly correlated with fishery yields than primary productivity itself (Friedland et al., 2012). Shifts in fish and macroinvertebrates are well documented in this ecosystem, yet those of zooplankton and phytoplankton are relatively under examined. To understand how the different trophic levels of the NEUS ecosystem are responding to climate change and ocean warming, we investigated prospective distribution shifts for phytoplankton, zooplankton, and fish and macroinvertebrate taxa over a period of five decades. Three centre of gravity metrics were used to characterize distributional change: distance along the coast, distance to the coast, and depth. The different metrics allowed us to compare the various potential responses a given trophic level may exhibit in response to a set of forcing factors. These metrics account for potential movements latitudinal along the coast and for response to gradients that may be associated with depth and coastal influences. The observed changes in distribution were further tested using change-point detection statistics to determine whether event-scale processes drove the changes in observed distribution and whether species return to their previously established locations after an event. Finally, the temperatures of occurrence for species were compared with ocean temperature to evaluate its role in distributional shifts across trophic levels. Material and methods Study system and distribution metrics A long-term monitoring program for the NEUS ecosystem has been measuring fish and macroinvertebrate populations for approximately five decades over the period 1968–2016. Complimentary programs have measured zooplankton populations within a similar period, 1977–2015, noting that were interruptions to these time series in the 1990s. Chlorophyll a concentration has not been measured synoptically on the Northeast Shelf in a consistent fashion by any monitoring program; therefore, we relied on remote sensing data to characterize the distribution of primary producers during the period 1979–2016, noting there was an interruption in the remote sensing data during the late 1980s into the early 1990s. Collectively these data, which will be described in detail below, allow for the comparison of spatial–temporal shifts over multiple trophic levels within the ecosystem. Change in distribution was characterized with three spatial distribution metrics applied to chlorophyll a concentration, zooplankton bio-volume, zooplankton taxon abundance, and fish and macroinvertebrate taxon abundance within the extent of the ecosystem (Figure 1 a). The three centre of gravity metrics included along-shelf distance, depth of occurrence, and distance to the coastline. Along-shelf distance was taken as the distance from the origin of a transect originating at 76.53°W 34.60°N extending to 65.71°W 43.49°N at a point closest to the position of the subject taxon. The distance is expressed in km, with examples of along-shelf position estimates illustrated in Figure 1a; lower along-shelf distances correspond to positions in the southwest portion of the ecosystem and higher values more in the northeast. The along-shelf distance centre of gravity was the weighted mean distance using the abundance measure of the subject taxon as the weighting factor. Depth of occurrence represents the depth (depth of the seabed) associated the abundance measure, the centre of gravity depth expressed in meter. Distance to the coast is the distance to the closest position on the coastline associated with the abundance measure, the centre of gravity distance expressed in units of kilometers. Figure 1. Open in new tabDownload slide Map of the US Northeast Shelf showing the extent of the ecosystem as shaded region (a). Line marks along-shelf reference line with reference distances marked. Maps of differential depth (b) and distance to the coast (c) in the ecosystem with lighter red regions representing the areas less than the median in terms of depth and distance, darker blue regions greater than the median. Figure 1. Open in new tabDownload slide Map of the US Northeast Shelf showing the extent of the ecosystem as shaded region (a). Line marks along-shelf reference line with reference distances marked. Maps of differential depth (b) and distance to the coast (c) in the ecosystem with lighter red regions representing the areas less than the median in terms of depth and distance, darker blue regions greater than the median. Water temperature Surface and bottom water temperature was considered a potential factor in actuating distribution events. Temperature fields for the extent of the ecosystem were developed using an optimal interpolation approach where annual data were combined with seasonal climatologies over the period 1968–2016. Temperature on the NEUS was collected using conductivity/temperature/depth (CTD) instruments, with the most complete sample coverage in spring (February–April) and fall (September–November). To correct for the differences in date of collection between years, temperatures were standardized to a collection date of April 3 for spring collections and 11 October for fall, which were the mean dates for the data collections by season. The corrections were based on linear regressions of temperature versus day of the year for a half-degree grid of the ecosystem. For the same grid, mean bottom temperature was calculated by year and season. For grid locations that had data for at least 80% of the time series, which preserved most of the locations on the shelf, the data from those locales were used to calculate a seasonal mean temperature. The annual seasonal means were used to calculate temperature anomalies, which were combined over the time series to provide seasonal surface and bottom temperature anomaly climatology. Returning to the raw data, the observations for a year, season, and depth were then used to estimate an annual field using universal kriging with depth as a covariate (R package automap ver. 1.0–14). The annual field was then combined with the climatology anomaly field, adjusted by the annual mean, using the variance field from the kriging as the basis for a weighted mean between the two. The variance field was divided into quartiles with the first quartile (lowest kriging variance) carrying with it a weighting of 4:1 between the annual to climatology values. Hence, the optimal interpolated field at these locations were skewed toward the annual data reflecting their proximity to actual data locations, reflected by low kriging variance associated with them. The weighting ratio shifted to 1:1 in the highest variance quartile reflecting less information from the annual field and more from the climatology. The temperature fields were applied as yearly seasonal means and as temperature differentials based on two partitioning schemes associated with the centre of gravity metrics. In one scheme, the ecosystem was divided by depth with a shallow zone based on the depth locations below the median depth and a deep zone for those above the median (Figure 1b). The temperature depth differential was computed as the mean for the shallow minus the mean of the deep. In the other scheme, the ecosystem was divided by distance to the coast with a close zone based on the distance locations below the median distance and a distance zone for those above the median (Figure 1c). The temperature distance differential was computed as the mean for the close minus the mean of the far. Chlorophyll concentration We analysed centre of gravity of chlorophyll a concentration using data extracted from remote-sensing databases. Chlorophyll a concentration (mg m−3) data from 1979 to 1986 were provided by the Coastal Zone Color Scanner (CZCS) sensor (Gregg and Conkright, 2002), available from the Ocean Color Website (oceancolor.gsfc.nasa.gov/). The data were analysed at a 9 km-resolution and resampled to match a 0.1° grid over the extent of the ecosystem (Figure 1a). Chlorophyll a concentration during the period 1997–2016 was based on measurements made with the Sea-viewing Wide Field of View Sensor (SeaWiFS), Moderate Resolution Imaging Spectroradiometer on the Aqua satellite (MODIS), Medium Resolution Imaging Spectrometer (MERIS), and Visible and Infrared Imaging/Radiometer Suite (VIIRS) sensors. We used the Garver, Siegel, Maritorena Model (GSM) merged data product obtained from the Hermes GlobColour website (hermes.acri.fr/index.php). These four-sensor time series were combined using a bio-optical model inversion algorithm (Maritorena et al., 2010). The data was downloaded at a 4 km resolution and resampled to match a 0.1° grid over the extent of the ecosystem. Annual mean spring and fall chlorophyll a concentration maps were developed, from which along-shelf distance, depth, and distance to the coast centres of gravity for chlorophyll a concentration were calculated. We evaluated the time series changes in distribution metrics using Mann–Kendall non-parametric trend analysis. We calculated Kendall’s tau test for the significance (two-tailed test) of a monotonic time series trend (Mann, 1945) for centre of gravity of along-shelf distance, depth, and distance to the coast. We also calculated Theil–Sen slopes of trend, which is the median slope joining all pairs of observations (R package wql, version 4.9). Zooplankton bio-volume and abundance Zooplankton abundance has been measured within the context of multiple sampling programs with varying sampling designs. The two most comprehensive monitoring programs over the study period were the Marine Resources Monitoring, Assessment, and Prediction Program or MARMAP (1977–1987) and the Ecosystem Monitoring Program or EcoMon (1992–present) programs, both serving as shelf-wide surveys of the ecosystem (Sherman et al., 1998; Kane, 2007). In addition, there were two sampling programs with more specific spatial and temporal foci, the Herring–Sandlance study (1988–1993) and the US GLOBEC program (1994–1999), both of which were mainly focused on sampling in the Georges Bank area (Beardsley et al., 2003). The raw zooplankton abundance data are made publicly available at ftp://ftp.nefsc.noaa.gov/pub/hydro/zooplankton_data/. In all of these surveys, zooplankton were sampled obliquely through the water column to within 5 m of the sea floor or a maximum depth of 200 m using paired 61-cm Bongo samplers equipped with 333-micron mesh nets. However, sample distribution has not been consistent through time, with segments of the time series comprised of differing combinations of fixed, random, and regionally focused stations. In addition, vessel availability has resulted in years lacking a distributional balance in zooplankton samples over the ecosystem. From this collection of sample stations, we considered candidate taxa to represent the distribution of secondary production in the ecosystem. For each station, we analysed the settled bio-volume (a proxy for total planktonic biomass) in the units of ml per m3, and taxa-specific abundance in units of number per m3. From a set of 38 candidate taxa (i.e. these higher abundance taxa), 21 met the criterion of occurring in at least 38 years of one or both of the seasonal time series. This metric resulted in 17 taxa for the spring data (9 of which were copepod species) and 20 taxa in fall, with 11 representing copepods. Of these taxa, 16 occurred in both seasons (Supplementary Table S1). Because of the aforementioned issues of sample distribution, we elected to interpolate the zooplankton data over the extent of the ecosystem (Figure 1a) using ordinary kriging; the data was logarithm (log10(x + 1)) transformed before interpolation. The interpolated fields by year and season were then used to determine the centre of gravity of along-shelf distance, depth, and distance to the coast for seasonal bio-volume and zooplankton taxa. The overall time series of zooplankton data extended from 1977 to 2015; however, due to data adequacy concerns, the years 1989, 1990, 1991, and 1994 of the spring data and years 1989, 1990, and 1992 of the fall data were excluded from analyses. Fish and macroinvertebrate species abundance The principal fishery independent survey on the NEUS is the bottom trawl survey conducted by the Northeast Fisheries Science Center, which provides both spatial and temporal depictions of fish and macroinvertebrate abundances (Grosslein, 1969). The survey began in the fall of 1963, but for consistency, we restricted the analysis to the years 1968–2016. The survey data are publicly available at http://comet.nefsc.noaa.gov/cgi-bin/ioos/ioos.pl. Surveys are conducted in the spring and fall and are based on a stratified random design. Catch from the survey tows can be identified to species, genus, functional group, or an unidentified category. We restricted this analysis to catch identified to species since most of the catch is identified to this level and provides the most reliable information on change in fish and macroinvertebrate distribution. Species were also assigned functional groups based on their adult prey preference and vertical presence to assess broader ecosystem changes: benthivores, demersal piscivores, pelagic piscivores, or planktivores. We first considered seasonal lists of candidate species that included taxa that occurred in at least 100 tows during the study time series. This exercise produced lists of 122 and 174 species in spring and fall, respectively. The distribution metrics were then calculated for each candidate species by season. In spring, 72 of the candidate species had catches distributed over the time series and were thus included in the analysis (Supplementary Table S2). Another 16 taxa were found to have catch distributed over only part of the time series, and though of interest in terms of distributional change, these partial time series raised concerns over change in availability and identification of these taxa in the survey. In fall, 79 of the candidate species had catches distributed over the time series (Supplementary Table S3); another 34 were found to have partial time series. Our centre of gravity calculations were similar to the methods used in previous analyses of these data (Nye et al., 2009; Bell et al., 2015) that applied calculations to catch from a core set of survey strata consistently sampled over the time series. In addition, we applied a tow per strata weighting to ameliorate the differences in the allocation of stations to strata (Bell et al., 2015). Change in distribution events Distributional change events for fish and macroinvertebrates were identified using change point statistics and contextualized with step change data from bottom and surface temperature time series. Change events for fish and macroinvertebrates were detected using the sequential averaging algorithm Sequential t-test Analysis of Regime Shifts, or STARS (Rodionov, 2004, 2006), which finds the change points in a time series. Consistent with other applications of this approach (Lavaniegos and Ohman, 2007), the STARS algorithm parameters were specified a priori: the alpha level used to test for a change in the mean was set to α = 0.1; the length criteria, the number of time steps to use when calculating the mean level of a new regime, was set to 10; and, the Huber weight parameter, which determines the relative weighting of outliers in the calculation of the regime mean, was set to 3. Positive and negative change points were summed each year. A year with a total number of change points equal or greater than two standard deviations above the mean were considered of interest for further analysis. For events that occurred early in the times, the species that had a change in distribution were examined further to determine how long after the event they remained at the new distribution. The intention was to provide evidence as to whether an event results in a persistent change in distribution. For comparison, time series of step change in bottom and surface temperature were represented as residuals from a loess fit curve, noting that the fits were not used to suggest significance, but instead to offer a consistent basis to visualize potential outliers. The same treatment was applied to the bottom and surface depth differential temperature and distance to coast temperature differential. Occupancy bottom temperature Occupancy temperature was taken as the mean standardized bottom temperature associated with each tow by species. For each spring taxon with data over the full time series, the standardized bottom temperature for tows within the study domain were averaged, and in turn, an annual mean was taken of the species means. The same procedure was done for the fall taxa with data over the full time series. Species trends were evaluated with Mann–Kendall non-parametric trend analysis. Multivariate analysis Multivariate relationships between fish and macroinvertebrate distribution responses and zooplankton and temperature explanatory variables were analysed with canonical correlation analysis. We correlated a response variable set that included the along-shelf distribution metrics for benthivores, demersal piscivores, pelagic piscivores, and planktivores to an independent variable set consisting of distribution metrics for other zooplankton and copepods and bottom and surface temperature indices (R package CCA ver. 1.2). For each zooplankton category, along-shelf, depth, and distance to the coast metrics were variables. The temperature variable set included bottom and surface temperature, depth temperature differential, and distant to the coast temperature differential. This was repeated for two other response sets of variables including fish and macroinvertebrate depth and distance to the coast distance metrics with the same independent variable sets. Results Trends in chlorophyll distributions The centre of gravity distribution metrics for chlorophyll a concentration had some discernable trends. The spring along-shelf distance for chlorophyll a concentration was approximately 850 km at both the beginning and end of the time series (Figure 2 a). Along-shelf distance in fall was less dispersed at the beginning of the time series and suggested a centre of gravity closer to 900 km (Figure 2b). However, neither the spring nor the fall chlorophyll a concentration along-shelf distance time series were found to have significant trends (Table 1 ). The spring centre of gravity depth distribution was generally at the 80 m isobaths during the start of the time series and increased to approximately 100 m during the more contemporary time period (Figure 2c). The fall centre of gravity depth was without trend and remained at approximately 100 m throughout (Figure 2d). The spring depth trend was significant whereas the fall trend was non-significant. The spring and fall distance to the coast centre of gravity trended in similar fashion as depth, tending toward further distances from the coast. The spring time series suggested a change from 80 to 95 km from the coast (Figure 2e), whereas the fall time series suggested an increase from 85 to 95 km (Figure 2f). The spring trend was significant whereas the fall was non-significant. Table 1. Theil–Sen slope and Mann–Kendall trend test probability (p, bold indicates significance at p=0.05) for time series of annual centre of gravity of along-shelf distance, depth, and distance to the coast for spring and fall spring and fall chlorophyll concentration, zooplankton abundance and bio-volume, and fish and macroinvertebrate abundance. Along-shelf distance Depth Distance to coast Group Season Slope p Slope p Slope p Chlorophyll Spring 0.020 0.999 0.435 0.033 0.345 0.020 Chlorophyll Fall 0.211 0.707 −0.002 0.999 0.155 0.338 Zooplankton abundance Spring 0.758 0.293 −1.893 0.173 0.096 0.532 Zooplankton bio-volume Spring −1.030 0.244 −0.198 0.609 0.234 0.410 Zooplankton abundance Fall 0.216 0.817 −3.787 0.037 0.009 0.902 Zooplankton bio-volume Fall −0.060 0.892 0.103 0.614 0.237 0.105 Fish and macroinvertebrates Spring 1.807 0.000 0.075 0.365 −0.007 0.938 Fish and macroinvertebrates Fall 0.843 0.000 0.023 0.724 −0.100 0.006 Along-shelf distance Depth Distance to coast Group Season Slope p Slope p Slope p Chlorophyll Spring 0.020 0.999 0.435 0.033 0.345 0.020 Chlorophyll Fall 0.211 0.707 −0.002 0.999 0.155 0.338 Zooplankton abundance Spring 0.758 0.293 −1.893 0.173 0.096 0.532 Zooplankton bio-volume Spring −1.030 0.244 −0.198 0.609 0.234 0.410 Zooplankton abundance Fall 0.216 0.817 −3.787 0.037 0.009 0.902 Zooplankton bio-volume Fall −0.060 0.892 0.103 0.614 0.237 0.105 Fish and macroinvertebrates Spring 1.807 0.000 0.075 0.365 −0.007 0.938 Fish and macroinvertebrates Fall 0.843 0.000 0.023 0.724 −0.100 0.006 Open in new tab Table 1. Theil–Sen slope and Mann–Kendall trend test probability (p, bold indicates significance at p=0.05) for time series of annual centre of gravity of along-shelf distance, depth, and distance to the coast for spring and fall spring and fall chlorophyll concentration, zooplankton abundance and bio-volume, and fish and macroinvertebrate abundance. Along-shelf distance Depth Distance to coast Group Season Slope p Slope p Slope p Chlorophyll Spring 0.020 0.999 0.435 0.033 0.345 0.020 Chlorophyll Fall 0.211 0.707 −0.002 0.999 0.155 0.338 Zooplankton abundance Spring 0.758 0.293 −1.893 0.173 0.096 0.532 Zooplankton bio-volume Spring −1.030 0.244 −0.198 0.609 0.234 0.410 Zooplankton abundance Fall 0.216 0.817 −3.787 0.037 0.009 0.902 Zooplankton bio-volume Fall −0.060 0.892 0.103 0.614 0.237 0.105 Fish and macroinvertebrates Spring 1.807 0.000 0.075 0.365 −0.007 0.938 Fish and macroinvertebrates Fall 0.843 0.000 0.023 0.724 −0.100 0.006 Along-shelf distance Depth Distance to coast Group Season Slope p Slope p Slope p Chlorophyll Spring 0.020 0.999 0.435 0.033 0.345 0.020 Chlorophyll Fall 0.211 0.707 −0.002 0.999 0.155 0.338 Zooplankton abundance Spring 0.758 0.293 −1.893 0.173 0.096 0.532 Zooplankton bio-volume Spring −1.030 0.244 −0.198 0.609 0.234 0.410 Zooplankton abundance Fall 0.216 0.817 −3.787 0.037 0.009 0.902 Zooplankton bio-volume Fall −0.060 0.892 0.103 0.614 0.237 0.105 Fish and macroinvertebrates Spring 1.807 0.000 0.075 0.365 −0.007 0.938 Fish and macroinvertebrates Fall 0.843 0.000 0.023 0.724 −0.100 0.006 Open in new tab Figure 2. Open in new tabDownload slide Time series of chlorophyll concentration centre of gravity based on remote sensing date. Spring time frame data along-shelf distance (a), depth (c), and distance to coast (e); fall data for along-shelf distance (b), depth (d), and distance to coast (f). Red lines are linear regression lines. Figure 2. Open in new tabDownload slide Time series of chlorophyll concentration centre of gravity based on remote sensing date. Spring time frame data along-shelf distance (a), depth (c), and distance to coast (e); fall data for along-shelf distance (b), depth (d), and distance to coast (f). Red lines are linear regression lines. Trends in zooplankton distributions Most of the metrics for centre of gravity distribution for zooplankton biomass and abundance did not exhibit significant trends through time. Spring zooplankton abundance and bio-volume along-shelf distances were found to be without trend (Table 1), with positions at approximately 800 and 900 km for abundance and bio-volume, respectively (Figure 3 a and d). Along-shelf distance centre of gravity time series were nearly the same for fall abundance and bio-volume (Figure 3g and j, respectively) and their trends were found to be non-significant. Centre of gravity depth distributions for abundance and bio-volume were dramatically different. In both spring and fall, zooplankton abundance depth centre of gravity tended to exceed 100 m (Figure 3b and h, respectively). In contrast, bio-volume depth centre of gravity tended to be less than 100 m in both seasons (Figure 3e and k, respectively). The fall abundance depth centre of gravity time series trend was significant, whereas the other trends were non-significant. There were no significant trends in the zooplankton distance to the coast centre of gravity data. Zooplankton abundance distance to the coast centre of gravity data tended to be approximately 100 km (Figure 3c and i, spring and fall, respectively) whereas the bio-volume distant to the coast was closer to 80 km (Figure 3f and l, spring and fall, respectively). Figure 3. Open in new tabDownload slide Mean along-shelf distance for spring zooplankton abundance (a) and bio-volume (b) and for fall abundance (c) and bio-volume (d). Mean depth for spring zooplankton abundance (e) and bio-volume (f) and for fall abundance (g) and bio-volume (h). Mean distance to coast for spring zooplankton abundance (i) and bio-volume (j) and for fall abundance (k) and bio-volume (l). Red lines are linear regression lines. Figure 3. Open in new tabDownload slide Mean along-shelf distance for spring zooplankton abundance (a) and bio-volume (b) and for fall abundance (c) and bio-volume (d). Mean depth for spring zooplankton abundance (e) and bio-volume (f) and for fall abundance (g) and bio-volume (h). Mean distance to coast for spring zooplankton abundance (i) and bio-volume (j) and for fall abundance (k) and bio-volume (l). Red lines are linear regression lines. The trends among zooplankton bio-volume, abundance of non-copepod taxa, and copepod taxa suggest differing distributional responses among these groups. Theil–Sen slopes of centre of gravity of along-shelf distance for copepods averaged approximately 2 km year−1 where the slope for bio-volume and the mean slope for the other taxa were negative (Figure 4 a and b, spring and fall, respectively). The positive slopes indicated a shift in location to the northeast. Theil–Sen slopes of the centre of gravity for depth were near zero for bio-volume and other taxa, but the mean slopes for copepod taxa were negative in both spring and fall (Figure 4c and d, spring and fall, respectively). The negative slopes suggested a shift to deeper depths. Theil–Sen slopes of centre of gravity of distance to the coast for copepods were negative, indicating a shift to locations closer to the coast; the slope for bio-volume and the mean slope for the other taxa were both positive, indicating shifts further from shore (Figure 4e and f, spring and fall, respectively). Figure 4. Open in new tabDownload slide Mean Theil–Sen slope of centre of gravity of spring zooplankton bio-volume (volume), non-copepod zooplankton taxa (other), and copepods (copepods) along-shelf distance (a), depth (c), and distance to the coast (e); for fall along-shelf distance (b), depth (d), and distance to the coast (f). Error bars are 95% confidence intervals. Figure 4. Open in new tabDownload slide Mean Theil–Sen slope of centre of gravity of spring zooplankton bio-volume (volume), non-copepod zooplankton taxa (other), and copepods (copepods) along-shelf distance (a), depth (c), and distance to the coast (e); for fall along-shelf distance (b), depth (d), and distance to the coast (f). Error bars are 95% confidence intervals. Trends in fish and macroinvertebrate species distributions Since the late 1960s, most fish and macroinvertebrate taxa have shifted northeast, as indicated by the higher along-shelf distance centres of gravity. The mean along-shelf distance for fish and macroinvertebrates was approximately 800 km at the beginning of the time series and has increased to greater than 850 km in the most recent years in both spring and fall seasonal surveys (Figure 5 a and d). The trends for these time series were also significant (Table 1). Mean depth centre of gravity for fish and macroinvertebrates time series had no discernable trends; however, mean depth tended to be slightly higher in spring than in fall (Figure 5b and e). Seasonal differences in distance to the coast centre of gravity reflect the differences noted in depth distributions, with mean spring distances at approximately 100 km and fall distance at 90 km (Figure 5c and f). The time series trend in the fall data was significant; however, the trends may have been anomalously high values in the beginning of the time series. Figure 5. Open in new tabDownload slide Fish and macroinvertebrate mean spring along-shelf distance (a), depth (c), and distance to the coast (e), and mean fall along-shelf distance (b), depth (d), and distance to the coast (f). Red lines are linear regression lines. Figure 5. Open in new tabDownload slide Fish and macroinvertebrate mean spring along-shelf distance (a), depth (c), and distance to the coast (e), and mean fall along-shelf distance (b), depth (d), and distance to the coast (f). Red lines are linear regression lines. The trends among benthivores, demersal piscivores, pelagic piscivores, and planktivores taxa suggest differing distributional responses among these groups. Theil–Sen slopes of centre of gravity of along-shelf distance for planktivores were generally twice the rate for other taxonomic groups (Figure 6 a and b, spring and fall, respectively). All these groups had positive slopes, indicating a shift in location to the northeast, with the exception of pelagic piscivores in fall, which had mean slope of zero. Theil–Sen slopes of the centre of gravity for depth were similar for benthivores, demersal piscivores, and planktivores and distinctly lower for pelagic piscivores (Figure 6c and d, spring and fall, respectively). The negative slopes for pelagic piscivores suggest a shift to deeper depths. Theil–Sen slopes of centre of gravity of distance to the coast were similar for all groups (Figure 6e and f, spring and fall, respectively). Figure 6. Open in new tabDownload slide Mean Theil–Sen slope of centre of gravity of spring fish and macroinvertebrates by benthivore, demersal piscivore, pelagic piscivore, and planktivore along-shelf distance (a), depth (c), and distance to the coast (e); fall along-shelf distance (b), depth (d), and distance to the coast (f). Error bars are 95% confidence intervals. Figure 6. Open in new tabDownload slide Mean Theil–Sen slope of centre of gravity of spring fish and macroinvertebrates by benthivore, demersal piscivore, pelagic piscivore, and planktivore along-shelf distance (a), depth (c), and distance to the coast (e); fall along-shelf distance (b), depth (d), and distance to the coast (f). Error bars are 95% confidence intervals. Distribution change events Twelve distributional change events were identified in the time series of fish and macroinvertebrate distribution metrics, two of which occurred early enough in the time series to test whether these events led to persistent changes in distribution. On average, four taxa had an identifiable change point in a given year. Most change point events were found in the along-shelf distance data. Two events, in 1986 and 2013, were found in the spring along-shelf data (Figure 7 a). The greatest number of events were found in the fall along-shelf data including an event in 1977 and four additional events at the end of the time series (Figure 7b; Table 2 ). There were three events in the depth data (Figure 7c and d) and two in the distance to the coast data (Figure 7e and f); these events were near the end of the series and were not analysed further since persistent changes could not be tested. Table 2. Year and season associated with total change points at or above two standard deviations above the mean. Season Parameter Event year Year bin Count Species Spring Along-shelf distance 1986 6 1 Scyliorhinus retifer 11 1 Conger oceanicus 18 1 Geryon quinquedens 24 2 Malacoraja senta, Loligo pealeii 32 10 Dipturus laevis, Alosa pseudoharengus, Urophycis regia, Pseudopleuronectes americanus, Helicolenus dactylopterus, Myoxocephalus octodecemspinosus, Hemitripterus americanus, Macrozoarces americanus, Lophius americanus, Homarus americanus Along-shelf distance 2013 Depth 2016 Fall Along-shelf distance 1977 11 3 Prionotus carolinus, Centropristis striata, Prionotus evolans 12 1 Leucoraja erinacea 15 1 Trachurus lathami 25 1 Cancer irroratus 28 1 Merluccius bilinearis 40 4 Mustelus canis, Enchelyopus cimbrius, Raja eglanteria, Micropogonias undulates Along-shelf distance 2009 Along-shelf distance 2014 Along-shelf distance 2015 Along-shelf distance 2016 Depth 2009 Depth 2015 Distance to coast 2015 Distance to coast 2016 Season Parameter Event year Year bin Count Species Spring Along-shelf distance 1986 6 1 Scyliorhinus retifer 11 1 Conger oceanicus 18 1 Geryon quinquedens 24 2 Malacoraja senta, Loligo pealeii 32 10 Dipturus laevis, Alosa pseudoharengus, Urophycis regia, Pseudopleuronectes americanus, Helicolenus dactylopterus, Myoxocephalus octodecemspinosus, Hemitripterus americanus, Macrozoarces americanus, Lophius americanus, Homarus americanus Along-shelf distance 2013 Depth 2016 Fall Along-shelf distance 1977 11 3 Prionotus carolinus, Centropristis striata, Prionotus evolans 12 1 Leucoraja erinacea 15 1 Trachurus lathami 25 1 Cancer irroratus 28 1 Merluccius bilinearis 40 4 Mustelus canis, Enchelyopus cimbrius, Raja eglanteria, Micropogonias undulates Along-shelf distance 2009 Along-shelf distance 2014 Along-shelf distance 2015 Along-shelf distance 2016 Depth 2009 Depth 2015 Distance to coast 2015 Distance to coast 2016 Notes: For along-shelf position in spring 1986 and in fall 1977, the number of years a taxa remained at the new level associated with the initial change is summarized by duration of the change (year bin) and the frequency of these changes (count), and the mean and median of the duration. Open in new tab Table 2. Year and season associated with total change points at or above two standard deviations above the mean. Season Parameter Event year Year bin Count Species Spring Along-shelf distance 1986 6 1 Scyliorhinus retifer 11 1 Conger oceanicus 18 1 Geryon quinquedens 24 2 Malacoraja senta, Loligo pealeii 32 10 Dipturus laevis, Alosa pseudoharengus, Urophycis regia, Pseudopleuronectes americanus, Helicolenus dactylopterus, Myoxocephalus octodecemspinosus, Hemitripterus americanus, Macrozoarces americanus, Lophius americanus, Homarus americanus Along-shelf distance 2013 Depth 2016 Fall Along-shelf distance 1977 11 3 Prionotus carolinus, Centropristis striata, Prionotus evolans 12 1 Leucoraja erinacea 15 1 Trachurus lathami 25 1 Cancer irroratus 28 1 Merluccius bilinearis 40 4 Mustelus canis, Enchelyopus cimbrius, Raja eglanteria, Micropogonias undulates Along-shelf distance 2009 Along-shelf distance 2014 Along-shelf distance 2015 Along-shelf distance 2016 Depth 2009 Depth 2015 Distance to coast 2015 Distance to coast 2016 Season Parameter Event year Year bin Count Species Spring Along-shelf distance 1986 6 1 Scyliorhinus retifer 11 1 Conger oceanicus 18 1 Geryon quinquedens 24 2 Malacoraja senta, Loligo pealeii 32 10 Dipturus laevis, Alosa pseudoharengus, Urophycis regia, Pseudopleuronectes americanus, Helicolenus dactylopterus, Myoxocephalus octodecemspinosus, Hemitripterus americanus, Macrozoarces americanus, Lophius americanus, Homarus americanus Along-shelf distance 2013 Depth 2016 Fall Along-shelf distance 1977 11 3 Prionotus carolinus, Centropristis striata, Prionotus evolans 12 1 Leucoraja erinacea 15 1 Trachurus lathami 25 1 Cancer irroratus 28 1 Merluccius bilinearis 40 4 Mustelus canis, Enchelyopus cimbrius, Raja eglanteria, Micropogonias undulates Along-shelf distance 2009 Along-shelf distance 2014 Along-shelf distance 2015 Along-shelf distance 2016 Depth 2009 Depth 2015 Distance to coast 2015 Distance to coast 2016 Notes: For along-shelf position in spring 1986 and in fall 1977, the number of years a taxa remained at the new level associated with the initial change is summarized by duration of the change (year bin) and the frequency of these changes (count), and the mean and median of the duration. Open in new tab Figure 7. Open in new tabDownload slide Number of positive and negative change points in spring along-shelf distance (a), depth (c), and distance to the coast (e); fall along-shelf distance (b), depth (d), and distance to the coast for fish and macroinvertebrate species. Dashed line marks two standard deviations above the mean. Figure 7. Open in new tabDownload slide Number of positive and negative change points in spring along-shelf distance (a), depth (c), and distance to the coast (e); fall along-shelf distance (b), depth (d), and distance to the coast for fish and macroinvertebrate species. Dashed line marks two standard deviations above the mean. Two along-shelf distance events occurred early in the time series and were evaluated in respect to persistence in distributional change. The spring 1986 event involved 15 species that had an abrupt change in distribution (Table 2). Ten of these species stayed at the new distribution centre of gravity for the rest of the time series with three appearing to return to their original distribution within two decades and the rest after approximately one decade. The group of species persisting at new distributions was taxonomically diverse including an elasmobranch, a clupeid, a pelagic piscivore, a crustacean, and a host of benthic species. The species responding to this event stayed at their new distribution for more than two decades on average. The fall 1977 event involved 11 species, 4 of which stayed at the new distribution to the end of the time series with 2 returning to their original distribution after approximately three decades and the rest after approximately one decade. A Crustacean, two elasmobranchs, a pelagic piscivore, and two benthic species were among the persistent migrating species, so like the 1986 event a cross section of taxa was impacted. The species responding to this event also stayed at their new distribution for more than two decades on average. Distributional events were compared with step changes in temperature and temperature differential time series. Bottom and surface temperatures have significantly increased on the order of 1°C over the time series in both spring and fall seasons (Figure 8 a, b, e, and f; Table 3 ). Loess residuals of these time series reveal several distinct temperature step changes in spring bottom and surface temperatures, specifically surface temperature in 1976 and bottom and surface temperatures in 2012 (Figure 8c and g). The only fall temperature change of note was in the bottom temperature during fall 1985 (Figure 8d and h). The fall 1985 bottom temperature change appeared potentially related to the distribution shift in the spring of 1986 given their one season of separation. The spring 1976 temperature change may be related to the fall 1977 distribution shift, but causality with a three-season lag seems unlikely. There are numerous distributional shifts beginning in spring of 2013 and in following years that may be related to the exceptional step change in temperature that occurred in 2012. Table 3. Theil–Sen slope and Mann–Kendall trend test probability (p, bold indicates significance at p=0.05) for time series of spring and fall bottom and surface temperature, temperature depth differential, temperature distance to the coast differential, and mean occupancy bottom temperature. Temperature Depth differential Distance to coast differential Group Season Slope p Slope p Slope p Bottom temperature Spring 0.017 0.039 0.001 0.776 0.001 0.803 Bottom temperature Fall 0.028 0.002 0.012 0.046 0.008 0.131 Surface temperature Spring 0.023 0.020 0.005 0.413 0.007 0.115 Surface temperature Fall 0.021 0.001 −0.003 0.464 −0.008 0.024 Occurrence bottom temperature Spring 0.006 0.413 Occurrence bottom temperature Fall 0.017 0.004 Temperature Depth differential Distance to coast differential Group Season Slope p Slope p Slope p Bottom temperature Spring 0.017 0.039 0.001 0.776 0.001 0.803 Bottom temperature Fall 0.028 0.002 0.012 0.046 0.008 0.131 Surface temperature Spring 0.023 0.020 0.005 0.413 0.007 0.115 Surface temperature Fall 0.021 0.001 −0.003 0.464 −0.008 0.024 Occurrence bottom temperature Spring 0.006 0.413 Occurrence bottom temperature Fall 0.017 0.004 Open in new tab Table 3. Theil–Sen slope and Mann–Kendall trend test probability (p, bold indicates significance at p=0.05) for time series of spring and fall bottom and surface temperature, temperature depth differential, temperature distance to the coast differential, and mean occupancy bottom temperature. Temperature Depth differential Distance to coast differential Group Season Slope p Slope p Slope p Bottom temperature Spring 0.017 0.039 0.001 0.776 0.001 0.803 Bottom temperature Fall 0.028 0.002 0.012 0.046 0.008 0.131 Surface temperature Spring 0.023 0.020 0.005 0.413 0.007 0.115 Surface temperature Fall 0.021 0.001 −0.003 0.464 −0.008 0.024 Occurrence bottom temperature Spring 0.006 0.413 Occurrence bottom temperature Fall 0.017 0.004 Temperature Depth differential Distance to coast differential Group Season Slope p Slope p Slope p Bottom temperature Spring 0.017 0.039 0.001 0.776 0.001 0.803 Bottom temperature Fall 0.028 0.002 0.012 0.046 0.008 0.131 Surface temperature Spring 0.023 0.020 0.005 0.413 0.007 0.115 Surface temperature Fall 0.021 0.001 −0.003 0.464 −0.008 0.024 Occurrence bottom temperature Spring 0.006 0.413 Occurrence bottom temperature Fall 0.017 0.004 Open in new tab Figure 8. Open in new tabDownload slide Annual spring (a) and fall (b) bottom temperatures for the Northeast Shelf ecosystem with residuals from their respective loess fits (c, d). Spring (e) and fall (f) surface temperature with residuals from loess fits (g) and (h), respectively. Red lines are loess fits. Figure 8. Open in new tabDownload slide Annual spring (a) and fall (b) bottom temperatures for the Northeast Shelf ecosystem with residuals from their respective loess fits (c, d). Spring (e) and fall (f) surface temperature with residuals from loess fits (g) and (h), respectively. Red lines are loess fits. Bottom and surface depth temperature differential time series were without significant trends (Supplementary Figure S1a, b, e, and f). There was a spring bottom temperature differential step change in 1975 that preceded the 1977 distribution event by 2.5 years (Supplementary Figure S1g); with such a long lag between thermal and distribution events it does not appear that these events are related. The other step changes in these data were unrelated to the 1977 or 1986 distributional events, but did include spring change in 2012 and a notable change in 1994 (Supplementary Figure S1c, d, and h). Bottom and surface distance to the coast temperature differential time series were significant in only one instance: the fall surface time series (Supplementary Figure S2a, b, e, and f). None of the step changes in these data were in proximity to the 1977 or 1986 distributional events or related to the recent events beginning in 2012 (Supplementary Figure S2c, d, g, and h). Occupancy bottom temperature Occupancy temperature for fish and macroinvertebrate species in spring had a differential response compared with the response for fall species. Occupancy temperature was without trend for the mean of all spring species (Figure 9 a), which is in contrast to the increasing trend in spring ecosystem temperatures (same data repeated from Figure 8a). The trend of the ecosystem temperature, as noted previously, was significant, whereas the trend for the mean occupancy temperature was non-significant (Table 3). The rate of change in occupancy temperature average about zero for pelagic piscivores and planktivores and were only slightly positive for benthic and demersal species (Figure 9c). This result would be consistent with fish and macroinvertebrates adjusting their distributions to maintain a constant thermal habitat over time despite changes in the distribution of thermal conditions in the ecosystem. In contrast, occupancy temperature for fish and macroinvertebrate species in fall changed over time, albeit not to the same extent as ecosystem temperatures (Figure 9b). The trends suggested by both time series were significant (Table 3). The excursion for ecosystem temperature was about 1.25°C as suggested by the start and end of the regression line, whereas the excursion of occupancy temperature was less, approximately 0.75°C. All fall functional groups had positive mean trends, exceeding the spring means for all groups (Figure 9d). The fall thermal regime has warmed faster than the spring regime; and, fall taxa have occupied progressively warmer water over time, thus not maintaining a consistent thermal habitat. Figure 9. Open in new tabDownload slide Annual spring (a) and fall (b) bottom temperatures for the Northeast Shelf ecosystem with mean occupancy temperature for fish and macroinvertebrates species. Lines are linear regression. Mean Theil–Sen slope of spring (c) and fall (d) occupancy temperature of fish and macroinvertebrates by benthivore, demersal piscivore, pelagic piscivore, and planktivore functional groups. Error bars are 95% confidence intervals. Figure 9. Open in new tabDownload slide Annual spring (a) and fall (b) bottom temperatures for the Northeast Shelf ecosystem with mean occupancy temperature for fish and macroinvertebrates species. Lines are linear regression. Mean Theil–Sen slope of spring (c) and fall (d) occupancy temperature of fish and macroinvertebrates by benthivore, demersal piscivore, pelagic piscivore, and planktivore functional groups. Error bars are 95% confidence intervals. Multivariate analysis Canonical correlation analysis suggests closer multivariate relationships between zooplankton and fish and macroinvertebrate distributions than with temperature variables. Fish and macroinvertebrate along-shelf distribution metrics were most clearly separated from zooplankton and temperature variables along canonical variate 2 (Figure 10 a and b). Proximity of variables suggest an association with the along-shelf distribution of copepods in spring and other zooplankton taxa in fall. The fish and macroinvertebrate depth differential metrics were separated along canonical variate 1 (Figure 10c and d). The spring variable separation suggests a weak correspondence to the temperature variables, where the fall data suggests a close relationship between fish and macroinvertebrate distribution and bottom and surface water temperature variables. The fish and macroinvertebrate distance to the coast metrics were clearly separated along canonical variate 1 in the spring (Figure 10e), suggesting weak correlation with the temperature variables and stronger correlation with the zooplankton data. There are few discernable patterns in the fall data (Figure 10f). Figure 10. Open in new tabDownload slide Canonical correlation analysis between fish and macroinvertebrate distribution responses (benthivores B, demersal piscivores DP, pelagic piscivores PP, and planktivores P), zooplankton (other zooplankton ASD Oasd, copepod ASD Casd, other zooplankton depth Odep, copepod depth Cdep, other zooplankton DTC Odtc, copepod DTC Cdtc), and temperature (bottom temperature BT, surface temperature ST, bottom depth temperature difference Bdep, surface temperature difference Sdep, bottom distance to coast temperature difference Bdis, surface distance to coast temperature difference Sdis) explanatory variables. Analyses are presented for spring along-shelf coast distance (a), depth (c), and distance to the coast (e), and fall along-shelf distance (b), depth (d), and distance to the coast (f). Figure 10. Open in new tabDownload slide Canonical correlation analysis between fish and macroinvertebrate distribution responses (benthivores B, demersal piscivores DP, pelagic piscivores PP, and planktivores P), zooplankton (other zooplankton ASD Oasd, copepod ASD Casd, other zooplankton depth Odep, copepod depth Cdep, other zooplankton DTC Odtc, copepod DTC Cdtc), and temperature (bottom temperature BT, surface temperature ST, bottom depth temperature difference Bdep, surface temperature difference Sdep, bottom distance to coast temperature difference Bdis, surface distance to coast temperature difference Sdis) explanatory variables. Analyses are presented for spring along-shelf coast distance (a), depth (c), and distance to the coast (e), and fall along-shelf distance (b), depth (d), and distance to the coast (f). Discussion Centre of gravities Aggregated fish and macroinvertebrate centre of gravities in the NEUS shifted poleward, and to an extent inshore through time, concurrently with warming water conditions and changes in secondary production. Similar distributional shifts have been reported for several species within the NEUS and linked to thermal conditions and climate oscillations (Nye et al., 2009; Lucey and Nye, 2010; Pinsky et al., 2013b; Kleisner et al., 2017). These results further support the operating hypothesis that a warming NEUS directly affects marine population distributions through their ecophysiology. Centre of gravity metrics, patterns may be weaker or stronger based on feeding strategies, with planktivores often exhibiting the strongest responses. For species living at the southern edge of their ecological range in the NEUS, distributional shifts may cause a contraction of the population and a functional extinction of the population within this region. For example, in the western Gulf of Maine, changes in the timing of the northern shrimp hatch have been linked to water temperatures, with hatch initiating earlier and lasting longer concomitant with increased bottom temperature (Richards, 2012). These earlier hatches are associated with decreased survival of larval stages and declining abundance of the population (Richards and Jacobson, 2016). Similar thermal mechanisms have been identified with a range of marine species with associated change in their population dynamics, including impacts on American lobster and Atlantic cod (Pershing et al., 2015; Rheuban et al., 2017; Le Bris et al., 2018). Centre of gravity for aggregated zooplankton metrics do not indicate any significant poleward shifts in response to increasing temperatures, with the only significant shift indicating a shallowing depth distribution during the fall. However, within the zooplankton community, differences emerge in the distribution response of taxonomic groups; spring and fall copepod centre of gravity estimates indicate poleward, shallower, and inshore movements, while non-copepods and other zooplankton indicated little or no change in distribution through time. Demarcation between copepods and other zooplankton may relate to differences in vertical distribution in the water column through their life, with copepods primarily pelagic and many of the non-copepod taxa analysed here having sessile life stages, and thus limited in their ability to move with habitat changes. Copepod shifts in centre of gravity align with those of fish and macroinvertebrates and may explain the seasonal differences in the occupancy temperature of fish and macroinvertebrates. Planktivores feeding on copepods seem to drive the overall nekton shift, further supporting the bottom-up effect of zooplankton (Suca et al., 2018). Similar shifts suggest that thermal environment influences copepods and fish and macroinvertebrate populations in some equivalent way; or, fish and macroinvertebrate distributional shifts may be due to more complex interactions beyond thermal tolerances, more specifically prey availability. Copepods are dominant prey for the early life history stages of many marine species, including Atlantic cod, haddock, redfish, and Atlantic mackerel (Runge, 1988; Castonguay et al., 2008). Changes in zooplankton species composition and abundance can determine the prey available for larval fish and the local recruitment and abundance of components of a meta-population (Friedland et al., 2013), particularly for species that have fish larvae with copepod-specific diets (Robert et al., 2008; Wilson et al., 2018). Copepod community structure shifts have been linked to regime change in the recruitment of a number of fish stocks (Perretti et al., 2017); more specific linkages have been described between Gulf of Maine and Georges Bank cod and haddock recruitment and change in climate and zooplankton (Mountain and Kane, 2010; Friedland et al., 2013). Our multivariate analysis suggests stronger concordance between copepods and specific fish and macroinvertebrate groups than with thermal covariates; however, the relationships are highly variable depending on the centre of gravity metric considered. The maintenance of a constant occupancy temperature by fish and macroinvertebrates in spring would suggest distributions were thermally driven whereas the change in fall occupancy temperature over time would suggest trophic interactions played a relatively larger role. It is worth noting that the rate of along-shelf movement of copepods was lower in the fall than the spring and the excursion of ecosystem temperatures was less in the spring than in the fall. It would appear that fall fish and macroinvertebrate distribution were at variance to thermal trends because of the less pronounced shift in distributions of copepods in that season. Phytoplankton centre of gravity metrics generally indicated little change in the distribution within the NEUS through time. Trophic influences between primary and secondary production may operate in the spring, but do not appear in the fall data. While phytoplankton biomass does not appear to have shifted through time, bloom phenology has changed, particularly with blooms initiating earlier and lasting longer (Friedland et al., 2015). Further, changes in bloom dynamics have in part been attributed to grazing effects (Friedland et al., 2015; 2016), indicating that the interaction reported between phytoplankton and zooplankton may be stationary or spatially variable through time. Bloom phenology changes attributed have been attributed to warming waters (Friedland et al., 2018); however, warming sea temperatures do not correspond to geographic shifts in the bloom centre of gravity. In addition to temperature, phytoplankton biomass and primary production rates are influenced by solar irradiance, mixed-layer depth, freshwater inputs, stratification, nutrient concentrations, and grazing pressure (Sverdrup, 1953; Friedland et al., 2015). In the Gulf of Maine and Georges Bank areas, physical forcing such as wind stress, heat flux, stratification, and source water entering the deep Gulf of Maine have been considered as or of greater importance to bloom formation than nitrate availability (Ji et al., 2008; Mountain and Kane, 2010; Saba et al., 2015). Our work supports the use of multiple metrics in identifying centre of gravity shifts since species and communities may respond to environmental changes differently. In an extensive analysis of adult fish and ichthyoplankton within the NEUS, Walsh et al. (2015) documented how geospatial shifts through time in marine fish are not necessarily uniform across species or life history stage. While centre of gravity shifts allow the assessment of population movements, climate change can affect marine fish distributions through other spatial effects, such as change in habitat areal extent or range modification. Adams (2017) highlighted how the methods used to assess spatial movements can influence perception of climate effects on the distribution of butterfish (Peprilus triacanthus) demonstrating differences in distribution when the data is disaggregated by age or size. Differential response to thermal conditions Our study suggests that lower trophic levels differentially respond to change in thermal regime compared with upper trophic levels, a difference potentially attributable to the distinct scales at which these organisms interact with their environment as modulated by order of magnitude differences in their vital rates. Whereas upper trophic level taxa integrate thermal conditions across broad spatial and temporal scales, lower trophic level taxa, composed of phytoplankton and zooplankton, tend to experience thermal conditions at much finer scales. When coupled with shorter turnover times associated with lower trophic levels, such taxa have the opportunity and ability to optimize the timing and microscale expression of their life histories at a given locale. In marine ecosystems, we have seen shifts in phenology of phytoplankton and zooplankton blooms (Friedland et al., 2015; Kristiansen et al., 2016). We have also seen shifts in fine-scale vertical location due to mixed-layer depth, stratification, nutrient concentrations, and predation pressure (Sverdrup, 1953; Friedland et al., 2015) and shifts in horizontal location due to wind stress, heat flux, freshwater inputs, and water mass movements (Ji et al., 2008; Mountain and Kane, 2010; Saba et al., 2015). Metabolic vital rate considerations, expressed as nutrient uptake for phytoplankton or grazing for zooplankton, mediate these shifts. For phytoplankton or zooplankton, given the speed of their vital rates that accumulate into what are relatively short production and generation times, the ability to wait until “conditions are right” (Tilman, 1981) is a reasonable outcome and likely why large-scale shifts in biomass are generally not widely observed for these taxa. The exception might be when current regimes change and thereby alter both the location of nutrients and phytoplankton cells at large spatial scales (Polovina et al., 2008). Furthermore, we need to be mindful that chlorophyll concentration is indicative of community level abundance and does not address the potential for species level change in the phytoplankton community that may be responding differentially to climate forcing (Winder and Sommer, 2012). Lower trophic levels would appear to have a higher capacity to adapt to thermal change, whereas upper trophic levels with more complex life histories, and relatively slower vital rates, cannot afford to wait for the “right conditions” and hence move to more suitable habitat—expressed as shifts in distribution. Ambient temperature affects all poikilotherms, but these effects are moderated by the organism’s surface area relative to volume exposed to ambient temperatures (Froese, 2006). Lower trophic level organisms have a much higher ratio of surface area to volume, and thus are more directly susceptible to the influence of thermal conditions on their vital rates than larger organisms with a lower surface area to volume ratio (Planinsic and Vollmer, 2008). As such, lower trophic level taxa respond rapidly to environmental conditions and are able to make up for suboptimal conditions by taking advantage of thermal conditions that shift to suitable temperatures. Conversely, upper trophic level taxa are relatively less adaptable, having a lower range of possible vital rate changes, which results in lower metabolic rates and ultimately lower population-level production. When viewed at a population level at broad spatial scales, the lower metabolic rates and lowered production of upper trophic level taxa would lead to lower realized biomass at a given locale, which coupled with their need to move to suitable habitats, is expressed as a shift in distribution. Another consideration that would lead to the differences observed in lower and upper trophic level shift in location is the ability of the former to enter dormant states. Phytoplankton can enter various cysts or similar resting stages that then respond to environment cues (such as temperature) that signal the activation of regular metabolism (Ellegaard and Ribeiro, 2018). Some zooplankton have diapausing capabilities (Baumgartner and Tarrant, 2017) that provide a similar function. Upper trophic level taxa, especially fishes, do not have dormant stages. Similar to the logic noted above, these dormant stages allow lower trophic level taxa to await suitable thermal conditions, and hence would resist the need to shift distribution. The instance of copepods exhibiting a weak, but discernable distribution shift represents an intermediary example between the patterns shown here for upper and lower trophic levels. Although true for many zooplankton, certain key facets of copepod life histories can more closely resemble, in both mechanism and magnitude, dynamics of upper trophic level taxa. These include the production of egg sac broods, the large potential migration distances relative to body size, relatively lower productivity and fecundity compared with other zooplankton taxa, and a lack of vegetative reproduction (Purcell, 2018). Additionally, copepod hatching is often mismatched to production cycles because of thermal conditions (Baumgartner and Tarrant, 2017) and copepods have the capacity to store energy to a greater extent than most other zooplankton (Jager et al., 2017). Collectively these facets of copepods life history likely contribute to the distributional behaviour we observed. Change points The greater number of distributional change points observed in recent years suggests greater system variability, and perhaps instability, than at other points in the time series. However, these recent shifts should be assessed with caution and reevaluated as data becomes available. The distinct shift seen in spring 2013 fish and macroinvertebrate along-shelf centre of gravity followed the anomalously warm year of 2012. This 2012 warming was pervasive throughout the NEUS, and most pronounced in the Gulf of Maine. The temperatures have been associated with the rare occurrence of longfin squid in the region and accelerated moulting of American lobsters that ultimately caused more legal-sized lobsters to recruit to the fishery and a lengthening of the fishing season (Mills et al., 2013). In the case of the 2012 warming, effects were noticed within the same year; however, change points in fish and macroinvertebrate centre of gravities may be lagged responses from environmental or bottom-up changes in years prior. For example, altered season lengths can influence fish growth, maturity, and reproduction in a given year, with the resulting biology apparent years later in recruitment and spawning stock abundance (Henderson et al., 2017). While this work analyses the NEUS in totality to capture holistic changes within the region, the heterogeneity between the NEUS subunits should be noted. Ecoregions within the NEUS have been shown to vary based on temperature (Thomas et al., 2017), oceanography (Townsend et al., 2006), phytoplankton blooms (Friedland et al., 2015), zooplankton (Morse et al., 2017), and fish assemblage (Lucey and Nye, 2010). Ecoregions with the NEUS may show differences in both central shifts and change points through time. Within the NEUS, spring and fall zooplankton community regime shifts have been documented between the years 1989 and 2006 (Morse et al. 2017). The fish and macroinvertebrate centre of gravity shifts were largely found outside of this period, but those species with significant change points in their centre of gravities in 1977 and 1986 tended to show a persistent shift in distribution. Zooplankton community regime shifts have been linked predominantly to changes in temperature, stratification, and climate oscillations (Morse et al., 2017), indicating that these forces may be greater determinants in zooplankton shifts in centre of gravity than the lower trophic level production. These regime shifts have also been presented as size shifts in the zooplankton community for ecoregions within the NEUS. The ratio of small (e.g. Pseudocalanus sp.) to large (e.g. Calanus finmarchicus) copepods have gone through three dominant phases: low ratio in the 1970s and 1980s, followed by a high ratio from the 1980s through 1990s, and a return to the smaller-sized regime in the early 2000s (Perretti et al., 2017). These three copepod regimes correspond to low-high-low regimes of fish recruitment for several stocks in the NEUS (Perretti et al., 2017), suggesting specifically the bottom-up effect of smaller-sized copepods on fish recruitment, and corroborating the synchronous centre of gravity shifts for copepods and higher trophic level nekton and the potential for a persistent shift (Rindorf and Lewy, 2006). As with fish and temperature effects, zooplankton have also exhibited lagged relations with climate. For example, sea temperatures and Calanus finmarchicus have been shown to relate to basin-scale oceanographic changes from climate oscillations by up to 4 years (Xu et al., 2015; Thomas et al., 2017). Conclusions This work highlights the importance of both the availability of lower trophic level organisms and the physical environment in shaping higher trophic level distributional patterns over mesoscale to macroscale of an ecosystem. Movement of fish and macroinvertebrate taxa northeast and inshore seems to correspond with both warming waters and shifts in copepod populations. Dissimilar trends across multiple trophic levels further highlights how ecosystem responses can be non-linear or different across trophic levels, with trophic levels impacted differently by the environment. The phytoplankton standing stock does not exhibit similar shifts, but can vary in its centre of gravity annually and is likely still a prominent driver in higher tropic level production. The incongruous shifts between phytoplankton and fish and macroinvertebrates represent the differences in physiological reliance to temperature, or the suite of factors that differentially influence each trophic level. While sea temperature and prey fields are primary habitat components, several other habitat determinants not reviewed here—including salinity, sediment, dissolved oxygen, pH, and ocean circulation—have also changed through time (Poloczanska et al., 2016) and likely contribute to distributional variability. Further, as highlighted by copepod and larger nekton interactions, habitat requirements can vary by life stage. Thus, changes at a given life stage can alter population connectivity between successive life stages, overall recruitment, and spawning stock biomass (Cowen et al., 2007; Llopiz et al., 2014). Finally, regardless of which specific mechanism, or cumulative effects thereof, the scales of interaction with the environment, vital rates, and life history traits differentially result in lower likelihood for lower trophic levels to exhibit large-scale distribution shifts in biomass relative to upper trophic level fish and macroinvertebrates. Acknowledgements We thank B. Smith and S. Lucey for input on fish and macroinvertebrate data and S. Jennings and S. Large for comments on an early draft. We also thank three anonymous reviewers for productive comments. References Adams C. F. 2017 . Age-specific differences in the seasonal spatial distribution of butterfish (Peprilus triacanthus . ICES Journal of Marine Science , 74 : 170 – 179 . Google Scholar Crossref Search ADS WorldCat Alheit J. , Niquen M. 2004 . Regime shifts in the Humboldt Current ecosystem . Progress in Oceanography , 60 : 201 – 222 . Google Scholar Crossref Search ADS WorldCat Anderson J. J. , Gurarie E. , Bracis C. , Burke B. J. , Laidre K. L. 2013 . Modeling climate change impacts on phenology and population dynamics of migratory marine species . Ecological Modelling , 264 : 83 – 97 . Google Scholar Crossref Search ADS WorldCat Anderson J. T. 1988 . A review of size-dependent survival during pre-recruit stages of fishes in relation to recruitment . Journal of Northwest Atlantic Fishery Science , 8 : 55 – 66 . Google Scholar Crossref Search ADS WorldCat Baumgartner M. F. , Tarrant A. M. 2017 . The physiology and ecology of diapause in marine copepods . Annual Review of Marine Sciences , 9 : 387 – 411 . Google Scholar Crossref Search ADS WorldCat Beardsley R. C. , Smith P. C. , Lee C. M. 2003 . Introduction to special section: U.S. GLOBEC: physical processes on Georges Bank (GLOBEC) . Journal of Geophysical Research-Oceans , 108 : 1 – 3 . WorldCat Beaugrand G. , Brander K. M. , Lindley J. A. , Souissi S. , Reid P. C. 2003 . Plankton effect on cod recruitment in the North Sea . Nature , 426 : 661 – 664 . Google Scholar Crossref Search ADS PubMed WorldCat Belkin I. M. 2009 . Rapid warming of large marine ecosystems . Progress in Oceanography , 81 : 207 – 213 . Google Scholar Crossref Search ADS WorldCat Bell R. J. , Richardson D. E. , Hare J. A. , Lynch P. D. , Fratantoni P. S. 2015 . Disentangling the effects of climate, abundance, and size on the distribution of marine fish: an example based on four stocks from the Northeast US Shelf . ICES Journal of Marine Science , 72 : 1311 – 1322 . Google Scholar Crossref Search ADS WorldCat Bi H. S. , Peterson W. T. , Lamb J. , Casillas E. 2011 . Copepods and salmon: characterizing the spatial distribution of juvenile salmon along the Washington and Oregon coast, USA . Fisheries Oceanography , 20 : 125 – 138 . Google Scholar Crossref Search ADS WorldCat Brett J. R. 1979 . Environmental factors and growth . In Fish Physiolog , pp. 599 – 675 . Ed. by Hoar W. S. , Randall D. J. , Brett J. R. Academic Press , London . Google Preview WorldCat COPAC Canales T. M. , Law R. , Blanchard J. L. 2016 . Shifts in plankton size spectra modulate growth and coexistence of anchovy and sardine in upwelling systems . Canadian Journal of Fisheries and Aquatic Sciences , 73 : 611 – 621 . Google Scholar Crossref Search ADS WorldCat Castonguay M. , Plourde S. , Robert D. , Runge J. A. , Fortier L. 2008 . Copepod production drives recruitment in a marine fish . Canadian Journal of Fisheries and Aquatic Sciences , 65 : 1528 – 1531 . Google Scholar Crossref Search ADS WorldCat Cowen R. K. , Gawarkiewic G. , Pineda J. , Thorrold S. R. , Werner F. E. 2007 . Population connectivity in marine systems: an overview . Oceanography , 20 : 14 – 21 . Google Scholar Crossref Search ADS WorldCat Ellegaard M. , Ribeiro S. 2018 . The long-term persistence of phytoplankton resting stages in aquatic ‘seed banks’ . Biological Reviews , 93 : 166 – 183 . Google Scholar Crossref Search ADS PubMed WorldCat Friedland K. D. , Kane J. , Hare J. A. , Lough R. G. , Fratantoni P. S. , Fogarty M. J. , Nye J. A. 2013 . Thermal habitat constraints on zooplankton species associated with Atlantic cod (Gadus morhua) on the US Northeast Continental Shelf . Progress in Oceanography , 116 : 1 – 13 . Google Scholar Crossref Search ADS WorldCat Friedland K. D. , Leaf R. T. , Kane J. , Tommasi D. , Asch R. G. , Rebuck N. , Ji R. 2015 . Spring bloom dynamics and zooplankton biomass response on the US Northeast Continental Shelf . Continental Shelf Research , 102 : 47 – 61 . Google Scholar Crossref Search ADS WorldCat Friedland K. D. , Mouw C. B. , Asch R. G. , Ferreira A. S. A. , Henson S. , Hyde K. J. W. , Morse R. E. , et al. . 2018 . Phenology and time series trends of the dominant seasonal phytoplankton bloom across global scales . Global Ecology and Biogeography , 27 : 551 – 569 . Google Preview WorldCat COPAC Friedland K. D. , Record N. R. , Asch R. G. , Kristiansen T. , Saba V. S. , Drinkwater K. F. , Henson S. , et al. . 2016 . Seasonal phytoplankton blooms in the North Atlantic linked to the overwintering strategies of copepods . Elementa , 99 : 1 – 19 . Friedland K. D. , Stock C. , Drinkwater K. F. , Link J. S. , Leaf R. T. , Shank B. V. , Rose J. M. , et al. . 2012 . Pathways between primary production and fisheries yields of large marine ecosystems . PLoS One , 7 : 1 – 11 . Google Scholar Crossref Search ADS WorldCat Froese R. 2006 . Cube law, condition factor and weight-length relationships: history, meta-analysis and recommendations . Journal of Applied Ichthyology , 22 : 241 – 253 . Google Scholar Crossref Search ADS WorldCat Gregg W. W. , Conkright M. E. 2002 . Decadal changes in global ocean chlorophyll . Geophysical Research Letters , 29 : 1 – 4 . Google Scholar Crossref Search ADS WorldCat Grosslein M. D. 1969 . Groundfish survey program of BCF woods hole . Commercial Fisheries Review , 31 : 22 – 35 . WorldCat Hare J. A. , Manderson J. P. , Nye J. A. , Alexander M. A. , Auster P. J. , Borggaard D. L. , Capotondi A. M. , et al. . 2012 . Cusk (Brosme brosme) and climate change: assessing the threat to a candidate marine fish species under the US Endangered Species Act . Ices Journal of Marine Science , 69 : 1753 – 1768 . Google Scholar Crossref Search ADS WorldCat Hare J. A. , Morrison W. E. , Nelson M. W. , Stachura M. M. , Teeters E. J. , Griffis R. B. , Alexander M. A. , et al. . 2016 . A vulnerability assessment of fish and invertebrates to climate change on the Northeast US Continental Shelf . PLoS One , 11 : 1 – 30 . Google Scholar Crossref Search ADS WorldCat Henderson M. E. , Mills K. E. , Thomas A. C. , Pershing A. J. , Nye J. A. 2017 . Effects of spring onset and summer duration on fish species distribution and biomass along the Northeast United States Continental Shelf . Reviews in Fish Biology and Fisheries , 27 : 411 – 424 . Google Scholar Crossref Search ADS WorldCat Hipfner J. M. 2008 . Matches and mismatches: ocean climate, prey phenology and breeding success in a zooplanktivorous seabird . Marine Ecology Progress Series , 368 : 295 – 304 . Google Scholar Crossref Search ADS WorldCat Hunt G. L. , Coyle K. O. , Eisner L. B. , Farley E. V. , Heintz R. A. , Mueter F. , Napp J. M. , et al. . 2011 . Climate impacts on eastern Bering Sea foodwebs: a synthesis of new data and an assessment of the oscillating control hypothesis . ICES Journal of Marine Science , 68 : 1230 – 1243 . Google Scholar Crossref Search ADS WorldCat Jager T. , Salaberria I. , Altin D. , Nordtug T. , Hansen B. H. 2017 . Modelling the dynamics of growth, development and lipid storage in the marine copepod Calanus finmarchicus . Marine Biology , 164 : 1 – 15 . Google Scholar Crossref Search ADS PubMed WorldCat Ji R. B. , Davis C. S. , Chen C. S. , Townsend D. W. , Mountain D. G. , Beardsley R. C. 2008 . Modeling the influence of low-salinity water inflow on winter-spring phytoplankton dynamics in the Nova Scotian Shelf-Gulf of Maine region . Journal of Plankton Research , 30 : 1399 – 1416 . Google Scholar Crossref Search ADS WorldCat Kane J. 2007 . Zooplankton abundance trends on Georges Bank, 1977–2004 . ICES Journal of Marine Science , 64 : 909 – 919 . Google Scholar Crossref Search ADS WorldCat Kleisner K. M. , Coll M. , Lynam C. P. , Bundy A. , Shannon L. , Shin Y. J. , Boldt J. L. , et al. . 2015 . Evaluating changes in marine communities that provide ecosystem services through comparative assessments of community indicators . Ecosystem Services , 16 : 413 – 429 . Google Scholar Crossref Search ADS WorldCat Kleisner K. M. , Fogarty M. J. , McGee S. , Barnette A. , Fratantoni P. , Greene J. , Hare J. A. , et al. . 2016 . The effects of sub-regional climate velocity on the distribution and spatial extent of marine species assemblages . PLoS One , 11 : 1 – 21 . Google Scholar Crossref Search ADS WorldCat Kleisner K. M. , Fogarty M. J. , McGee S. , Hare J. A. , Moret S. , Perretti C. T. , Saba V. S. 2017 . Marine species distribution shifts on the US Northeast Continental Shelf under continued ocean warming . Progress in Oceanography , 153 : 24 – 36 . Google Scholar Crossref Search ADS WorldCat Kristiansen I. , Gaard E. , Hatun H. , Jonasdottir S. , Ferreira A. S. A. 2016 . Persistent shift of Calanus spp. in the southwestern Norwegian Sea since 2003, linked to ocean climate . ICES Journal of Marine Science , 73 : 1319 – 1329 . Google Scholar Crossref Search ADS WorldCat Lavaniegos B. E. , Ohman M. D. 2007 . Coherence of long-term variations of zooplankton in two sectors of the California Current System . Progress in Oceanography , 75 : 42 – 69 . Google Scholar Crossref Search ADS WorldCat Le Bris A. , Mills K. E. , Wahle R. A. , Chen Y. , Alexander M. A. , Allyn A. J. , Schuetz J. G. , et al. . 2018 . Climate vulnerability and resilience in the most valuable North American Fishery . Proceedings of the National Academy of Sciences of the United States of America , 115 : 1831 – 1836 . Link J. S. , Fulton E. A. , Gamble R. J. 2010 . The northeast US application of ATLANTIS: a full system model exploring marine ecosystem dynamics in a living marine resource management context . Progress in Oceanography , 87 : 214 – 234 . Google Scholar Crossref Search ADS WorldCat Llopiz J. K. , Cowen R. K. , Hauff M. J. , Ji R. B. , Munday P. L. , Muhling B. A. , Peck M. A. , et al. . 2014 . Early life history and fisheries oceanography new questions in a changing world . Oceanography , 27 : 26 – 41 . Google Scholar Crossref Search ADS WorldCat Lucey S. M. , Nye J. A. 2010 . Shifting species assemblages in the Northeast US Continental Shelf Large Marine Ecosystem . Marine Ecology Progress Series , 415 : 23 – 33 . Google Scholar Crossref Search ADS WorldCat Lynch P. D. , Nye J. A. , Hare J. A. , Stock C. A. , Alexander M. A. , Scott J. D. , Curti K. L. , et al. . 2015 . Projected ocean warming creates a conservation challenge for river herring populations . ICES Journal of Marine Science , 72 : 374 – 387 . Google Scholar Crossref Search ADS WorldCat Mackas D. L. , Batten S. , Trudel M. 2007 . Effects on zooplankton of a warmer ocean: recent evidence from the Northeast Pacific . Progress in Oceanography , 75 : 223 – 252 . Google Scholar Crossref Search ADS WorldCat Mann H. B. 1945 . Nonparametric tests against trend . Econometrica , 13 : 245 – 259 . Google Scholar Crossref Search ADS WorldCat Maritorena S. , d’Andon O. H. F. , Mangin A. , Siegel D. A. 2010 . Merged satellite ocean color data products using a bio-optical model: characteristics, benefits and issues . Remote Sensing of Environment , 114 : 1791 – 1804 . Google Scholar Crossref Search ADS WorldCat McManus M. C. , Hare J. A. , Richardson D. E. , Collie J. S. 2018 . Tracking shifts in Atlantic mackerel (Scomber scombrus) larval habitat suitability on the Northeast US Continental Shelf . Fisheries Oceanography , 27 : 49 – 62 . Google Scholar Crossref Search ADS WorldCat Mills K. E. , Pershing A. J. , Brown C. J. , Chen Y. , Chiang F. S. , Holland D. S. , Lehuta S. , et al. . 2013 . Fisheries management in a changing climate lessons from the 2012 ocean heat wave in the Northwest Atlantic . Oceanography , 26 : 191 – 195 . Google Scholar Crossref Search ADS WorldCat Morse R. E. , Friedland K. D. , Tommasi D. , Stock C. , Nye J. 2017 . Distinct zooplankton regime shift patterns across ecoregions of the US Northeast Continental Shelf Large Marine Ecosystem . Journal of Marine Systems , 165 : 77 – 91 . Google Scholar Crossref Search ADS WorldCat Mountain D. G. , Kane J. 2010 . Major changes in the Georges Bank ecosystem, 1980s to the 1990s . Marine Ecology Progress Series , 398 : 81 – 91 . Google Scholar Crossref Search ADS WorldCat Neill W. H. , Miller J. M. , Van Der Veer H. W. , Winemiller K. O. 1994 . Ecophysiology of marine fish recruitment - a conceptual-framework for understanding interannual variability . Netherlands Journal of Sea Research , 32 : 135 – 152 . Google Scholar Crossref Search ADS WorldCat Nye J. A. , Link J. S. , Hare J. A. , Overholtz W. J. 2009 . Changing spatial distribution of fish stocks in relation to climate and population size on the Northeast United States Continental Shelf . Marine Ecology Progress Series , 393 : 111 – 129 . Google Scholar Crossref Search ADS WorldCat Pepin P. 1991 . Effect of temperature and size on development, mortality, and survival rates of the pelagic early life-history stages of marine fish . Canadian Journal of Fisheries and Aquatic Sciences , 48 : 503 – 518 . Google Scholar Crossref Search ADS WorldCat Perretti C. T. , Fogarty M. J. , Friedland K. D. , Hare J. A. , Lucey S. M. , McBride R. S. , Miller T. J. , et al. . 2017 . Regime shifts in fish recruitment on the Northeast US Continental Shelf . Marine Ecology Progress Series , 574 : 1 – 11 . Google Scholar Crossref Search ADS WorldCat Pershing A. J. , Alexander M. A. , Hernandez C. M. , Kerr L. A. , Le Bris A. , Mills K. E. , Nye J. A. , et al. . 2015 . Slow adaptation in the face of rapid warming leads to collapse of the Gulf of Maine cod fishery . Science , 350 : 809 – 812 . Google Scholar Crossref Search ADS PubMed WorldCat Pinsky M. L. , Fogarty M. 2012 . Lagged social-ecological responses to climate and range shifts in fisheries . Climatic Change , 115 : 883 – 891 . Google Scholar Crossref Search ADS WorldCat Pinsky M. L. , Guannel G. , Arkema K. K. 2013a . Quantifying wave attenuation to inform coastal habitat conservation . Ecosphere , 4 : 1 – 16 . Google Scholar Crossref Search ADS WorldCat Pinsky M. L. , Worm B. , Fogarty M. J. , Sarmiento J. L. , Levin S. A. 2013b . Marine taxa track local climate velocities . Science , 341 : 1239 – 1242 . Google Scholar Crossref Search ADS WorldCat Planinsic G. , Vollmer M. 2008 . The surface-to-volume ratio in thermal physics: from cheese cube physics to animal metabolism . European Journal of Physics , 29 : 369 – 384 . Google Scholar Crossref Search ADS WorldCat Poloczanska E. S. , Burrows M. T. , Brown C. J. , Molinos J. G. , Halpern B. S. , Hoegh-Guldberg O. , Kappel C. V. , et al. . 2016 . Responses of marine organisms to climate change across oceans . Frontiers in Marine Science , 3 : 1 – 21 . Google Scholar Crossref Search ADS WorldCat Polovina J. J. , Howell E. A. , Abecassis M. 2008 . Ocean’s least productive waters are expanding . Geophysical Research Letters , 35 : 1 – 5 . Google Scholar Crossref Search ADS WorldCat Purcell J. E. 2018 . Successes and challenges in jellyfish ecology: examples from Aequorea spp . Marine Ecology Progress Series , 591 : 7 – 27 . Google Scholar Crossref Search ADS WorldCat Rheuban J. E. , Kavanaugh M. T. , Doney S. C. 2017 . Implications of future northwest Atlantic bottom temperatures on the American Lobster (Homarus americanus) Fishery . Journal of Geophysical Research-Oceans , 122 : 9387 – 9398 . Google Scholar Crossref Search ADS WorldCat Richards R. A. 2012 . Phenological shifts in hatch timing of northern shrimp Pandalus borealis . Marine Ecology Progress Series , 456 : 149 – 158 . Google Scholar Crossref Search ADS WorldCat Richards R. A. , Jacobson L. D. 2016 . A simple predation pressure index for modeling changes in natural mortality: application to Gulf of Maine northern shrimp stock assessment . Fisheries Research , 179 : 224 – 236 . Google Scholar Crossref Search ADS WorldCat Rindorf A. , Lewy P. 2006 . Warm, windy winters drive cod north and homing of spawners keeps them there . Journal of Applied Ecology , 43 : 445 – 453 . Google Scholar Crossref Search ADS WorldCat Robert D. , Castonguay M. , Fortier L. 2008 . Effects of intra- and inter-annual variability in prey field on the feeding selectivity of larval Atlantic mackerel (Scomber scombrus) . Journal of Plankton Research , 30 : 673 – 688 . Google Scholar Crossref Search ADS WorldCat Rodionov S. N. 2004 . A sequential algorithm for testing climate regime shifts . Geophysical Research Letters , 31 : 1 – 4 . Google Scholar Crossref Search ADS WorldCat Rodionov S. N. 2006 . Use of prewhitening in climate regime shift detection . Geophysical Research Letters , 33 : 1 – 4 . Google Scholar Crossref Search ADS WorldCat Runge J. A. 1988 . Should we expect a relationship between primary production and fisheries - the role of copepod dynamics as a filter of trophic variability . Hydrobiologia , 167–168 : 61 – 71 . Google Scholar Crossref Search ADS WorldCat Ryther J. H. 1969 . Photosynthesis and fish production in the sea . Science , 166 : 72 – 76 . Google Scholar Crossref Search ADS PubMed WorldCat Saba V. S. , Griffies S. M. , Anderson W. G. , Winton M. , Alexander M. A. , Delworth T. L. , Hare J. A. , et al. . 2016 . Enhanced warming of the Northwest Atlantic Ocean under climate change . Journal of Geophysical Research-Oceans , 121 : 118 – 132 . Google Scholar Crossref Search ADS WorldCat Saba V. S. , Hyde K. J. W. , Rebuck N. D. , Friedland K. D. , Hare J. A. , Kahru M. , Fogarty M. J. 2015 . Physical associations to spring phytoplankton biomass interannual variability in the US Northeast Continental Shelf . Journal of Geophysical Research-Biogeosciences , 120 : 205 – 220 . Google Scholar Crossref Search ADS WorldCat Sherman K. , Solow A. , Jossi J. , Kane J. 1998 . Biodiversity and abundance of the zooplankton of the Northeast Shelf ecosystem . ICES Journal of Marine Science , 55 : 730 – 738 . Google Scholar Crossref Search ADS WorldCat Suca J. J. , Pringle J. W. , Knorek Z. R. , Hamilton S. L. , Richardson D. E. , Llopiz J. K. 2018 . Feeding dynamics of Northwest Atlantic small pelagic fishes . Progress in Oceanography , 165 : 52 – 62 . Google Scholar Crossref Search ADS WorldCat Sverdrup H. U. 1953 . On conditions for the vernal blooming of phytoplankton . ICES Journal of Marine Science , 18 : 287 – 295 . Google Scholar Crossref Search ADS WorldCat Tanasichuk R. W. , Routledge R. 2011 . An investigation of the biological basis of return variability for sockeye salmon (Oncorhynchus nerka) from Great Central and Sproat lakes, Vancouver Island . Fisheries Oceanography , 20 : 462 – 478 . Google Scholar Crossref Search ADS WorldCat Thomas A. C. , Pershing A. J. , Friedland K. D. , Nye J. A. , Mills K. E. , Alexander M. A. , Record N. R. , et al. . 2017 . Seasonal trends and phenology shifts in sea surface temperature on the North American Northeastern Continental Shelf . Elementa: Science of the Anthropocene , 5 : 1 – 17 . WorldCat Thorson J. T. , Ianelli J. N. , Larsen E. A. , Ries L. , Scheuerell M. D. , Szuwalski C. , Zipkin E. F. 2016 . Joint dynamic species distribution models: a tool for community ordination and spatio-temporal monitoring . Global Ecology and Biogeography , 25 : 1144 – 1158 . Google Scholar Crossref Search ADS WorldCat Tilman D. 1981 . Tests of resource competition theory using 4 species of Lake-Michigan Algae . Ecology , 62 : 802 – 815 . Google Scholar Crossref Search ADS WorldCat Townsend D. W. , Thomas A. C. , Mayer L. M. , Thomas M. , Quinlan J. 2006 . Oceanography of the Northwest Atlantic Continental Shelf . In The Sea , pp. 119 – 168 . Ed. by Robinson A. R. , Brink K. H. Harvard University Press , Cambridge . Google Preview WorldCat COPAC Walsh H. J. , Richardson D. E. , Marancik K. E. , Hare J. A. 2015 . Long-term changes in the distributions of larval and adult fish in the Northeast US Shelf Ecosystem . PLoS One , 10 : 1 – 31 . WorldCat Wilson S. K. , Depcyznski M. , Fisher R. , Holmes T. H. , Noble M. M. , Radford B. T. , Rule M. , et al. . 2018 . Climatic forcing and larval dispersal capabilities shape the replenishment of fishes and their habitat-forming biota on a tropical coral reef . Ecology and Evolution , 8 : 1918 – 1928 . Google Scholar Crossref Search ADS PubMed WorldCat Winder M. , Sommer U. 2012 . Phytoplankton response to a changing climate . Hydrobiologia , 698 : 5 – 16 . Google Scholar Crossref Search ADS WorldCat Woillez M. , Poulard J. C. , Rivoirard J. , Petitgas P. , Bez N. 2007 . Indices for capturing spatial patterns and their evolution in time, with application to European hake (Merluccius merluccius) in the Bay of Biscay . Ices Journal of Marine Science , 64 : 537 – 550 . Google Scholar Crossref Search ADS WorldCat Xu H. K. , Kim H. M. , Nye J. A. , Hameed S. 2015 . Impacts of the North Atlantic Oscillation on sea surface temperature on the Northeast US Continental Shelf . Continental Shelf Research , 105 : 60 – 66 . Google Scholar Crossref Search ADS WorldCat Published by International Council for the Exploration of the Sea 2018. This work is written by US Government employees and is in the public domain in the US. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
Effects of ocean acidification on the respiration and feeding of juvenile red and blue king crabs (Paralithodes camtschaticus and P. platypus)Long, William, Christopher;Pruisner,, Paige;Swiney, Katherine, M;Foy, Robert, J
doi: 10.1093/icesjms/fsz090pmid: N/A
Abstract Ocean acidification is a decrease in pH resulting from dissolution of anthropogenic CO2 in the oceans that has physiological effects on many marine organisms. Juvenile red and blue king crabs (Paralithodes camtschaticus and P. platypus) exhibit both increased mortality and decreased growth in acidified waters. In this study, we determined how ocean acidification affects oxygen consumption, feeding rates, and growth in both species. Juvenile crab were exposed to three pH levels: ambient (pH 8.1), pH 7.8, and pH 7.5 for 3 weeks. Oxygen consumption and feeding ration were determined immediately after exposure to treatment water and after 3 weeks’ exposure. Growth was calculated as a change in wet mass. Both species exhibited initially increased oxygen consumption at pH 7.5, but not after 3 weeks. Feeding rations did not vary with pH or exposure time. Red king crabs that moulted grew more in ambient water than in pH 7.5. The initial increase in oxygen consumption at pH 7.5 suggests the crab increased metabolism and expended more energy in osmo-/iono-regulation. Without an increase in feeding ration, it is likely the crab reduced energy expenditure in other areas, explaining the reduced growth and increased mortality observed in this and other studies. Introduction The dissolution of anthropogenic CO2 into the world’s oceans is causing a phenomenon called ocean acidification (Feely et al., 2004). The average global pH of surface waters has decreased 0.1 pH units over the past century, and the rate of acidification is projected to accelerate with increasing CO2 emissions (Caldeira and Wickett, 2003; Fabry et al., 2008). High latitude waters, in which CO2 is more soluble due to colder temperatures, are likely to acidify faster than the global average (Fabry et al., 2009). As a result, surface waters of the Bering Sea are projected to be undersaturated with respect to aragonite within 60 years (Mathis et al., 2015). Bottom waters in the Bering Sea shelf have already seasonally high pCO2 levels (1600 µatm) due to increases in respiration (Mathis et al., 2014). These projected changes in water chemistry can have physiological effects on marine organisms, although the effects are highly variable, both within and among taxa (Kroeker et al., 2010). Crustaceans comprise a diverse group of calcifying marine organisms that can be vulnerable to ocean acidification. The crab exoskeleton comprised calcium carbonate (primarily calcite and amorphous calcium carbonate) embedded in a chitin matrix (Raabe et al., 2006; Boßelmann et al., 2007; Chen et al., 2008). Although their internal calcification process and organic-layer shielded calcified structures may protect them against direct dissolution (Whiteley, 2011), ocean acidification can still cause physiological stress, affecting macro-scale responses including growth and survival (Keppel et al., 2012; Baragi and Anil, 2015). Most decapod crustaceans buffer their haemolymph in response to ocean acidification (Meseck et al., 2016) primarily via active bicarbonate transport at the gills (Appelhans et al., 2012). This can prevent or reverse acidosis, but it is not always fully effective (Zittier et al., 2013), and some crustaceans may not use this mechanism at all (Pane and Barry, 2007). In addition, this mechanism is energetically expensive, and may divert energy away from other vital biological processes, including growth and reproduction (Whiteley, 2011; Swiney et al., 2016). As many crustaceans are important commercial species, there are concerns that ocean acidification could result in reduced populations and fishery yields (Punt et al., 2014,, 2016). Red and blue king crabs, Paralithodes camtschaticus and P. platypus, are two important fisheries species in Alaska waters which have a similar life-history and overlapping range, though blue king crab may be more cold-tolerant than red king crab (Somerton, 1985). Both species release larvae in the late spring and early summer after brooding the embryos for about a year (Jensen and Armstrong, 1989; Stevens and Swiney, 2007). The larvae are planktonic in the water column so that their pH exposure can vary diurnally (due to diel vertical migration) and seasonally (Wainwright et al., 1991; Cross et al., 2013). The larvae pass through four zoeal stages before moulting to the glaucothoe or settling stage (Shirley and Shirley, 1989; Stevens et al., 2008). The glaucothoe settle in complex habitats (Stevens, 2003; Tapella et al., 2009), usually in nearshore areas (McMurray et al., 1984; Armstrong et al., 1987), where ocean pH is more stable than the upper water column or deeper benthic areas but may be affected by seasonal upwelling, where they use cryptic behaviour in the available complex habitat (Daly and Long, 2014b) to avoid predation (Daly et al., 2012) and cannibalism (Long and Whitefleet-Smith, 2013; Daly and Long, 2014a). For both species, distribution generally shifts to deeper, soft-bottomed habitats, where pH can be seasonally undersaturated with respect to calcite, as crab grow large enough to deter most predators (Armstrong et al., 1987; Dew, 1990). Both species are sensitive to ocean acidification. When exposed to acidified waters, red king crab embryos show altered development (Long, Swiney, and Foy, 2013). Exposure at both the embryonic stage (carryover effects) and larval stage reduces the time to larval starvation in the laboratory (Long, Swiney, and Foy, 2013). Juveniles suffer increased mortality and decreased growth in acidified water (Long, Swiney, Harris, et al., 2013), and there is a synergistic negative effect on survival when acidification is combined with increased temperatures (Swiney et al., 2017). Blue king crab are less well-studied but, similarly to red king crab, the juveniles have decreased survival and growth in acidified water (Long et al., 2017). Red king crab appear to be more sensitive; juvenile red king crab are affected at a pH of 7.8, whereas blue king crab are not affected at pHs above 7.8. Although the calcium content is not affected, the juveniles of both species have reduced microhardness in the chela (though not in the carapace) when held at a pH of 7.5 (Coffey et al., 2017). Given these results, it is projected that the level of ocean acidification projected for the next 50–100 years will have a substantial negative effect on red king crab populations and fisheries in the absence of any acclimation (Punt et al., 2014). These studies set the stage for our beginning to understand the energetic and physiological responses of these species to ocean acidification. Understanding how metabolic demand changes during exposure to ocean acidification is critical to interpreting the whole-organism responses found in these initial studies. In addition, many species are able to balance the increased energetic demands of ocean acidification on organismal physiology, provided sufficient food is available to meet the demand (Hettinger, Sanford, Hill, Hosfelt, et al., 2013; Pansch et al., 2014). Consequently, understanding how feeding changes in light of metabolic demand is also important. In this study, we determine how respiration, feeding ration, and growth in juvenile red and blue king crabs are affected by ocean acidification. Methods Juvenile king crab for this experiment were reared by the Alutiiq Pride Shellfish Hatchery using standard published protocols (Swingle et al., 2013; Long, 2016) from red king crab broodstock obtained from Alitak Bay in 2013 and blue king crab broodstock obtained from the St. Matthew stock in 2012. All broodstock were caught in crab pots. Year-0 red king crab, ∼3 weeks from settlement, and year-1 blue king crab, ∼1 year from settlement, were used for this study. Juvenile crab were shipped to the Kodiak Laboratory in insulated containers, and were held in tanks with flowing, unfiltered seawater at ambient pH and temperature. Juvenile crab were fed a diet of frozen Artemia (Brine Shrimp Direct, Ogden, UT, USA), frozen bloodworms (Brine Shrimp Direct), frozen Cyclop-eeze (Argent Laboratories, Redmond, WA, USA), Cyclop-eeze flakes, and Gelly Belly (Florida Aqua Farms, Dade City, FL, USA) mixed with Cyclop-eeze powder, and walleye pollock (Gadus chalcogrammus) bone powder (US Department of Agriculture, Agricultural Research Service, Kodiak, AK, USA) three times per week to excess prior to the experiment. The average mass of red king crab at the beginning of the experiment was 0.0111 ± 0.0032 (SD) g and crabs were at the first to second crab stage. Blue king crab were 0.0486 ± 0.0156 (SD) g and were probably at the fifth to sixth crab stage. During these experiments, crab were held in individual cells made out of PVC pipe (52 mm diameter, 5 cm tall) with mesh bottoms, which were placed in three larger experimental tubs [53 (L) × 38 (W) × 23 (H) cm], water volume 24 l. Although there was no replication at the tub level (meaning that treatment differences cannot be fully disentangled from tub) given the crabs were isolated from each other, the flow rate and temperatures were checked daily, the high exchange rate of water (1 tank exchange every 24 min), and the very low biomass (∼370 mg crab per tank or ∼15 mg/l) the effect of tub can be reasonably assumed to be negligible. Cells received flow-through water at 100 ml/min and were large enough to avoid causing stress to the animals (Swiney et al., 2013). Crab were fed Gelly Belly (above) three times a week to excess during the experiment. Water temperature was maintained at 5°C, using a recirculating chiller, which is well within the natural range experienced by both species in the wild (incoming ambient seawater in our laboratory ranges from ∼0 to 12°C annually and Kodiak is within the geographic range of both species) and within the thermal tolerance range for both species (Long and Daly, 2017). Each of the tubs was fed with flow-through seawater at 1 l/min at one of three experimental pH levels. Water flow, into both the tubs and the individual inserts, was checked daily and adjusted if necessary. Seawater was acidified as described by Long, Swiney, and Foy (2013). In brief, CO2 was bubbled into seawater to reduce the pH to 5.5. This low pH water was mixed with ambient filtered seawater into treatment head tanks. The flow rate of pH 5.5 water was controlled via feedback from pH probes in the head tanks that adjusted the speed of peristaltic pumps. Three pH treatments were used: ambient (pH ∼8.1), pH 7.8 (pH expected in global surface waters in ∼2100), and pH 7.5 (∼2200) (Caldeira and Wickett, 2003). These pH treatments were, in part, use to allow direct comparisons between this experiment and previous experiments performed on the same species and life-history stages (Long, Swiney, Harris, et al., 2013; Long et al., 2017). The pH and temperature were measured in a randomly selected cell in each treatment once a day using a Durafet III pH probe which was calibrated daily using TRIS buffer made according to Millero (1986). Previous experiments using the same experimental setup have demonstrated that the variability in the pH and temperature measurements among cells in the same tank is less than the nominal precision of the probe, and typically 0 (Long et al., 2017). Weekly water samples were taken from each treatment, poisoned with 0.02% of total sample volume of saturated mercuric chloride solution according to the best practice (Dickson et al., 2007), and sent to an analytic laboratory for dissolved inorganic carbon (DIC) and total alkalinity (TA) analysis. DIC and TA were determined using a VINDTA 3 C (Marianda, Kiel, Germany) coupled to a 5012 Coulometer (UIC, Inc., Joliet, IL, USA), using Certified Reference Material from the Dickson Laboratory (Scripps Institution, San Diego, CA, USA) and the procedures in DOE (1994). Other components of the carbonate system were calculated from the measured pH and DIC using seacarb package in R (V2.2.3, Vienna, Austria, Lavigne and Gattuse, 2011). As an internal check of our measurements we also calculated the pH of the seawater from the measured DIC and Alkalinity using seacarb and compared it to the measured pH. The average difference between the calculated and measured pH was −0.01 ± 0.018 (s.e.) pH units indicating a good level of agreement. In this experiment, we measured respiration and feeding ration twice for each crab once immediately after exposure to treatment water and once after a 3-week acclimation period in treatment water. A 3-week exposure period was selected, in part based on previous experiments where juvenile red king crab mortality rate in acidified conditions increased substantially after the 3-week mark (Long, Swiney, Harris, et al., 2013). Thus, 3 weeks represent the maximum exposure time possible to ensure a sufficient sample size at the end of the experiment in the lowest pH treatment while getting measurements over the longest time frame possible. In addition, keeping mortality to a minimum avoided complications in interpretation inherent if differential mortality among the treatments weeded out the individuals most sensitive to low pH. Sample size was 6 crab per species per treatment, except for red king crab at pH 7.5, which we increased to 10 crab in anticipation of a higher mortality rate at that treatment. Total stocking density was 12–16 crabs/tub. As no more than five respirometry measurements could be made per day, trials for individual crab were staggered, and crab were started in a random order, with each crab assigned to its treatment randomly. Partway through the initial sets of measurements for the experiment, an equipment failure caused mass mortality in the pH 7.5 treatment. All crabs (14 out of 16 total for that treatment) present in the tub at the time were removed regardless of whether they had died, and the initial respiration/feeding trials were re-run using new crabs according to the same protocol. Each crab was starved for 1 day prior to measuring the respiration and feeding ration to standardize hunger levels. Respiration was measured in a static, adjustable volume (max 5 ml) Plexiglas cell with an integrated Clark electrode oxygen sensor which continuously recorded the concentration of dissolved O2 in mg/l to the nearest 0.01 mg/l. This sensor was calibrated daily using a two-point calibration procedure at 0 and 100% saturation. The cell was jacketed in a secondary chamber that allowed flow-through water to maintain the cell at a constant temperature (but this water flowed around the cell and did not flow into the cell), and the whole apparatus was placed inside a temperature-controlled room at 5°C. To measure respiration rates, crabs were taken out of the ambient pH water in their holding tank and placed into the cell with a known volume of water (2 ml for red king crab and 3 ml for blue king crab) at the appropriate treatment pH. A vented plunger was depressed into the cell to purge all air in the cell. The water within the cell was continuously mixed with a magnetic stir bar and was not exchanged during the trial. A small piece of fibreglass mesh inserted into the cell was used to keep the crab away from the stir bar. Trials were run for ∼1.0 to 1.5 h. Immediately after the trial, the crab were removed from the chamber and blotted dry, and the individual wet mass was determined. The rate of oxygen consumption in the cell was calculated by determining the slope of the change in oxygen concentration over time once the trend became linear, usually within ∼15 min, and was normalized to the wet mass of each crab. After the initial respiration trials, the crab were placed in their holding cells within the experimental holding system (see above). Feeding ration was determined the same day as respiration measurements were taken. A pre-massed piece of squid mantle (blotted dry) ∼50% of the mass of the given individual crab was placed into each cell. Crab were allowed to feed for 24 h, after which the remaining food was collected, blotted dry, and massed. As the red king crab were smaller than the blue king crab, the mass of food given to each species differed accordingly. Control trials without crab were performed in each pH treatment for each species (to account for any potential difference in the initial mass of the food samples), with three replicates of each pH/species combination. On average, the mass of squid increased by 0.8% ± 7.8 (s.e.) and did not differ among pH treatments (two-way ANOVA, F2,13 = 0.184, p = 0.834) or species (two-way ANOVA, F1,13 = 1.318, p = 0.272), so the overall mean was used when calculating the feeding ration. The mass of food consumed was determined, and the feeding ration was calculated as the percent of the crab’s mass consumed corrected for mass change in control trials. The crab were held in their treatment water for ∼21 days (range: 20–24 days), and checked daily for moults or mortalities. After the ∼3 weeks, the respiration and feeding ration for each crab was then determined a second time following the above procedures. Respiration and feeding ration were analysed via repeated-measures ANOVAs, with time (initial and final measurements) fully crossed with pH treatment, species, and crab number (nested within pH treatment and species) as factors and with crab mass as a covariate. In all analyses, treatment could not be unambiguously disentangled from any potential tub effects because of a lack of replication at the tub level. Homogeneity of variance was verified with Levene’s test and post hoc comparisons were tested using Fishers least significant difference tests in all ANOVA analyses. Growth was calculated as the percent change in wet weight between the initial and final measurements. As few blue king crab moulted (only 2 in the pH 7.5 treatments), there was insufficient data for analysis. Growth in red king crab, where three crabs moulted in each treatment, was analysed with a fully crossed two-way ANOVA, with pH treatment and moult status (whether they moulted or not) as factors. Data availability The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request. Results The pH in the ambient tank averaged 8.14 ± 0.03 (SD) while the pHs in the two treatment tanks averaged 7.80 ± 0.03 and 7.50 ± 0.03 (Table 1). As expected, TA did not vary among the treatments. The water in the pH 7.8 treatment was undersaturated with respect to aragonite, and the water in the pH 7.5 treatment was undersaturated with respect to both calcite and aragonite (Table 1). Table 1. Water chemistry parameters in the three treatment tanks during experimental exposure of red and blue king crabs to varying levels of ocean acidification. pHF pCO2 HCO3− CO3−2 DIC Alkalinity ΩAragonite ΩCalcite Treatment µatm mmol/kg mmol/kg mmol/kg mmol/kg Ambient 8.14 ± 0.03 331.83 ± 28.95 1.86 ± 0.01 0.11 ± 0.01 1.98 ± 0.01 2.12 ± 0.01 1.63 ± 0.11 2.60 ± 0.18 pH 7.8 7.80 ± 0.03 767.92 ± 53.53 1.98 ± 0.01 0.05 ± 0.00 2.07 ± 0.01 2.10 ± 0.02 0.80 ± 0.06 1.27 ± 0.10 pH 7.5 7.50 ± 0.03 1555.49 ± 44.22 2.04 ± 0.00 0.03 ± 0.00 2.15 ± 0.00 2.11 ± 0.01 0.41 ± 0.01 0.66 ± 0.02 pHF pCO2 HCO3− CO3−2 DIC Alkalinity ΩAragonite ΩCalcite Treatment µatm mmol/kg mmol/kg mmol/kg mmol/kg Ambient 8.14 ± 0.03 331.83 ± 28.95 1.86 ± 0.01 0.11 ± 0.01 1.98 ± 0.01 2.12 ± 0.01 1.63 ± 0.11 2.60 ± 0.18 pH 7.8 7.80 ± 0.03 767.92 ± 53.53 1.98 ± 0.01 0.05 ± 0.00 2.07 ± 0.01 2.10 ± 0.02 0.80 ± 0.06 1.27 ± 0.10 pH 7.5 7.50 ± 0.03 1555.49 ± 44.22 2.04 ± 0.00 0.03 ± 0.00 2.15 ± 0.00 2.11 ± 0.01 0.41 ± 0.01 0.66 ± 0.02 The pH (free scale) was measured daily, the DIC and alkalinity were measured weekly (N = 5), and the other parameters were calculated. Values are means ± 1 SD. ΩAragonite and ΩCalcite are the saturation states of aragonite and calcite, respectively. Open in new tab Table 1. Water chemistry parameters in the three treatment tanks during experimental exposure of red and blue king crabs to varying levels of ocean acidification. pHF pCO2 HCO3− CO3−2 DIC Alkalinity ΩAragonite ΩCalcite Treatment µatm mmol/kg mmol/kg mmol/kg mmol/kg Ambient 8.14 ± 0.03 331.83 ± 28.95 1.86 ± 0.01 0.11 ± 0.01 1.98 ± 0.01 2.12 ± 0.01 1.63 ± 0.11 2.60 ± 0.18 pH 7.8 7.80 ± 0.03 767.92 ± 53.53 1.98 ± 0.01 0.05 ± 0.00 2.07 ± 0.01 2.10 ± 0.02 0.80 ± 0.06 1.27 ± 0.10 pH 7.5 7.50 ± 0.03 1555.49 ± 44.22 2.04 ± 0.00 0.03 ± 0.00 2.15 ± 0.00 2.11 ± 0.01 0.41 ± 0.01 0.66 ± 0.02 pHF pCO2 HCO3− CO3−2 DIC Alkalinity ΩAragonite ΩCalcite Treatment µatm mmol/kg mmol/kg mmol/kg mmol/kg Ambient 8.14 ± 0.03 331.83 ± 28.95 1.86 ± 0.01 0.11 ± 0.01 1.98 ± 0.01 2.12 ± 0.01 1.63 ± 0.11 2.60 ± 0.18 pH 7.8 7.80 ± 0.03 767.92 ± 53.53 1.98 ± 0.01 0.05 ± 0.00 2.07 ± 0.01 2.10 ± 0.02 0.80 ± 0.06 1.27 ± 0.10 pH 7.5 7.50 ± 0.03 1555.49 ± 44.22 2.04 ± 0.00 0.03 ± 0.00 2.15 ± 0.00 2.11 ± 0.01 0.41 ± 0.01 0.66 ± 0.02 The pH (free scale) was measured daily, the DIC and alkalinity were measured weekly (N = 5), and the other parameters were calculated. Values are means ± 1 SD. ΩAragonite and ΩCalcite are the saturation states of aragonite and calcite, respectively. Open in new tab There were no mortalities noted during the ∼3 weeks exposure for either species although two red king crabs, one in the ambient and one in the pH 7.8 treatment, died after the second set of respiration and feeding ration measurements. In respiration trials, initial crab respiration was typically high during the first 5–15 min of the trial, likely due to handling and air exposure (Figure 1). Linear fits used to determine respiration rates were all excellent fits with R2 averaging 0.981 and ranging from 0.878 to 0.999. Respiration rates did not differ between blue and red king crabs under all conditions (Table 2, Figure 2). There was a significant interactive effect between pH treatment and exposure time for both species; the respiration rate at the initial exposure in pH 7.5 water was higher than at either pH 7.8 or ambient water, but after 3 weeks there was no difference among the treatments (Table 2, Figure 2). In the initial measurements, respiration in pH 7.5 water was 73% higher than in ambient water in red king crab, and 178% higher in blue king crab. Feeding ration did not differ between species, among pH treatments, or with exposure time (Table 2, Figure 3). Figure 1. Open in new tabDownload slide Example of data collected during respirometry trial on a blue king crab. Dots are individual measurements of dissolved oxygen within the cell, vertical lines indicate the range of data chosen to determine respiration rate. The line is the best fit linear regression for the data and the p and R2 are given. Note data indicating higher rate of respiration during the initial part of the trial, likely caused by handling and air exposure, which was not used to determine the respiration rate. Figure 1. Open in new tabDownload slide Example of data collected during respirometry trial on a blue king crab. Dots are individual measurements of dissolved oxygen within the cell, vertical lines indicate the range of data chosen to determine respiration rate. The line is the best fit linear regression for the data and the p and R2 are given. Note data indicating higher rate of respiration during the initial part of the trial, likely caused by handling and air exposure, which was not used to determine the respiration rate. Figure 2. Open in new tabDownload slide Effect of pH and exposure time (Initial: immediately after exposure, solid bars; Final: after 3 weeks’ exposure, stippled bars) on the respiration rates in juvenile red and blue king crabs. Error bars are one standard error. Bars with different letters above them differ within each species. Figure 2. Open in new tabDownload slide Effect of pH and exposure time (Initial: immediately after exposure, solid bars; Final: after 3 weeks’ exposure, stippled bars) on the respiration rates in juvenile red and blue king crabs. Error bars are one standard error. Bars with different letters above them differ within each species. Figure 3. Open in new tabDownload slide Effect of pH and exposure time (Initial: immediately after exposure, solid bars; Final: after 3 weeks’ exposure, stippled bars) on the feeding ration in juvenile red and blue king crabs. Error bars are one standard error. No significant differences were found. Figure 3. Open in new tabDownload slide Effect of pH and exposure time (Initial: immediately after exposure, solid bars; Final: after 3 weeks’ exposure, stippled bars) on the feeding ration in juvenile red and blue king crabs. Error bars are one standard error. No significant differences were found. Table 2. RM-ANOVA results examining the effects of pH and exposure time on respiration and feeding ration in red and blue king crabs. df Mean squares F-ratio p-Value Respiration Time 1 0.002 0.073 0.788 pH 2 0.044 1.790 0.184 Species 1 0.010 0.422 0.521 Time*pH 2 0.142 5.732 0.008 Time*species 1 0.000 0.019 0.892 pH*species 2 0.026 1.032 0.368 Time*species*pH 2 0.016 0.637 0.536 Mass 1 0.036 1.469 0.235 Crab (pH*species) 34 0.021 0.830 0.702 Error 31 0.025 – – Feeding ration Time 1 0.845 0.005 0.941 pH 2 24.318 0.158 0.855 Species 1 119.780 0.778 0.385 Time*pH 2 175.866 1.142 0.332 Time*species 1 148.843 0.966 0.333 pH*species 2 274.258 1.781 0.185 Time*species*pH 2 37.859 0.246 0.784 Mass 1 74.651 0.485 0.492 Crab (pH*species) 34 370.709 2.407 0.008 Error 31 154.023 – – df Mean squares F-ratio p-Value Respiration Time 1 0.002 0.073 0.788 pH 2 0.044 1.790 0.184 Species 1 0.010 0.422 0.521 Time*pH 2 0.142 5.732 0.008 Time*species 1 0.000 0.019 0.892 pH*species 2 0.026 1.032 0.368 Time*species*pH 2 0.016 0.637 0.536 Mass 1 0.036 1.469 0.235 Crab (pH*species) 34 0.021 0.830 0.702 Error 31 0.025 – – Feeding ration Time 1 0.845 0.005 0.941 pH 2 24.318 0.158 0.855 Species 1 119.780 0.778 0.385 Time*pH 2 175.866 1.142 0.332 Time*species 1 148.843 0.966 0.333 pH*species 2 274.258 1.781 0.185 Time*species*pH 2 37.859 0.246 0.784 Mass 1 74.651 0.485 0.492 Crab (pH*species) 34 370.709 2.407 0.008 Error 31 154.023 – – Time is either initial or final measurements, pH the pH treatment, species the crab species, mass the mass of the crab, and crab the individual crab on which multiple measurements were done. Significant p-values (<0.05) are highlighted in bold. Open in new tab Table 2. RM-ANOVA results examining the effects of pH and exposure time on respiration and feeding ration in red and blue king crabs. df Mean squares F-ratio p-Value Respiration Time 1 0.002 0.073 0.788 pH 2 0.044 1.790 0.184 Species 1 0.010 0.422 0.521 Time*pH 2 0.142 5.732 0.008 Time*species 1 0.000 0.019 0.892 pH*species 2 0.026 1.032 0.368 Time*species*pH 2 0.016 0.637 0.536 Mass 1 0.036 1.469 0.235 Crab (pH*species) 34 0.021 0.830 0.702 Error 31 0.025 – – Feeding ration Time 1 0.845 0.005 0.941 pH 2 24.318 0.158 0.855 Species 1 119.780 0.778 0.385 Time*pH 2 175.866 1.142 0.332 Time*species 1 148.843 0.966 0.333 pH*species 2 274.258 1.781 0.185 Time*species*pH 2 37.859 0.246 0.784 Mass 1 74.651 0.485 0.492 Crab (pH*species) 34 370.709 2.407 0.008 Error 31 154.023 – – df Mean squares F-ratio p-Value Respiration Time 1 0.002 0.073 0.788 pH 2 0.044 1.790 0.184 Species 1 0.010 0.422 0.521 Time*pH 2 0.142 5.732 0.008 Time*species 1 0.000 0.019 0.892 pH*species 2 0.026 1.032 0.368 Time*species*pH 2 0.016 0.637 0.536 Mass 1 0.036 1.469 0.235 Crab (pH*species) 34 0.021 0.830 0.702 Error 31 0.025 – – Feeding ration Time 1 0.845 0.005 0.941 pH 2 24.318 0.158 0.855 Species 1 119.780 0.778 0.385 Time*pH 2 175.866 1.142 0.332 Time*species 1 148.843 0.966 0.333 pH*species 2 274.258 1.781 0.185 Time*species*pH 2 37.859 0.246 0.784 Mass 1 74.651 0.485 0.492 Crab (pH*species) 34 370.709 2.407 0.008 Error 31 154.023 – – Time is either initial or final measurements, pH the pH treatment, species the crab species, mass the mass of the crab, and crab the individual crab on which multiple measurements were done. Significant p-values (<0.05) are highlighted in bold. Open in new tab Although growth was not of primary interest in this experiment, sufficient red king crab moulted that we present the data for comparisons to other studies. There was a significant interactive effect between pH treatment and moulting on growth of red king crab (Table 3). Under all pH treatments, red king crab that did not moult increased in mass by ∼10% over the 3-week experiment (Figure 4). However, there was a large difference in growth among treatments for red king crab that did moult, with crab in ambient water growing more than those in pH 7.5 (Figure 4). Indeed, at pH 7.5, red king crab that did moult had a smaller (albeit non-significant) increase in mass than those that did not moult (Figure 3) and the average increase in mass was statistically indistinguishable from 0 (one sample t-test, p = 0.281). Ambient crab also grew 83% more than those in pH 7.8 (Figure 4), but the difference, though large, was not quite significant (p = 0.064). Blue king crab that did not moult increased in mass by 1.7 ± 1.1% and those that did by 21.8 ± 10.5% (s.e.). Figure 4. Open in new tabDownload slide Effect of pH and moulting on growth in juvenile red king crabs. Error bars are one standard error. Bars with different letters above them differ statistically. Figure 4. Open in new tabDownload slide Effect of pH and moulting on growth in juvenile red king crabs. Error bars are one standard error. Bars with different letters above them differ statistically. Table 3. ANOVA results for effect of pH on growth as measured as a percent change in wet mass in red king crab. df Mean squares F-ratio p-Value Moulted 1 1 492.253 6.845 0.020 pH 2 1 018.971 4.674 0.028 Moulted*pH 2 938.653 4.305 0.035 Error 14 218.015 df Mean squares F-ratio p-Value Moulted 1 1 492.253 6.845 0.020 pH 2 1 018.971 4.674 0.028 Moulted*pH 2 938.653 4.305 0.035 Error 14 218.015 Moulting represents whether the crab moulted during the experiment and pH the pH treatment. Open in new tab Table 3. ANOVA results for effect of pH on growth as measured as a percent change in wet mass in red king crab. df Mean squares F-ratio p-Value Moulted 1 1 492.253 6.845 0.020 pH 2 1 018.971 4.674 0.028 Moulted*pH 2 938.653 4.305 0.035 Error 14 218.015 df Mean squares F-ratio p-Value Moulted 1 1 492.253 6.845 0.020 pH 2 1 018.971 4.674 0.028 Moulted*pH 2 938.653 4.305 0.035 Error 14 218.015 Moulting represents whether the crab moulted during the experiment and pH the pH treatment. Open in new tab Discussion This study helps to shine light on the energetic and physiological underpinnings of the macro-responses of red and blue king crabs to ocean acidification. The initial higher respiration rate in both species at the lowest pH as compared to ambient pH suggests an energetic cost from the crabs buffering their haemolymph in response to an immediate change in water pH; although given the crab’s sizes we were not able to directly measure this buffering response. This higher metabolism at the onset of the exposure did not lead to increased food intake, however. After 3 weeks in treatment water, respiration rates were the same as those in ambient pH water and feeding rates were unchanged; however, decreased growth [which occurred here for red king crab, but has been demonstrated more robustly for both species in longer-term studies (Long, Swiney, Harris, et al., 2013; Long et al., 2017)] suggests that the crab continued to expend a greater amount of energy on osmo-/iono-regulation in acidified water, compared to crabs held in ambient water without an increase in feeding, thus reducing the energy available for growth. As the two species used in this study were at different ages and stages it must be acknowledged that any direct comparisons between the two in this study could have been affected by this; however, this does not change the interpretation of the results in regards to the physiological effects of ocean acidification. Furthermore, for all measurements other than the initial respiration rate, the effect of tub could not be estimated because of a lack of replication at the tub level; however, given the flow rates and the isolation of the crabs within the tubs (see Methods section) the effects of tub are assumed to be negligible. More experiments will be necessary to completely understand the physiological response of these crabs to reduced pH water, and to estimate the degree to which acclimation and adaptation may ameliorate the negative effects of ocean acidification at a population level. To maintain homeostasis, crustaceans exposed to decreased pH/increased pCO2 must respond by either buffering their haemolymph or reducing CO2 production. Failure to do so results in decreased protein function and the accompanying disruption of biological processes. One mechanism crustaceans use is to reduce metabolism and thus internal production of CO2, as is the case for both the Dungeness crab, Metacarcinus magister, and the krill, Euphausia pacifica (Hans et al., 2014; Cooper et al., 2016). An alternative, non-exclusive mechanism is active transport of bicarbonate into the haemolymph at the gills, primarily through Cl−/HCO3− exchange, which buffers the haemolymph and reduces, reverses, or prevents acidosis (Whiteley, 2011). This mechanism is very common in marine and estuarine crustaceans (Pane and Barry, 2007; Dissanayake et al., 2010; Knapp et al., 2015). Given the near-ubiquity of this response and the fact that active ion transport is energetically costly, it seems likely that the initial higher respiration rate that occurred in both red and blue king crabs in low pH water compared to those in ambient water was at least partially due to the energetic cost of buffering the haemolymph with bicarbonate. This conclusion is further substantiated by the fact that all red king crab life history stages examined, including larval, juvenile, and adult, have either increased or unchanged calcium content in acidified water (Long, Swiney, and Foy, 2013; Long, Swiney, Harris, et al., 2013). Increasing the bicarbonate ion concentration in the haemolymph may make the internal calcification process employed by crustaceans more effective, thus explaining the increase in calcification that is a frequent result of exposure to CO2 acidified water in this taxa (Ries et al., 2009); however, calcium content is not necessarily correlated with other measures of exoskeleton structure or function (Coffey et al., 2017; Glandon et al., 2018). In this study, neither red nor blue king crab increased their feeding rate under acidified conditions in either the short or long term despite a very high initial increase in respiration. This could be because both species are at their maximum possible feeding rate at this temperature. In endotherms, feeding rates increase with temperature (Brett, 1971). In this study, both species fed more (based on interpolation between measurements made at 2 and 6°C for red king crab and 4 and 6°C for blue king crab) at 5°C than in previous studies (Stoner et al., 2010; Long and Daly, 2017) suggesting that these crab may be incapable of increasing feeding, at least at this temperature. This may be why high food availability did not reduce or eliminate negative effects of ocean acidification as can occur in both barnacles, Amphibalanus improvisus, and the Olympia oyster, Ostrea lurida (Hettinger, Sanford, Hill, Hosfelt, et al., 2013; Pansch et al., 2014). Long-term trials suggest that slightly (+2°C) increased temperatures may reduce negative effects on mortality in red king crab juveniles (Swiney et al., 2017), and this could be because higher temperatures allow for a higher feeding rate; however, a larger increase in temperature in the same study (+4°C) resulted in a synergistic increase in the mortality rate suggesting that at that point the increase in metabolic demand due to the combined stressors of pH and temperature could not be compensated for by increased feeding. Further work on the interaction between pH and temperature on the physiology of these species, particularity blue king crab, is warranted. The initial higher oxygen consumption did not persist after 3 weeks exposure to low pH water in red or blue king crab which, combined with the lack of an increase in feeding, suggests that the crabs had reached a physiological equilibrium by that point. Palaemon elegans and P. serratus both suffer initial acidosis over at least the first 14 days, but after 30 days of exposure they are able to completely compensate for the reduced ambient pH (Dissanayake et al., 2010) and Dungeness crabs do the same but within 24 h after exposure to acidified water (Pane and Barry, 2007). Thus, our timeframe for apparent compensation within 3 weeks is comparable to other crustaceans. However, the reduced growth in red king crab that occurred in this study, and the increased mortality and decreased growth that occurs in both species in acidified conditions in longer-term, 1 year for blue and 6 months for red king crab, experiments (Long, Swiney, Harris, et al., 2013; Long et al., 2017), suggest an ongoing energetic cost to maintaining acid/base balance against an ever-present pH gradient. This higher cost of maintaining homeostasis, if uncompensated by increased feeding, would necessarily divert energy away from other biological functions such as growth. A similar effect occurs in the brittlestar, Amphiura filiformis, which increases both metabolism and calcification under acidified conditions but suffers reduced muscle mass (Wood et al., 2008). Future research should look at other functions that could be affected by decreased energy availability such as immune function (Meseck et al., 2016) or, in mature animals, reproductive output (Dupont et al., 2013). More research will be necessary to better quantify and model the physiological response in these species. It is worth noting that in this experiment as well as in virtually all ocean acidification experiments, the pH change experienced by the experimental animals is more akin to an upwelling or other acute event rather than the gradual change in pH predicted from rising atmospheric CO2 levels. Thus while the acute exposure experiments conducted to date suggest substantial population and fishery level effects within a few decades for at least red king crab (Punt et al., 2014), there is potential for evolutionary adaptation in both species which should be investigated. In addition, carryover effects from previous life-history stages (Hettinger et al., 2012; Hettinger, Sanford, Hill, Lenz, et al., 2013) as well as transgenerational effects from parental exposure to low pH (Parker et al., 2012; Long et al., 2016) can both either ameliorate or exacerbate the effect on an organism and also need to be explored further for these species. Finally, co-occurring stressors, especially temperature (Swiney et al., 2017), could alter the effects of ocean acidification, and should also be considered (Breitburg et al., 2015). Acknowledgements We thank Jeff Hetrick and the Alutiiq Pride Shellfish Hatchery for rearing the juvenile crab, and staff of the Kodiak Laboratory, particularly B. Daly and L. Harver, for help with animal husbandry and performing the experiments. . Previous versions of this paper were improved by comments from J. Richar and J. Long. The findings and conclusions in the paper are those of the authors and do not necessarily represent the views of the National Marine Fisheries Service, NOAA. Reference to trade names or commercial firms does not imply endorsement by the National Marine Fisheries Service, NOAA. Funding We thank the NOAA Ocean Acidification Program for support of the project. P. Pruisner was funded through a NOAA Holling’s Scholar fellowship. References Appelhans Y. S. , Thomsen J. , Pansch C. , Melzner F. , Wahl M. 2012 . Sour times: seawater acidification effects on growth, feeding behaviour and acid–base status of Asterias rubens and Carcinus maenas . Marine Ecology Progress Series , 459 : 85 – 98 . Google Scholar Crossref Search ADS WorldCat Armstrong D. A. , Armstrong J. L. , Jenson G. , Palacios R. , Williams G. 1987 . Distribution, abundance, and biology of blue king and Korean hair crabs around the Pribilof Islands . Outer Continental Shelf Environmental Assessment Program: Final Reports of Principal Investigators , 67 : 1 – 278 . WorldCat Baragi L. V. , Anil A. C. 2015 . Interactive effect of elevated pCO2 and temperature on the larval development of an inter-tidal organism, Balanus amphitrite Darwin (Cirripedia: Thoracica) . Journal of Experimental Marine Biology and Ecology , 471 : 48 – 57 . Google Scholar Crossref Search ADS WorldCat Boßelmann F. , Romano P. , Fabritius H. , Raabe D. , Epple M. 2007 . The composition of the exoskeleton of two crustacea: the American lobster Homarus americanus and the edible crab Cancer pagurus . Thermochimica Acta , 463 : 65 – 68 . Google Scholar Crossref Search ADS WorldCat Breitburg D. L. , Salisbury J. , Bernhard J. M. , Cai W.-J. , Dupont S. , Doney S. C. , Kroeker K. J. 2015 . And on top of all that …. Coping with ocean acidification in the midst of many stressors . Oceanography , 28 : 48 – 61 . Google Scholar Crossref Search ADS WorldCat Brett J. R. 1971 . Energetic responses of salmon to temperature. A study of some thermal relations in the physiology and freshwater ecology of sockeye salmon (Oncorhynchus nerka) . American Zoologist , 11 : 99 – 113 . Google Scholar Crossref Search ADS WorldCat Caldeira K. , Wickett M. E. 2003 . Anthropogenic carbon and ocean pH . Nature , 425 : 365 . Google Scholar Crossref Search ADS PubMed WorldCat Chen P.-Y. , Lin A. Y.-M. , McKittrick J. , Meyers M. A. 2008 . Structure and mechanical properties of crab exoskeletons . Acta Biomaterialia , 4 : 587 – 596 . Google Scholar Crossref Search ADS PubMed WorldCat Coffey W. D. , Nardone J. A. , Yarram A. , Long W. C. , Swiney K. M. , Foy R. J. , Dickinson G. H. 2017 . Ocean acidification leads to altered micromechanical properties of the mineralized cuticle of juvenile red and blue king crabs . Journal of Experimental Marine Biology and Ecology , 495 : 1 – 12 . Google Scholar Crossref Search ADS WorldCat Cooper H. L. , Potts D. C. , Paytan A. 2016 . Metabolic responses of the North Pacific krill, Euphausia pacifica, to short- and long-term pCO2 exposure . Marine Biology , 163 : 206. Google Scholar Crossref Search ADS WorldCat Cross J. N. , Mathis J. T. , Bates N. R. , Byrne R. H. 2013 . Conservative and non-conservative variations of total alkalinity on the southeastern Bering Sea shelf . Marine Chemistry , 154 : 100 – 112 . Google Scholar Crossref Search ADS WorldCat Daly B. , Long W. C. 2014a . Inter-cohort cannibalism of early benthic phase blue king crabs (Paralithodes platypus): alternate foraging strategies in different habitats lead to different functional responses . PLoS One , 9 : e88694. Google Scholar Crossref Search ADS WorldCat Daly B. , Long W. C. 2014b . Intra-guild predation among early benthic phase red and blue king crabs: evidence for a habitat-mediated competitive advantage . Journal of Experimental Marine Biology and Ecology , 451 : 98 – 104 . Google Scholar Crossref Search ADS WorldCat Daly B. , Stoner A. W. , Eckert G. L. 2012 . Predator-induced behavioral plasticity of juvenile red king crabs (Paralithodes camtschaticus) . Journal of Experimental Marine Biology and Ecology , 429 : 47 – 54 . Google Scholar Crossref Search ADS WorldCat Dew C. B. 1990 . Behavioral ecology of podding red king crab, Paralithodes camtschatica . Canadian Journal of Fisheries and Aquatic Sciences , 47 : 1944 – 1958 . Google Scholar Crossref Search ADS WorldCat Dickson A. G. , Sabine C. L. , Christian J. R. 2007 . Guide to Best Practices for Ocean CO2 Measurements . PICES Special Publication , 3 . 191 pp. WorldCat Dissanayake A. , Clough R. , Spicer J. I. , Jones M. B. 2010 . Effects of hypercapnia on acid–base balance and osmo-/iono-regulation in prawns (Decapoda: Palaemonidae) . Aquatic Biology , 11 : 27 – 36 . Google Scholar Crossref Search ADS WorldCat DOE. 1994 . Handbook of Methods for the Analysis of the Various Parameters of the Carbon Dioxide System in Sea Water. Version 2. ORNL/CDIAC-74. 197 pp. Dupont S. , Dorey N. , Stumpp M. , Melzner F. , Thorndyke M. 2013 . Long-term and trans-life-cycle effects of exposure to ocean acidification in the green sea urchin Strongylocentrotus droebachiensis . Marine Biology , 160 : 1835 – 1843 . Google Scholar Crossref Search ADS WorldCat Fabry V. J. , McClintock J. B. , Mathis J. T. , Grebmeier J. M. 2009 . Ocean acidification at high latitudes: the bellwether . Oceanography , 22 : 160 – 171 . Google Scholar Crossref Search ADS WorldCat Fabry V. J. , Seibel B. A. , Feely R. A. , Orr J. C. 2008 . Impacts of ocean acidification on marine fauna and ecosystem processes . ICES Journal of Marine Science , 65 : 414 – 432 . Google Scholar Crossref Search ADS WorldCat Feely R. A. , Sabine C. L. , Lee K. , Berelson W. , Kleypas J. , Fabry V. J. , Millero F. J. 2004 . Impact of anthropogenic CO2 on the CaCO3 system in the oceans . Science , 305 : 362 – 366 . Google Scholar Crossref Search ADS PubMed WorldCat Glandon H. L. , Kilbourne K. H. , Schijf J. , Miller T. J. 2018 . Counteractive effects of increased temperature and pCO2 on the thickness and chemistry of the carapace of juvenile blue crab, Callinectes sapidus, from the Patuxent River, Chesapeake Bay . Journal of Experimental Marine Biology and Ecology , 498 : 39 – 45 . Google Scholar Crossref Search ADS WorldCat Hans S. , Fehsenfeld S. , Treberg J. R. , Weihrauch D. 2014 . Acid–base regulation in the Dungeness crab (Metacarcinus magister) . Marine Biology , 161 : 1179 – 1193 . Google Scholar Crossref Search ADS WorldCat Hettinger A. , Sanford E. , Hill T. M. , Hosfelt J. D. , Russell A. D. , Gaylord B. 2013 . The influence of food supply on the response of Olympia oyster larvae to ocean acidification . Biogeosciences , 10 : 6629 – 6638 . Google Scholar Crossref Search ADS WorldCat Hettinger A. , Sanford E. , Hill T. M. , Lenz E. A. , Russell A. D. , Gaylord B. 2013 . Larval carry-over effects from ocean acidification persist in the natural environment . Global Change Biology , 19 : 3317 – 3326 . Google Scholar PubMed WorldCat Hettinger A. , Sanford E. , Hill T. M. , Russell A. D. , Sato K. N. S. , Hoey J. , Forsch M. , et al. . 2012 . Persistent carry-over effects of planktonic exposure to ocean acidification in the Olympia oyster . Ecology , 93 : 2758 – 2768 . Google Scholar Crossref Search ADS PubMed WorldCat Jensen G. C. , Armstrong D. A. 1989 . Biennial reproductive cycle of blue king crab, Paralithodes platypus, at the Pribilof Islands, Alaska and comparison to a congener, P. camtschatica . Canadian Journal of Fisheries and Aquatic Sciences , 46 : 932 – 940 . Google Scholar Crossref Search ADS WorldCat Keppel E. A. , Scrosati R. A. , Courtenay S. C. 2012 . Ocean acidification decreases growth and development in American lobster (Homarus americanus) larvae . Journal of Northwest Atlantic Fishery Science , 44 : 61 – 66 . Google Scholar Crossref Search ADS WorldCat Knapp J. L. , Bridges C. R. , Krohn J. , Hoffman L. C. , Auerswald L. 2015 . Acid–base balance and changes in haemolymph properties of the South African rock lobsters, Jasus lalandii, a palinurid decapod, during chronic hypercapnia . Biochemical and Biophysical Research Communications , 461 : 475 – 480 . Google Scholar Crossref Search ADS PubMed WorldCat Kroeker K. J. , Kordas R. L. , Crim R. N. , Singh G. G. 2010 . Meta-analysis reveals negative yet variable effects of ocean acidification on marine organisms . Ecology Letters , 13 : 1419 – 1434 . Google Scholar Crossref Search ADS PubMed WorldCat Lavigne H. , Gattuse J. 2011 . seacarb: Seawater Carbonate Chemistry With R. http://CRAN.R-project.org/package=seacarb (last accessed 18 November 2016). R package version 2.4.6 edn. Long W. C. 2016 . A new quantitative model of multiple transitions between discrete stages, applied to the development of crustacean larvae . Fishery Bulletin , 114 : 58 – 66 . Google Scholar Crossref Search ADS WorldCat Long W. C. , Daly B. 2017 . Upper thermal tolerance in red and blue king crab: sublethal and lethal effects . Marine Biology , 164 : 162. Google Scholar Crossref Search ADS WorldCat Long W. C. , Swiney K. M. , Foy R. J. 2013 . Effects of ocean acidification on the embryos and larvae of red king crab, Paralithodes camtschaticus . Marine Pollution Bulletin , 69 : 38 – 47 . Google Scholar Crossref Search ADS PubMed WorldCat Long W. C. , Swiney K. M. , Foy R. J. 2016 . Effects of high pCO2 on Tanner crab reproduction and early life history, Part II: carryover effects on larvae from oogenesis and embryogenesis are stronger than direct effects . ICES Journal of Marine Science , 73 : 836 – 848 . Google Scholar Crossref Search ADS WorldCat Long W. C. , Swiney K. M. , Harris C. , Page H. N. , Foy R. J. 2013 . Effects of ocean acidification on juvenile red king crab (Paralithodes camtschaticus) and Tanner crab (Chionoecetes bairdi) growth, condition, calcification, and survival . PLoS One , 8 : e60959 . Google Scholar Crossref Search ADS PubMed WorldCat Long W. C. , Van Sant S. B. , Swiney K. M. , Foy R. 2017 . Survival, growth, and morphology of blue king crabs: effect of ocean acidification decreases with exposure time . ICES Journal of Marine Science , 74 : 1033 – 1041 . Google Scholar Crossref Search ADS WorldCat Long W. C. , Whitefleet-Smith L. 2013 . Cannibalism in red king crab: habitat, ontogeny, and the predator functional response . Journal of Experimental Marine Biology and Ecology , 449 : 142 – 148 . Google Scholar Crossref Search ADS WorldCat Mathis J. T. , Cross J. N. , Evans W. , Doney S. C. 2015 . Ocean acidification in the surface waters of the Pacific-Arctic boundary regions . Oceanography , 28 : 122 – 135 . Google Scholar Crossref Search ADS WorldCat Mathis J. T. , Cross J. N. , Monacci N. , Feely R. A. , Stabeno P. 2014 . Evidence of prolonged aragonite undersaturations in the bottom waters of the southern Bering Sea shelf from autonomous sensors . Deep-Sea Research Part II: Topical Studies in Oceanography , 109 : 125 – 133 . Google Scholar Crossref Search ADS WorldCat McMurray G. , Vogel A. H. , Fishman P. A. 1984 . Distribution of larval and juvenile red king crab (Paralithodes camtschatica) in Bristol Bay . Outer Continental Shelf Environmental Assessment Program: Final Reports of Principal Investigators , 53 : 267 – 477 . WorldCat Meseck S. L. , Alix J. H. , Swiney K. M. , Long W. C. , Wikfors G. H. , Foy R. J. 2016 . Ocean acidification affects hemocyte physiology in the Tanner crab (Chionoecetes bairdi) . PLoS One , 11 : e0148477. Google Scholar Crossref Search ADS PubMed WorldCat Millero F. J. 1986 . The pH of estuarine waters . Limnology and Oceanography , 31 : 839 – 847 . Google Scholar Crossref Search ADS WorldCat Pane E. F. , Barry J. P. 2007 . Extracellular acid–base regulation during short-term hypercapnia is effective in a shallow-water crab, but ineffective in a deep-sea crab . Marine Ecology Progress Series , 334 : 1 – 9 . Google Scholar Crossref Search ADS WorldCat Pansch C. , Schaub I. , Havenhand J. , Wahl M. 2014 . Habitat traits and food availability determine the response of marine invertebrates to ocean acidification . Global Change Biology , 20 : 765 – 777 . Google Scholar Crossref Search ADS PubMed WorldCat Parker L. M. , Ross P. M. , O'Connor W. A. , Borysko L. , Raftos D. A. , Pörtner H. O. 2012 . Adult exposure influences offspring response to ocean acidification in oysters . Global Change Biology , 18 : 82 – 92 . Google Scholar Crossref Search ADS WorldCat Punt A. E. , Foy R. J. , Dalton M. G. , Long W. C. , Swiney K. M. 2016 . Effects of long term exposure to ocean acidification on future southern Tanner crab (Chionoecetes bairdi) fisheries management . ICES Journal of Marine Science , 73 : 849 – 864 . Google Scholar Crossref Search ADS WorldCat Punt A. E. , Poljak D. , Dalton M. G. , Foy R. J. 2014 . Evaluating the impact of ocean acidification on fishery yields and profits: the example of red king crab in Bristol Bay . Ecological Modelling , 285 : 39 – 53 . Google Scholar Crossref Search ADS WorldCat Raabe D. , Romano P. , Sachs C. , Fabritius H. , Al-Sawalmih A. , Yi S.-B. , Servos G. , et al. . 2006 . Microstructure and crystallographic texture of the chitin–protein network in the biological composite material of the exoskeleton of the lobster Homarus americanus . Materials Science and Engineering: A , 421 : 143 – 153 . Google Scholar Crossref Search ADS WorldCat Ries J. B. , Cohen A. L. , McCorkle D. C. 2009 . Marine calcifiers exhibit mixed responses to CO2-induced ocean acidification . Geology , 37 : 1131 – 1134 . Google Scholar Crossref Search ADS WorldCat Shirley S. M. , Shirley T. C. 1989 . Interannual variability in density, timing and survival of Alaskan red king crab Paralithodes camtschatica larvae . Marine Ecology Progress Series , 54 : 51 – 59 . Google Scholar Crossref Search ADS WorldCat Somerton D. A. 1985 . The disjunct distribution of blue king crab, Paralithodes platypus, in Alaska: some hypotheses. In Proceedings of the International King Crab Symposium , pp. 13 – 21 . Ed. by Melteff B. University of Alaska Sea Grant Program , Anchorage, AK . Google Preview WorldCat COPAC Stevens B. G. 2003 . Settlement, substratum preference, and survival of red king crab Paralithodes camtschaticus (Tilesius, 1815) glaucothoe on natural substrata in the laboratory . Journal of Experimental Marine Biology and Ecology , 283 : 63 – 78 . Google Scholar Crossref Search ADS WorldCat Stevens B. G. , Persselin S. , Matweyou J. 2008 . Survival of blue king crab Paralithodes platypus Brandt, 1850, larvae in cultivation: effects of diet, temperature and rearing density . Aquaculture Research , 39 : 390 – 397 . Google Scholar Crossref Search ADS WorldCat Stevens B. G. , Swiney K. M. 2007 . Hatch timing, incubation period, and reproductive cycle for captive primiparous and multiparous red king crab, Paralithodes camtschaticus . Journal of Crustacean Biology , 27 : 37 – 48 . Google Scholar Crossref Search ADS WorldCat Stoner A. W. , Ottmar M. L. , Copeman L. A. 2010 . Temperature effects on the molting, growth, and lipid composition of newly-settled red king crab . Journal of Experimental Marine Biology and Ecology , 393 : 138 – 147 . Google Scholar Crossref Search ADS WorldCat Swiney K. M. , Long W. C. , Foy R. J. 2016 . Effects of high pCO2 on Tanner crab reproduction and early life history, Part I: long-term exposure reduces hatching success and female calcification, and alters embryonic development . ICES Journal of Marine Science , 73 : 825 – 835 . Google Scholar Crossref Search ADS WorldCat Swiney K. M. , Long W. C. , Foy R. J. 2017 . Decreased pH and increased temperatures affect young-of-the-year red king crab (Paralithodes camtschaticus ). ICES Journal of Marine Science , 74 : 1191 – 1200 . Google Scholar Crossref Search ADS WorldCat Swiney K. M. , Long W. C. , Persselin S. L. 2013 . The effects of holding space on juvenile red king crab (Paralithodes camtschaticus) growth and survival . Aquaculture Research , 44 : 1007 – 1016 . Google Scholar Crossref Search ADS WorldCat Swingle J. S. , Daly B. , Hetrick J. 2013 . Temperature effects on larval survival, larval period, and health of hatchery-reared red king crab, Paralithodes camtschaticus . Aquaculture , 384–387 : 13 – 18 . Google Scholar Crossref Search ADS WorldCat Tapella F. , Romero M. C. , Stevens B. G. , Buck C. L. 2009 . Substrate preferences and redistribution of blue king crab Paralithodes platypus glaucothoe and first crab on natural substrates in the laboratory . Journal of Experimental Marine Biology and Ecology , 372 : 31 – 35 . Google Scholar Crossref Search ADS WorldCat Wainwright T. C. , Armstrong D. A. , Andersen H. , Dinnel P. A. , Herren D. , Jensen G. C. , Orensanz J. M. , et al. . 1991 . Port Moller King Crab Studies. Annual Report, FRI-UW-9203. 38 pp. Whiteley N. M. 2011 . Physiological and ecological responses of crustaceans to ocean acidification . Marine Ecology Progress Series , 430 : 257 – 271 . Google Scholar Crossref Search ADS WorldCat Wood H. L. , Spicer J. I. , Widdicombe S. 2008 . Ocean acidification may increase calcification rates, but at a cost . Proceedings of the Royal Society B: Biological Sciences , 275 : 1767 – 1773 . Google Scholar Crossref Search ADS WorldCat Zittier Z. M. C. , Hirse T. , Portner H. O. 2013 . The synergistic effects of increasing temperature and CO2 levels on activity capacity and acid–base balance in the spider crab, Hyas araneus . Marine Biology , 160 : 2049 – 2062 . Google Scholar Crossref Search ADS WorldCat Published by International Council for the Exploration of the Sea 2019. This work is written by US Government employees and is in the public domain in the US. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
Food habits of an endangered seabird indicate recent poor forage fish availability off western South AfricaCrawford, Robert J, M;Sydeman, William, J;Thompson, Sarah, Ann;Sherley, Richard, B;Makhado, Azwianewi, B
doi: 10.1093/icesjms/fsz081pmid: N/A
Abstract Large recent decreases of Cape gannets Morus capensis, Cape cormorants Phalacrocorax capensis and African penguins Spheniscus demersus in South Africa resulted in their being listed as Endangered. These seabirds, endemic to the Benguela upwelling system (BUS), primarily rely on anchovy Engraulis encrasicolus and sardine Sardinops sagax for food, yet decreased during periods of abundance of these prey species. In order to investigate this dichotomy, we examined long-term dietary characteristics for gannets in the region in relation to acoustically-derived biomass of prey. Principal component (PC) analysis of diet composition indicated an alternation in the use of anchovy and sardine (PC1), as well as a marked decrease, after the early 2000s, in the availability of these preferred forage resources (PC2). PC2, which we term the Forage Availability Index, was positively related to numbers of gannets and cormorants breeding each year and to estimates of survival of adult penguins at their two largest colonies in northwest South Africa. This indicates that recent availability of anchovy and sardine was insufficient to support these bird populations. Our results emphasize the need to account not only for overall abundance but also local availability of forage resources, when applying an ecosystem approach to managing fisheries for seabirds. Introduction Seabirds are more threatened and their status has deteriorated faster over recent decades than other comparable groups of birds (Croxall et al., 2012). The main at-sea threats are posed by fisheries and climatic impacts on food webs, mortality from fishing gear, and pollution (Croxall et al., 2012). Most marine predators, including cetaceans and large fish species, travel widely and can dive to considerable depths to obtain food (e.g. Leatherwood and Reeves, 1983). However, seabirds are central-place foragers when breeding and most feed in the upper (epipelagic) part of the water column (Shealer, 2002). Hence, it is important to account for the availability of forage resources when considering impacts of fishing and the environment on the prey of seabirds (Boyd et al., 2017; Sydeman et al., 2017). Yet, assessments of forage fish abundance are often undertaken for the entire water column, as is the case for anchovy Engraulis encrasicolus and sardine Sardinops sagax in the southern Benguela upwelling system (BUS; e.g. Augustyn et al., 2018). Furthermore, there have been recent large changes in the environments of several marine ecosystems, including the BUS (e.g. Blamey et al., 2015; Lamont et al., 2018), that may have influenced the availability of forage fish to epipelagic predators. In the BUS, four seabirds subsist mainly on anchovy and sardine (Hockey et al., 2005). Off northwest South Africa, there were recent large decreases in the populations of three of these, all endemic to the BUS: African penguin Spheniscus demersus (Crawford et al., 2011), Cape gannet Morus capensis (Crawford et al., 2014), and Cape cormorant Phalacrocorax capensis (Crawford et al., 2016). Consequently, each species is classified as Endangered by the International Union for Conservation of Nature (Birdlife International, 2017). In contrast, numbers of the fourth seabird, greater crested (swift) tern Thalasseus bergii, increased (Crawford, 2009). Unlike the three endangered species, greater crested terns undertake nomadism between breeding localities (Crawford et al., 2002; Crawford, 2003), can fly overland (Underhill et al., 1999) thereby reducing commuting distances, and have lower energy requirements than the larger penguins, gannets, and cormorants (Gaglio et al., 2018b). The population declines of the three endangered seabirds have been attributed mainly to food stress (Crawford et al., 2015, 2018), although recent stock assessments have indicated a concurrent high biomass of anchovy or sardine off western South Africa (west of Cape Agulhas, Figure 1; e.g. van der Sleen et al., 2018). This presents a conundrum for management and suggests that a greater understanding of food availability is required to elucidate the causes of the seabird decreases (cf. Sherley et al., 2013). Figure 1. Open in new tabDownload slide Important seabird breeding localities off western South Africa north of Cape Town. Triangles indicate localities where the diet of Cape gannets was studied. Figure 1. Open in new tabDownload slide Important seabird breeding localities off western South Africa north of Cape Town. Triangles indicate localities where the diet of Cape gannets was studied. In lieu of the apparent abundance of anchovy or sardine, we hypothesized that the availability of these nutritious resources for predators feeding in the epipelagic zone recently decreased off northwest South Africa. Monitoring changes in dietary preferences of seabirds can provide insights into the availability of prey resources (Votier et al., 2004). Thus, to test our hypothesis, we applied multivariate analyses to the long-term composition of the diet of Cape gannets to describe variation in their take of different prey species. Abundances of anchovy and sardine off South Africa have displayed large fluctuations (Augustyn et al., 2018) and previous studies showed that the contributions of both these species to the diet of gannets were linked to their abundance, with the use of anchovy and sardine inversely related (e.g. Berruti et al., 1993; Crawford and Dyer, 1995; Green et al., 2015). We anticipated that the analyses would reflect this contrast between the use of anchovy and sardine. In accord with our hypothesis, we also predicted a decrease over time in the combined contribution of anchovy and sardine to the diet, i.e. in a Forage Availability Index (FAI) for these preferred prey species. We further envisaged that the index would be better related than biomass estimates to demographic statistics of the three endangered seabirds. We expected the combined contribution of sardine and anchovy to gannet diet to be a good indicator of the availability of these fishes to the epipelagic community of the BUS because anchovy and sardine undertake extensive movements in this system (e.g. Crawford, 1980) and Cape gannets are wide-ranging animals that are known to switch to feeding on offal discarded by fisheries when their preferred prey is at low availability (Grémillet et al., 2008). In contrast African penguins, Cape cormorants, and greater crested terns have smaller foraging extents when breeding and do not generally utilize offal, so that forage fish may dominate their diet even when scarce (Crawford et al., 2014; Gaglio et al., 2018a). Methods Diet composition From 1978 to 2016, the diet of Cape gannets was sampled monthly throughout their breeding season (September–March; Hockey et al., 2005) at Lambert’s Bay and Malgas Island in western South Africa (Figure 1). Gannets returning to the colonies after foraging were caught with a hook on the end of a pole and upturned over an empty bucket, into which many regurgitated (Berruti et al., 1993). The contents of regurgitations were sorted into five categories (sardine, anchovy, saury Scomberesox saury, hakes Meluccius spp., and other species) and the mass of all categories was weighed. For each locality and month, the mass of different categories was summed for all regurgitations and their proportional contribution to the diet was calculated (Berruti et al., 1993). For each of the 38 breeding seasons (1978/1979–2015/2016) that were sampled, the average contribution of a prey category to the diet was obtained for both localities, applying equal weight to each month (Crawford et al., 2014). For the entire period, the average contribution of a prey category to the diet was obtained by applying equal weight to each breeding season. Identification of prey to species was straightforward because nearly all regurgitations comprised easily identified species and were not highly digested (Berruti et al., 1993). Regurgitations provide an indication of the quantity of food returned to colonies rather than the original mass eaten, but for Cape gannets were not thought excessively to influence the proportions by mass of different prey types ingested because often within a regurgitation, the degree of digestion was similar for different prey species. Furthermore, the prey species identified in regurgitations were similar to those found in previous studies that used the stomachs of Cape gannets shot at sea (Hockey et al. 2005). Further details of diet sampling are given in Supplementary data. Forage indices We obtained forage indices by applying principal component analysis (PCA) to the contributions of sardine and anchovy to the diet of Cape gannets at each site by month (September through March) from 1978/1979 to 2015/2016, a 7-month × 38-year matrix. We interpreted derived principal components (PCs) having eigenvalues ≥3 (Jolliffe, 2002) by examining loadings of monthly diet composition on the PC. One PC reflected prey switching between anchovy and sardine (PC1), whereas the other (PC2) appeared to reflect the combined availability of sardine and anchovy in the diet (see Results). We took PC2 as an Availability Index for sardine and anchovy and compared it to annual estimates of the spawner biomass of sardine and anchovy west of Cape Agulhas (Augustyn et al., 2018, Department of Agriculture, Forestry and Fisheries) by calculating the difference between the z-score normalized PC2 and the z-score normalized combined spawner biomass. We fit a GAM of the form Yi = α + S(Xi) + εi, where Yi is the value for the residual difference of the two z-scores in year i; α is the intercept; S(Xi) is a non-parametric smoothing function, specifying the effect of the year Xi; and εi ∼ N(0, σ2) is the residual error. We used thin plate regression splines and allowed the degrees of freedom of the smoothing function to be selected automatically by generalized cross-validation (GCV), with the option for the function to be linear [e.g. S(Xi) = β × Xi, where β is the slope]. Relationship of seabird parameters to forage indices Last, in order to investigate the influence of forage availability on South African seabirds, we related PC2 to nine time-series of population statistics: annual nesting attempts of (1) Cape gannets (1980–2015), (2) African penguins (1997–2015), (3) Cape cormorants (1985–2015) and (4) greater crested terns (1984–2015) (Crawford, 2009; Crawford, et al. 2011,, 2014,, 2016; Makhado, et al. 2016); (5) breeding success (chicks fledged per pair) of Cape gannets at Malgas Island (1988–2015; Grémillet, et al., 2008; updated); survival of adult African penguins at (6) Dassen and (7) Robben islands (1994–2011; Sherley, et al., 2014); and number of (8) adult and (9) immature African penguins moulting at Dassen Island (1994–2009; Robinson, 2013) [for African penguin, Cape gannet and greater crested tern, the numbers at all colonies in the region were used; for Cape cormorants, the numbers at eight well-monitored colonies in the region were used: Lambert’s Bay, Malgas, Jutten, Meeuw, Schaapen, Vondeling, Dassen, and Robben islands (Crawford et al., 2016)]. We selected these time-series (Supplementary Table S1 and Figure S1) because they reflect seabird abundance (i.e. nesting effort) as well as vital rates that drive population size (i.e. breeding success, survival, and counts of immature birds as a proxy for recruitment). We similarly compared them with estimates of the combined spawner biomass of anchovy and sardine west of Cape Agulhas. Seabirds often show non-linear responses to the abundance or availability of their prey (e.g. Cury et al., 2011; Sherley et al., 2017). We therefore used generalized additive models (GAMs) to explore the form of the relationships between the seabird response variables and PC2 or biomass. We initially used the “gamm” function from the mgcv library (version 1.8-22; Wood, 2017) for R (version 3.4.3) to fit a model of the form Yi = α + S(Xi) + εi, where Yi is the value for a seabird performance measure in year i; α is the intercept; S(Xi) is the non-parametric smoothing function, specifying the effect of the PC or biomass on each seabird performance measure; and other parameters are as described above. For each model, we assessed residual plots for normality, heterogeneity of variance and an absence of autocorrelation (using the “acf” function in R). In one case (number of Cape cormorant nests), the normality assumption was violated. We thus computed the natural log of the cormorant numbers, refit the models, and re-examined the residual plots. For four data series, the normalized residuals from the initial model demonstrated significant autocorrelation. We therefore refit the models specifying an error structure with first-order autocorrelation using the “corAR1” function from the nlme library (version 3.1-131; Pinheiro et al., 2017) for R. We then re-examined the residual plots to ensure an absence of autocorrelation in the normalized residuals of the models used for inference (Supplementary Figure S2). Results Diet composition Across the 38 breeding seasons (1978/1979–2015/2016), totals of 11 197 and 20 422 regurgitations were collected at Lambert’s Bay and Malgas Island, respectively, or averages of c. 300 and 540 per season. The relative contributions by mass of the main prey categories to the diets of Cape gannets at Lambert’s Bay and Malgas Island during the breeding season are shown in Figure 2 and Supplementary Table S2. At both localities, the proportion of sardine in the diet increased substantially after 1983 and then remained high until the end of the twentieth century. However, sardine fell to low levels from 2004–2008 and 2013–2014. The proportion of anchovy was high in most years from 1978 to 1988 but then decreased and showed substantial fluctuations. The proportion of saury increased after 2002 and hake was important at Malgas Island in the early 1980s and from 2004 to 2009. Figure 2. Open in new tabDownload slide The contributions of sardine, anchovy, hakes, saury, and other species to the diet of Cape gannets (% mass) at Lambert’s Bay and Malgas Island during the breeding season, 1978/1979–2015/2016. The x-axis refers to the year in which the breeding season was initiated. Values for Lambert’s Bay in 2006 were imputed (see Supplementary data for details). Figure 2. Open in new tabDownload slide The contributions of sardine, anchovy, hakes, saury, and other species to the diet of Cape gannets (% mass) at Lambert’s Bay and Malgas Island during the breeding season, 1978/1979–2015/2016. The x-axis refers to the year in which the breeding season was initiated. Values for Lambert’s Bay in 2006 were imputed (see Supplementary data for details). Forage indices The multivariate analysis provided two significant PCs (Table 1). PC1 explained 43% of the variance and increased from negative values in the late 1970s and early 1980s to mostly positive values after the mid-1980s (Figure 3). We interpret PC1 as an alternation between sardine and anchovy in the gannet diet. All 14 sardine loadings had positive values, whereas 13 of the 14 anchovy loadings were negative and the 14th was only marginally positive and less than all sardine loadings (Table 2). Figure 3. Open in new tabDownload slide The values for PC1 and PC2 (the FAI) obtained from analysis of monthly Cape gannet diet at Lambert’s Bay and Malgas Island during the breeding season, 1978/1979–2015/2016. The x-axis indicates the year in which the breeding season started. Figure 3. Open in new tabDownload slide The values for PC1 and PC2 (the FAI) obtained from analysis of monthly Cape gannet diet at Lambert’s Bay and Malgas Island during the breeding season, 1978/1979–2015/2016. The x-axis indicates the year in which the breeding season started. Table 1. Results of principal component analysis on sardine and anchovy in the diet of Cape gannets at Lambert’s Bay and Malgas Island in their September–March breeding season, 1978/1979–2015/2016. Principle component Eigenvalue Proportion of variance explained 1 12.04 0.43 2 3.97 0.14 3 2.02 0.07 Principle component Eigenvalue Proportion of variance explained 1 12.04 0.43 2 3.97 0.14 3 2.02 0.07 Open in new tab Table 1. Results of principal component analysis on sardine and anchovy in the diet of Cape gannets at Lambert’s Bay and Malgas Island in their September–March breeding season, 1978/1979–2015/2016. Principle component Eigenvalue Proportion of variance explained 1 12.04 0.43 2 3.97 0.14 3 2.02 0.07 Principle component Eigenvalue Proportion of variance explained 1 12.04 0.43 2 3.97 0.14 3 2.02 0.07 Open in new tab Table 2. Variable loadings for principal components 1 and 2. Variables Principle components 1 2 Sardine, Lambert’s Bay, September 0.19 0.09 Sardine, Lambert’s Bay, October 0.22 0.11 Sardine, Lambert’s Bay, November 0.20 −0.09 Sardine, Lambert’s Bay, December 0.21 −0.08 Sardine, Lambert’s Bay, January 0.22 0.13 Sardine, Lambert’s Bay, February 0.20 0.16 Sardine, Lambert’s Bay, March 0.16 0.21 Sardine, Malgas Island, September 0.13 0.31 Sardine, Malgas Island, October 0.24 0.05 Sardine, Malgas Island, November 0.23 0.16 Sardine, Malgas Island, December 0.24 0.11 Sardine, Malgas Island, January 0.20 0.25 Sardine, Malgas Island, February 0.17 0.32 Sardine, Malgas Island, March 0.06 0.40 Anchovy, Lambert’s Bay, September −0.14 0.08 Anchovy, Lambert’s Bay, October −0.19 0.03 Anchovy, Lambert’s Bay, November −0.17 0.26 Anchovy, Lambert’s Bay, December −0.19 0.29 Anchovy, Lambert’s Bay, January −0.21 0.17 Anchovy, Lambert’s Bay, February −0.18 0.11 Anchovy, Lambert’s Bay, March −0.16 −0.01 Anchovy, Malgas Island, September −0.18 0.03 Anchovy, Malgas Island, October −0.23 0.15 Anchovy, Malgas Island, November −0.24 0.13 Anchovy, Malgas Island, December −0.17 0.25 Anchovy, Malgas Island, January −0.17 0.25 Anchovy, Malgas Island, February −0.18 0.03 Anchovy, Malgas Island, March 0.04 −0.24 Variables Principle components 1 2 Sardine, Lambert’s Bay, September 0.19 0.09 Sardine, Lambert’s Bay, October 0.22 0.11 Sardine, Lambert’s Bay, November 0.20 −0.09 Sardine, Lambert’s Bay, December 0.21 −0.08 Sardine, Lambert’s Bay, January 0.22 0.13 Sardine, Lambert’s Bay, February 0.20 0.16 Sardine, Lambert’s Bay, March 0.16 0.21 Sardine, Malgas Island, September 0.13 0.31 Sardine, Malgas Island, October 0.24 0.05 Sardine, Malgas Island, November 0.23 0.16 Sardine, Malgas Island, December 0.24 0.11 Sardine, Malgas Island, January 0.20 0.25 Sardine, Malgas Island, February 0.17 0.32 Sardine, Malgas Island, March 0.06 0.40 Anchovy, Lambert’s Bay, September −0.14 0.08 Anchovy, Lambert’s Bay, October −0.19 0.03 Anchovy, Lambert’s Bay, November −0.17 0.26 Anchovy, Lambert’s Bay, December −0.19 0.29 Anchovy, Lambert’s Bay, January −0.21 0.17 Anchovy, Lambert’s Bay, February −0.18 0.11 Anchovy, Lambert’s Bay, March −0.16 −0.01 Anchovy, Malgas Island, September −0.18 0.03 Anchovy, Malgas Island, October −0.23 0.15 Anchovy, Malgas Island, November −0.24 0.13 Anchovy, Malgas Island, December −0.17 0.25 Anchovy, Malgas Island, January −0.17 0.25 Anchovy, Malgas Island, February −0.18 0.03 Anchovy, Malgas Island, March 0.04 −0.24 Open in new tab Table 2. Variable loadings for principal components 1 and 2. Variables Principle components 1 2 Sardine, Lambert’s Bay, September 0.19 0.09 Sardine, Lambert’s Bay, October 0.22 0.11 Sardine, Lambert’s Bay, November 0.20 −0.09 Sardine, Lambert’s Bay, December 0.21 −0.08 Sardine, Lambert’s Bay, January 0.22 0.13 Sardine, Lambert’s Bay, February 0.20 0.16 Sardine, Lambert’s Bay, March 0.16 0.21 Sardine, Malgas Island, September 0.13 0.31 Sardine, Malgas Island, October 0.24 0.05 Sardine, Malgas Island, November 0.23 0.16 Sardine, Malgas Island, December 0.24 0.11 Sardine, Malgas Island, January 0.20 0.25 Sardine, Malgas Island, February 0.17 0.32 Sardine, Malgas Island, March 0.06 0.40 Anchovy, Lambert’s Bay, September −0.14 0.08 Anchovy, Lambert’s Bay, October −0.19 0.03 Anchovy, Lambert’s Bay, November −0.17 0.26 Anchovy, Lambert’s Bay, December −0.19 0.29 Anchovy, Lambert’s Bay, January −0.21 0.17 Anchovy, Lambert’s Bay, February −0.18 0.11 Anchovy, Lambert’s Bay, March −0.16 −0.01 Anchovy, Malgas Island, September −0.18 0.03 Anchovy, Malgas Island, October −0.23 0.15 Anchovy, Malgas Island, November −0.24 0.13 Anchovy, Malgas Island, December −0.17 0.25 Anchovy, Malgas Island, January −0.17 0.25 Anchovy, Malgas Island, February −0.18 0.03 Anchovy, Malgas Island, March 0.04 −0.24 Variables Principle components 1 2 Sardine, Lambert’s Bay, September 0.19 0.09 Sardine, Lambert’s Bay, October 0.22 0.11 Sardine, Lambert’s Bay, November 0.20 −0.09 Sardine, Lambert’s Bay, December 0.21 −0.08 Sardine, Lambert’s Bay, January 0.22 0.13 Sardine, Lambert’s Bay, February 0.20 0.16 Sardine, Lambert’s Bay, March 0.16 0.21 Sardine, Malgas Island, September 0.13 0.31 Sardine, Malgas Island, October 0.24 0.05 Sardine, Malgas Island, November 0.23 0.16 Sardine, Malgas Island, December 0.24 0.11 Sardine, Malgas Island, January 0.20 0.25 Sardine, Malgas Island, February 0.17 0.32 Sardine, Malgas Island, March 0.06 0.40 Anchovy, Lambert’s Bay, September −0.14 0.08 Anchovy, Lambert’s Bay, October −0.19 0.03 Anchovy, Lambert’s Bay, November −0.17 0.26 Anchovy, Lambert’s Bay, December −0.19 0.29 Anchovy, Lambert’s Bay, January −0.21 0.17 Anchovy, Lambert’s Bay, February −0.18 0.11 Anchovy, Lambert’s Bay, March −0.16 −0.01 Anchovy, Malgas Island, September −0.18 0.03 Anchovy, Malgas Island, October −0.23 0.15 Anchovy, Malgas Island, November −0.24 0.13 Anchovy, Malgas Island, December −0.17 0.25 Anchovy, Malgas Island, January −0.17 0.25 Anchovy, Malgas Island, February −0.18 0.03 Anchovy, Malgas Island, March 0.04 −0.24 Open in new tab PC2 explained 14% of the variance. It fluctuated around 0 from 1978 to 1985, 2 from 1986 to 2003 and −2 from 2004 to 2015 (Figure 3). For PC2, 24 of the 28 loadings for sardine and anchovy had positive values, 18 of which were >0.1 (Table 2). We therefore considered PC2 as a FAI, an index of the availability of anchovy and sardine to seabirds. The GAM used to examine the relationship between the biomass of sardine and anchovy and the FAI showed a non-linear trend, with a clear transition from mainly positive residuals before 2000 to mainly negative residuals thereafter [Figure 4; effective degrees of freedom (edf) = 5.33, F = 10.2, p < 0.001]. Figure 4. Open in new tabDownload slide Results of generalized additive modelling of the change over time in the relationship between PC2, the FAI, and the combined spawner biomass of sardine and anchovy west of Cape Agulhas. Diamonds show residual differences between z-score normalized PC2 values and z-score normalized estimates of forage fish biomass (residual = ZFAI − Zbiomass). On the y-axis, s[x, y] indicates the smoothing term, where x is the explanatory variable and y is the estimated degrees of freedom of the smoothing term. The grey shading shows pointwise 95% confidence intervals. Figure 4. Open in new tabDownload slide Results of generalized additive modelling of the change over time in the relationship between PC2, the FAI, and the combined spawner biomass of sardine and anchovy west of Cape Agulhas. Diamonds show residual differences between z-score normalized PC2 values and z-score normalized estimates of forage fish biomass (residual = ZFAI − Zbiomass). On the y-axis, s[x, y] indicates the smoothing term, where x is the explanatory variable and y is the estimated degrees of freedom of the smoothing term. The grey shading shows pointwise 95% confidence intervals. Relationship of seabird parameters to forage indices GAMs provided evidence of non-linear relationships between the FAI and the breeding effort of both Cape gannets (edf = 4.40, F = 17.7, p < 0.001) and Cape cormorants (edf = 2.22, F = 3.75, p = 0.025). For gannets, the numbers breeding increased when PC2 (which increases as anchovy and sardine contribute more to the diet) was >c. − 1 (Figure 5a) and for cormorants when it was >c. 1 (Figure 5b). Additionally, the survival of adult African penguins showed linear relationships with PC2 at both Dassen (edf = 1, F = 24.7, p < 0.001) and Robben (edf = 1, F = 16.5, p < 0.001) islands, with survival rates generally lower when PC2 was negative than positive and decreasing markedly when PC2 was < c. −1.5 (Figure 5c and d). The remaining five seabird parameters showed non-significant linear relationships with PC2: four showed positive trends, with only the number of breeding greater crested terns having a weak negative trend (Supplementary Figure S3b). However, the breeding success of gannets at Malgas Island was only marginally not significant (p = 0.06) and was much more likely to be poor for negative than for positive values of PC2 (Supplementary Figure S3c). No seabird parameter was significantly related to the combined biomass of anchovy and sardine (Supplementary Figure S4). Figure 5. Open in new tabDownload slide Results of generalized additive modelling of the effects of PC2 on four indicators of seabird performance in South Africa. PC2, the Forage Availability Index (FAI), represents the combined contribution of sardine and anchovy to the diet of Cape gannets. The indicators of seabird performance are (a) number of nests at which Cape gannets were breeding between 1978/1979 and 2015/2016; (b) number of nests at which Cape cormorants were breeding between 1978/1979 and 2015/2016; (c) and (d) apparent survival of adult African penguins at Dassen and Robben islands, respectively, between 1994/1995 and 2011/2012. On the y-axis, s[x, y] indicates the smoothing term, where x is the explanatory variable and y is the estimated degrees of freedom of the smoothing term. The grey shading shows pointwise 95% confidence intervals and diamonds show the partial residuals around the significant covariate effects. Figure 5. Open in new tabDownload slide Results of generalized additive modelling of the effects of PC2 on four indicators of seabird performance in South Africa. PC2, the Forage Availability Index (FAI), represents the combined contribution of sardine and anchovy to the diet of Cape gannets. The indicators of seabird performance are (a) number of nests at which Cape gannets were breeding between 1978/1979 and 2015/2016; (b) number of nests at which Cape cormorants were breeding between 1978/1979 and 2015/2016; (c) and (d) apparent survival of adult African penguins at Dassen and Robben islands, respectively, between 1994/1995 and 2011/2012. On the y-axis, s[x, y] indicates the smoothing term, where x is the explanatory variable and y is the estimated degrees of freedom of the smoothing term. The grey shading shows pointwise 95% confidence intervals and diamonds show the partial residuals around the significant covariate effects. Discussion The multivariate analysis of gannet diet identified two significant PCs: PC1 appeared to reflect prey switching between the two high-quality forage species, whereas PC2 seemed to contrast good and poor prey regimes for anchovy and sardine combined. Consequently, we took PC2 to be an index of the availability of these fishes to seabirds off northwest South Africa, the Forage Availability Index (FAI). Its use in this manner was corroborated by clear positive relationships between PC2 and five independent measures of seabird performance that contrasted an absence of such relationships with biomass. Nutritious and “junk” prey In one study, the energy contents of sardine, anchovy, saury, and hakes, the main food of Cape gannets, were respectively 8.59, 6.74, 6.20 and 4.07 kJ g−1 (Batchelor and Ross, 1984). A later study similarly recorded reasonably high values of 6.59 and 6.03 kJ g−1 for sardine and anchovy, respectively (Balmelli and Wickens, 1994). On account of their higher energy content relative to other species, sardine and anchovy have often been regarded as preferred prey for Cape gannets (Berruti et al., 1993; Adams and Klages, 1999; Crawford et al., 2007,, 2014; Green et al., 2015). In contrast, hakes eaten by Cape gannets are mostly fishery discards and have been termed “junk food”, owing to their lower calorific content (Grémillet et al., 2008). Gannet chicks fed sardine also consumed less food and had higher fledging masses than those fed hakes (Batchelor and Ross, 1984). Foraging effort of Cape gannets increased and nest attendance decreased with reduced consumption of sardine and anchovy, whereas adult body condition was negatively impacted by increases of hakes in their diet (Cohen et al., 2014). During pronounced scarcity of sardine and anchovy in the mid-2000s, hake discards proved inadequate for many gannets to rear chicks (Grémillet et al., 2008), although appeared to be sufficient to maintain adult survival (Distiller et al., 2012). The other main alternate prey species is saury, which off western South Africa occurs outside the oceanic front in waters of 18–22°C roughly 80 km or more from the coast (Dudley et al., 1985). Except from January to March, when both the front and saury move inshore, its distribution requires gannets to expend considerable energy to access it (Berruti et al., 1993). In South Africa, condition of sardine deteriorated in the 2000s (Ndjaula et al., 2013). Elsewhere, a decreased condition of preferred prey species led to poor seabird performance (Wanless et al. 2005). Forage Availability Index and seabird performance metrics Whereas significant relationships with the FAI were obtained for numbers of Cape gannets and Cape cormorants that bred, this was not the case for African penguins or greater crested terns, or for numbers of adult and immature penguins moulting at Dassen Island, although all three measures of penguin abundance showed positive trends with the FAI. Unlike Cape gannets and Cape cormorants, which breed in the austral spring and summer, in South Africa many African penguins and greater crested terns breed in autumn and winter outside the September–March period used to obtain the FAI (Crawford et al., 2002,, 2013). African penguins may moult at colonies other than those at which they breed and may feed some distance away from their colony when not breeding (Whittington et al., 2005). The different ecology and relatively low energy requirements of greater crested terns have allowed them to sustain high survival (Payo-Payo et al., 2018) and successfully provision chicks on young-of-the-year anchovy and other small fish species (Gaglio et al., 2018a, b). At both Dassen and Robben islands, survival of adult African penguins was significantly related to the FAI. At these localities, penguins mostly moult from September–January (Underhill and Crawford, 1999; Wolfaardt et al., 2009). Moult is a critical time for African penguins as they remain onshore and fast for about 21 days while they replace their entire plumage; they must fatten before and after moult to build up and replenish their energy reserves (Randall et al., 1986). Hence, their survival may be influenced by food availability outside their breeding season (Sherley et al., 2014) but within the spring–summer period used to derive the FAI. The validity of the FAI as a measure of epipelagic availability of sardine and anchovy off northwest South Africa is supported by other observations of poor performance by, or unusual behaviour of, seabirds in that region, which followed the marked decrease of this index early in the twenty-first century (see Supplementary data). Forage availability and ecosystem management of seabirds The large decrease of the FAI in the early 2000s (Figure 3) followed prior shifts to the southeast of spawning sardine and anchovy, which may have been caused by altered environmental conditions or intense localized fishing (Roy et al., 2007; Coetzee et al., 2008). Although biomass of anchovy off South Africa’s west coast remained high after 2003 (van der Sleen et al., 2018), it appears then to have been less available to gannets and other seabirds during spring and summer than previously. In particular, the switch from mainly positive residuals to mostly negative residuals in the difference between the FAI and forage fish biomass indicates that the amount of sardine and anchovy that gannets were able to capture for a given level of biomass dropped abruptly at the turn of the recent century (Figure 4). The decreased availability of these prey species may have been caused by the observed geographic shift in their distribution, by them assuming a deeper position in the water column or a combination of both mechanisms. In a modelling study, depth of prey primarily determined foraging success of Peruvian boobies Sula variegata and guanay cormorants Phalacrocorax bougainvilliorum, which feed mainly on Peruvian anchoveta Engraulis ringens in the Humboldt upwelling system off western South America (Boyd et al., 2017). In that system, Peruvian boobies and guanay cormorants are the ecological equivalents of Cape gannets and Cape cormorants in the BUS (Crawford et al., 2006). Since 2001, South Africa’s purse-seine fishery has not taken its allowable catch of anchovy and in recent years, the extent of this under-catch has increased (DAFF, 2016), suggesting that availability of anchovy to the fishery also decreased. The recent low forage availability meant that, from 2004 to 2015, the FAI only once (2008) rose above the level of −1, at which the relevant GAM predicted numbers of Cape gannets breeding north of Cape Town would increase (Figure 5a), whereas between 1978 and 2003, it only fell below that threshold in 1984 (Figure 3). Cape cormorants had a higher threshold (c. 1, Figure 5b) at which numbers breeding were predicted to increase. This was exceeded in 13 of 17 years between 1986 and 2002 but not before or after that period (Figure 3). The higher threshold for Cape cormorants probably results from the fact that, unlike Cape gannets, they are mostly unable to access alternative food such as saury and hake offal. When the biomass west of Cape Agulhas of sardine spawners fell below c. 25% of its maximum observed value, survival of adult African penguins at Robben Island decreased markedly (Robinson et al., 2015). This was the case at both Dassen and Robben islands in most years when the FAI had a negative value (Figure 5). Similarly to seabirds in other ecosystems, breeding success of Cape gannets and African penguins in the southern BUS became more variable and on average decreased when the combined biomass of sardine and anchovy was less than about one-third of its maximum observed value (Cury et al., 2011). For Cape gannets at Malgas Island, this also happened when the FAI had negative values (Supplementary Figure S3c). Thresholds provide one means to implement an ecosystem approach to fisheries (EAF) and may relate to allowable by-catch mortality (e.g. Rollinson et al., 2017) or escapement levels for forage resources (e.g. Cury et al., 2011). In addition to linking prey abundance with seabird performance (e.g. Cury et al., 2011; Robinson et al., 2015), the development of indices of food availability will assist in identifying forage thresholds through accounting for portions of food resources that are unavailable to predators. In the southern BUS, an EAF should aim to maintain a positive FAI. Although implementation of an EAF may have short-term costs for forage fisheries, its non-application could disadvantage other users of these resources such as burgeoning marine ecotourism industries (e.g. Lewis et al., 2012) and fisheries exploiting species at higher trophic levels. Furthermore, failure to adopt an EAF may lead to irreversible ecosystem change and longer-term loss of ecosystem services, as was demonstrated for the northern BUS (Roux et al., 2013). In contrast, meaningful ecosystem-based management of forage resources in the California upwelling system has benefitted predators there (Ainley et al., 2018). In summary, the FAI for nutritious forage fishes in the epipelagic zone off northwest South Africa explained several of the trends in seabird populations and demographic parameters observed in this region. Importantly, these included large decreases in the number of breeding Cape gannets and Cape cormorants and in survival of African penguins at their most important colonies. In doing so, the FAI highlights the need to understand changes in the availability (as opposed to abundance) of prey species when identifying thresholds to be used in an EAF aimed at ensuring healthy ecosystem functioning. That Cape gannet diet produced such a useful index resulted in large measure from the species’ ability to switch between prey types and its wide foraging range. Acknowledgements This article is an output of the California Benguela Joint Investigation (Cal-BenJI). Funding for this collaboration was provided by NSF OCE-1434732 and OCE-1434530. We thank South Africa’s Department of Environmental Affairs for supporting this research and C. van der Lingen and J. Coetzee (Department of Agriculture, Forestry and Fisheries) for data on the biomass of anchovy and sardine. We thank S. C. Votier and two anonymous reviewers for most helpful comments on an initial draft of the manuscript, as well as all who assisted with collection of the data and CapeNature, South African National Parks and South African Defence Force for permission to visit colonies and logistical support. RJMC and ABM thank South Africa’s National Research Foundation for incentive funding (grant numbers 103359 and 103388, respectively). RBS was supported by the Leiden Conservation Foundation, the Bristol Zoological Society, and the Zoological Society of San Diego. References Adams N. J. , Klages N. T. W. 1999 . Foraging effort and prey choice in Cape gannets . South African Journal of Marine Science , 21 : 157 – 163 . Google Scholar Crossref Search ADS WorldCat Ainley D. G. , Santora J. A. , Capitolo P. J. , Field J. C. , Beck J. N. , Carle R. D. , Donnelly-Greenan E. , et al. . 2018 . Ecosystem-based management affecting Brandt’s Cormorant resources and populations in the central California Current region . Biological Conservation , 217 : 407 – 418 . Google Scholar Crossref Search ADS WorldCat Augustyn J. , Cockcroft A. , Kerwath S. , Lamberth S. , Githaiga-Mwicigi J. , Pitcher G. , Roberts M. , et al. . 2018 . South Africa. In Climate Change Impacts on Fisheries and Aquaculture: A Global Analysis , pp. 479 – 522 . Ed. by Phillips B. , Pérez-Ramírez M. John Wiley & Sons Ltd, Hoboken, USA . Google Preview WorldCat COPAC Balmelli W. , Wickens P. A. 1994 . Estimates of daily ration for the South African (Cape) fur seal . South African Journal of Marine Science , 14 : 151 – 157 . Google Scholar Crossref Search ADS WorldCat Batchelor A. L. , Ross G. J. B. 1984 . The diet and implications of dietary change of Cape Gannets on Bird Island, Algoa Bay . Ostrich , 55 : 45 – 63 . Google Scholar Crossref Search ADS WorldCat Berruti A. , Underhill L. G. , Shelton P. A. , Moloney C. , Crawford R. J. M. 1993 . Seasonal and interannual variation in the diet of two colonies of the Cape Gannet Morus capensis between 1977/1978 and 1989 . Colonial Waterbirds , 16 : 158 – 175 . Google Scholar Crossref Search ADS WorldCat BirdLife International. 2017 . IUCN Red List for birds. http://www.birdlife.org (last accessed 20 December 2017). Blamey L. , Shannon L. J. , Bolton J. J. , Crawford R. J. M. , Dufois F. , Evers-King H. , Griffiths C. L. , et al. . 2015 . Ecosystem change in the southern Benguela and the underlying processes . Journal of Marine Systems , 144 : 9 – 29 . Google Scholar Crossref Search ADS WorldCat Boyd C. , Grünbaum D. , Hunt G. L. , Punt A. E. , Weimerskirch H. , Bertrand S. 2017 . Effects of variation in the abundance and distribution of prey on the foraging success of central place foragers . Journal of Applied Ecology , 54 : 1362 – 1372 . Google Scholar Crossref Search ADS WorldCat Coetzee J. C. , van der Lingen C. D. , Hutchings L. , Fairweather T. P. 2008 . Has the fishery contributed to a major shift in the distribution of South African sardine? ICES Journal of Marine Science , 65 : 1676 – 1688 . Google Scholar Crossref Search ADS WorldCat Cohen L. A. , Pichegru L. , Grémillet D. , Coetzee J. , Upfold L. , Ryan P. G. 2014 . Changes in prey availability impact foraging behaviour and fitness of Cape gannets over a decade . Marine Ecology Progress Series , 505 : 281 – 293 . Google Scholar Crossref Search ADS WorldCat Crawford R. J. M. 1980 . Seasonal patterns in South Africa’s Western Cape purse-seine fishery . Journal of Fish Biology , 16 : 649 – 664 . Google Scholar Crossref Search ADS WorldCat Crawford R. J. M. 2003 . Influence of food on numbers breeding, colony size and fidelity to localities of Swift Terns in South Africa’s Western Cape, 1987–2000 . Waterbirds , 26 : 44 – 53 . Google Scholar Crossref Search ADS WorldCat Crawford R. J. M. 2009 . A recent increase of swift terns Thalasseus bergii off South Africa – the possible influence of an altered abundance and distribution of prey . Progress in Oceanography , 83 : 398 – 403 . Google Scholar Crossref Search ADS WorldCat Crawford R. J. M. , Dyer B. M. 1995 . Responses by four seabirds to a fluctuating availability of Cape Anchovy Engraulis capensis off South Africa . Ibis , 137 : 329 – 339 . Google Scholar Crossref Search ADS WorldCat Crawford R. J. M. , Altwegg R. , Barham B. J. , Barham P. J. , Durant J. M. , Dyer B. M. , Geldenhuys D. , et al. . 2011 . Collapse of South Africa’s penguins in the early 21st century: a consideration of food availability . African Journal of Marine Science , 33 : 139 – 156 . Google Scholar Crossref Search ADS WorldCat Crawford R. J. M. , Cooper J. , Dyer B. M. , Upfold L. , Venter A. D. , Whittington P. A. , Williams A. J. , et al. . 2002 . Longevity, inter-colony movements and breeding of Crested Terns in South Africa . Emu – Austral Ornithology , 102 : 1 – 9 . Google Scholar Crossref Search ADS WorldCat Crawford R. J. M. , Dundee B. L. , Dyer B. M. , Klages N. T. W. , Meÿer M. A. , Upfold L. 2007 . Trends in numbers of Cape gannets (Morus capensis), 1956/57–2005/06, with a consideration of the influence of food and other factors . ICES Journal of Marine Science , 64 : 169 – 177 . Google Scholar Crossref Search ADS WorldCat Crawford R. J. M. , Goya E. , Roux J.-P. , Zavalaga C. B. 2006 . Comparison of assemblages and some life-history traits of seabirds in the Humboldt and Benguela systems . African Journal of Marine Science , 28 : 553 – 560 . Google Scholar Crossref Search ADS WorldCat Crawford R. J. M. , Kemper J. , Underhill L. G. 2013 . African Penguin (Spheniscus demersus). In Penguins Natural History and Conservation , pp. 211 – 231 . Ed. by Garcia Borboroglu P. , Boersma P. D. University of Washington Press , Seattle and London . Google Preview WorldCat COPAC Crawford R. J. M. , Makhado A. B. , Oosthuizen W. H. 2018 . Bottom-up and top-down control of the Benguela ecosystem’s seabirds . Journal of Marine Systems , 188 : 133 – 141 . Google Scholar Crossref Search ADS WorldCat Crawford R. J. M. , Makhado A. B. , Waller L. J. , Whittington P. A. 2014 . Winners and losers – responses to recent environmental change by South African seabirds that compete with purse-seine fisheries for food . Ostrich , 85 : 111 – 117 . Google Scholar Crossref Search ADS WorldCat Crawford R. J. M. , Makhado A. B. , Whittington P. A. , Randall R. M. , Oosthuizen W. H. , Waller L. J. 2015 . A changing distribution of seabirds in South Africa – the possible impact of climate and its consequences . Frontiers in Ecology and Evolution , 3 : 1 – 10 . Google Scholar Crossref Search ADS WorldCat Crawford R. J. M. , Randall R. M. , Cook T. R. , Ryan P. G. , Dyer B. M. , Fox R. , Geldenhuys D. , et al. . 2016 . Cape cormorants decrease, move east and adapt foraging strategies following eastward displacement of their main prey . African Journal of Marine Science , 38 : 373 – 383 . Google Scholar Crossref Search ADS WorldCat Croxall J. P. , Butchart S. H. M. , Lascelles B. , Stattersfield A. J. , Sullivan B. , Symes A. , Taylor P. 2012 . Seabird conservation status, threats and priority actions: a global assessment . Bird Conservation International , 22 : 1 – 34 . Google Scholar Crossref Search ADS WorldCat Cury P. M. , Boyd I. L. , Bonhommeau S. , Anker-Nilssen T. , Crawford R. J. M. , Furness R. W. , Mills J. A. , et al. . 2011 . Global seabird response to forage fish depletion – one-third for the birds . Science , 334 : 1703 – 1706 . Google Scholar Crossref Search ADS PubMed WorldCat DAFF (Department of Agriculture, Forestry and Fisheries). 2016 . Status of the South African Marine Fisheries Resources 2016 . DAFF , Cape Town . WorldCat COPAC Distiller G. , Altwegg R. , Crawford R. J. M. , Klages N. T. W. , Barham B. 2012 . Factors affecting adult survival and inter-colony movement at the three South African colonies of Cape gannet . Marine Ecology Progress Series , 461 : 245 – 255 . Google Scholar Crossref Search ADS WorldCat Dudley S. F. J. , Field J. G. , Shelton P. A. 1985 . Distribution and abundance of eggs, larvae and early juveniles of saury Scomberesox saurus sconbroides (Richardson) off the south-western Cape, South Africa, 1977/78 . South African Journal of Marine Science , 3 : 229 – 237 . Google Scholar Crossref Search ADS WorldCat Gaglio D. , Cook T. R. , McInnes A. , Sherley R. B. , Ryan P. G. 2018a . Foraging plasticity in seabirds: a non-invasive study of the diet of greater crested terns breeding in the Benguela Region . PLoS One , 13 : e0190444. Google Scholar Crossref Search ADS WorldCat Gaglio D. , Sherley R. B. , Ryan P. G. , Cook T. R. 2018b . A non-invasive approach to estimate the energetic requirements of an increasing seabird population in a perturbed marine ecosystem . Scientific Reports , 8 : 8343 . Google Scholar Crossref Search ADS WorldCat Green D. B. , Klages N. T. W. , Crawford R. J. M. , Coetzee J. C. , Dyer B. M. , Rishworth G. M. , Pistorius P. A. 2015 . Dietary change in Cape gannets reflects distributional and demographic shifts in two South African commercial fish stocks . ICES Journal of Marine Science , 72 : 771 – 781 . Google Scholar Crossref Search ADS WorldCat Grémillet D. , Pichegru L. , Kuntz G. , Woakes A. G. , Wilkinson S. , Crawford R. J. M. , Ryan P. G. 2008 . A junk-food hypothesis for gannets feeding on fishery waste . Proceedings of the Royal Society, London Biological Series , 18 : 1 – 8 . WorldCat Hockey P. A. R. , Dean W. R. J. , Ryan P. G. (Eds). 2005 . Roberts Birds of Southern Africa , 7th edn. John Voelcker Bird Book Fund , Cape Town . Google Preview WorldCat COPAC Jolliffe I. T. 2002 . Principal Component Analysis , 2nd edn. Springer-Verlag , New York . Google Preview WorldCat COPAC Lamont T. , García-Reyes M. , Bograd S. J. , van der Lingen C. D. , Sydeman W. J. 2018 . Upwelling indices for comparative ecosystem studies: variability in the Benguela upwelling system . Journal of Marine Systems , 188 : 3 – 16 . Google Scholar Crossref Search ADS WorldCat Leatherwood S. , Reeves R. R. 1983 . The Sierra Club Handbook of Whales and Dolphins . Sierra Club Books , San Francisco, CA . Google Preview WorldCat COPAC Lewis S. E. F. , Turpie J. K. , Ryan P. G. 2012 . Are African penguins worth saving? The ecotourism value of the Boulders Beach colony . African Journal of Marine Science , 34 : 497 – 504 . Google Scholar Crossref Search ADS WorldCat Makhado A. B. , Crawford R. J. M. , Dyer B. M. , Upfold L. 2016 . Trends in the contribution of potential nests to the overall count of active nests of African penguins at Robben and Dassen islands . Department of Environmental Affairs, Branch Oceans & Coasts, Top Predator Working Group, Seabird Technical Team 2016 , 1 : 1 – 4 . WorldCat Ndjaula H. O. , Gerow K. G. , van der Lingen C. D. , Moloney C. L. , Jarre A. 2013 . Establishing a baseline for evaluating changes in body condition and population dynamics of sardine (Sardinops sagax) in the southern Benguela ecosystem . Fisheries Research , 147 : 253 – 263 . Google Scholar Crossref Search ADS WorldCat Payo-Payo A. , Sanz-Aguilar A. , Gaglio D. , Sherley R. B. , Cook T. R. , Altwegg R. , Ryan P. G. 2018 . Survival estimates for the greater crested tern Thalasseus bergii in southern Africa . African Journal of Marine Science , 40 : 43 – 50 . Google Scholar Crossref Search ADS WorldCat Pinheiro J. , Bates D. , DebRoy S. , Sarkar D. and R Core Team. 2017 . nlme: Linear and nonlinear mixed effects models. R package version 3.1-131. https://CRAN.R-project.org/package=nlme (last accessed 31 July 2018). Randall R. M. , Randall B. M. , Cooper J. , Frost P. G. H. 1986 . A new census method for penguins tested on Jackass Penguins Spheniscus demersus . Ostrich , 57 : 211 – 215 . Google Scholar Crossref Search ADS WorldCat Robinson W. M. L. 2013 . Modelling the impact of the South African small pelagic fishery on African penguin dynamics. PhD thesis, University of Cape Town. Robinson W. M. L. , Butterworth D. S. , Plaganyi E. E. 2015 . Quantifying the projected impact of the South African sardine fishery on the Robben Island penguin colony . ICES Journal of Marine Science , 72 : 1822 – 1833 . Google Scholar Crossref Search ADS WorldCat Rollinson D. P. , Wanless R. M. , Ryan P. G. 2017 . Patterns and trends in seabird bycatch in the pelagic longline fishery off South Africa . African Journal of Marine Science , 39 : 9 – 25 . Google Scholar Crossref Search ADS WorldCat Roux J.-P. , van der Lingen C. D. , Gibbons M. J. , Moroff N. E. , Shannon L. J. , Smith A. D. M. , Cury P. M. 2013 . Jellyfication of marine ecosystems as a likely consequence of overfishing small pelagic fishes: lessons from the Benguela . Bulletin of Marine Science , 89 : 249 – 284 . Google Scholar Crossref Search ADS WorldCat Roy C. , van der Lingen C. D. , Coetzee J. C. , Lutjeharms J. R. E. 2007 . Abrupt environmental shift associated with changes in the distribution of Cape anchovy Engraulis encrasicolus spawners in the southern Benguela . African Journal of Marine Science , 29 : 309 – 319 . Google Scholar Crossref Search ADS WorldCat Shealer D. A. 2002 . Foraging behavior and food of seabirds. In Biology of Marine Birds , pp. 137 – 177 . Ed. by Schreiber E. A. , Burger J. CRC Press , Boca Raton, FL . Google Preview WorldCat COPAC Sherley R. B. , Abadi F. , Ludynia K. , Barham B. J. , Clark A. E. , Altwegg R. 2014 . Age-specific survival and movement among major African Penguin Spheniscus demersus colonies . Ibis , 156 : 716 – 728 . Google Scholar Crossref Search ADS WorldCat Sherley R. B. , Both P. , Underhill L. G. , Ryan P. G. , van Zyl D. , Cockcroft A. C. , Crawford R. J. M. , et al. . 2017 . Defining ecologically relevant scales for spatial protection with long-term data on an endangered seabird and local prey availability . Conservation Biology , 31 : 1312 – 1321 . Google Scholar Crossref Search ADS PubMed WorldCat Sherley R. B. , Underhill L. G. , Barham B. J. , Barham P. J. , Coetzee J. C. , Crawford R. J. M. , Dyer B. M. , et al. . 2013 . Influence of local and regional prey availability on breeding performance of African penguins Spheniscus demersus . Marine Ecology Progress Series , 473 : 291 – 301 . Google Scholar Crossref Search ADS WorldCat Sydeman W. J. , Thompson S. A. , Anker-Nilssen T. , Arimitsu M. , Bennison A. , Bertrand S. , Boersch-Supan P. , et al. . 2017 . Best practices for assessing forage fish fisheries-seabird resource competition . Fisheries Research , 194 : 209 – 221 . Google Scholar Crossref Search ADS WorldCat Underhill L. G. , Crawford R. J. M. 1999 . Season of moult of African penguins at Robben Island, South Africa, and its variation, 1988–1998 . South African Journal of Marine Science , 21 : 437 – 441 . Google Scholar Crossref Search ADS WorldCat Underhill L. G. , Tree A. J. , Oschadleus H. D. , Parker V. 1999 . Review of Ring Recoveries of Waterbirds in Southern Africa . Avian Demography Unit , Cape Town . Google Preview WorldCat COPAC van der Sleen P. , Rykaczewski R. R. , Turley B. D. , Sydeman W. J. , García-Reyes M. , Bograd S. J. , van der Lingen C. D. , et al. . 2018 . Non-stationary responses in anchovy (Engraulis encrasicolus) recruitment to coastal upwelling in the Southern Benguela . Marine Ecology Progress Series , 596 : 155 – 164 . Google Scholar Crossref Search ADS WorldCat Votier S. C. , Furness R. W. , Bearhop S. , Crane J. E. , Caldow R. W. G. , Catry P. , Ensor K. , et al. . 2004 . Changes in fisheries discard rates and seabird communities . Nature , 427 : 727 – 730 . Google Scholar Crossref Search ADS PubMed WorldCat Wanless S. , Harris M. P. , Redman P. , Speakman J. R. 2005 . Low energy values of fish as a probable cause of a major seabird breeding failure in the North Sea . Marine Ecology Progress Series , 294 : 1 – 8 . Google Scholar Crossref Search ADS WorldCat Whittington P. A. , Randall R. M. , Randall B. M. , Wolfaardt A. C. , Crawford R. J. M. , Klages N. T. W. , Bartlett P. A. , et al. . 2005 . Patterns of movement of the African penguin in South Africa and Namibia . African Journal of Marine Science , 27 : 215 – 229 . Google Scholar Crossref Search ADS WorldCat Wolfaardt A. C. , Underhill L. G. , Crawford R. J. M. 2009 . Comparison of moult phenology of African penguins Spheniscus demersus at Robben and Dassen islands . African Journal of Marine Science , 31 : 19 – 29 . Google Scholar Crossref Search ADS WorldCat Wood S. N. 2017 . Generalized Additive Models: An Introduction with R , 2nd edn. Chapman and Hall/CRC, Boca Raton, USA . Google Preview WorldCat COPAC © International Council for the Exploration of the Sea 2019. All rights reserved. For permissions, please email: [email protected] This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)