TY - JOUR
AU - Bush,, Peter
AB - Abstract Maintaining a functionally connected network of high-quality habitat is one of the most effective responses to biodiversity loss. However, the spatial distribution of suitable habitat may shift over time in response to climate change. Taxa such as migratory forest landbirds are already undergoing climate-driven range shifts. Therefore, patches of climate-resilient habitat (also known as “climate refugia”) are especially valuable from a conservation perspective. Here, we performed maximum entropy (Maxent) species distribution modeling to predict suitable and potentially climate-resilient habitat in Nova Scotia, Canada, for 3 migratory forest landbirds: Rusty Blackbird (Euphagus carolinus), Olive-sided Flycatcher (Contopus cooperi), and Canada Warbler (Cardellina canadensis). We used a reverse stepwise elimination technique to identify covariates that influence habitat suitability for the target species at broad scales, including abiotic (topographic control of moisture and nutrient accumulation) and biotic (forest characteristics) covariates. As topography should be relatively unaffected by a changing climate and helps regulate the structure and composition of forest habitat, we posit that the inclusion of appropriate topographic features may support the identification of climate-resilient habitat. Of all covariates, depth to water table was the most important predictor of relative habitat suitability for the Rusty Blackbird and Canada Warbler, with both species showing a strong association with wet areas. Mean canopy height was the most important predictor for the Olive-sided Flycatcher, whereby the species was associated with taller trees. Our models, which comprise the finest-scale species distribution models available for these species in this region, further indicated that, for all species, habitat (1) remains relatively abundant and well distributed in Nova Scotia and (2) is often located in wet lowlands (a climate-resilient topographic landform). These findings suggest that opportunities remain to conserve breeding habitat for these species despite changing temperature and precipitation regimes. RÉSUMÉ Le maintien d’un réseau d’habitats de grande qualité fonctionnellement connectés est l’une des solutions les plus efficaces à la perte de biodiversité. Cependant, la répartition spatiale des habitats propices peut changer dans le temps en réponse aux changements climatiques. Des taxons comme les oiseaux forestiers migrateurs modifient déjà leur répartition en réaction aux changements climatiques. Par conséquent, des îlots d’habitats résilients aux changements climatiques (aussi connus sous le nom de « refuges climatiques ») sont particulièrement précieux du point de vue de la conservation. Nous avons effectué une modélisation de la répartition des espèces par la méthode de l’entropie maximale (Maxent) afin de prédire les habitats propices et potentiellement résilients aux changements climatiques en Nouvelle-Écosse, au Canada, pour trois oiseaux forestiers migrateurs: Euphagus carolinus, Contopus cooperi, et Cardellina canadensis. Nous avons utilisé une technique d’élimination par étapes inversée afin d’identifier les covariables qui influencent la qualité de l’habitat pour ces espèces cibles sur de grandes échelles, dont les covariables abiotiques (contrôle topographique de l’humidité et accumulation des nutriments) et biotiques (caractéristiques de la forêt). Puisque la topographie devrait demeurer relativement intouchée par les changements climatiques et qu’elle contribue à réguler la structure et la composition de l’habitat forestier, nous posons l’hypothèse que l’inclusion d’éléments topographiques appropriés peut appuyer l’identification d’habitats résilients aux changements climatiques. De toutes les covariables, la profondeur de la nappe phréatique était la plus importante pour prédire la qualité relative de l’habitat pour E. carolinus et C. canadensis, chacune de ces espèces présentant une forte association avec les zones humides. La hauteur moyenne de la canopée était la variable explicative la plus importante pour Contopus cooperi, alors que cette espèce était associée aux plus grands arbres. Nos modèles, qui comprennent les modèles de répartition des espèces aux échelles les plus fines disponibles pour ces espèces dans cette région, indiquent également que, pour toutes ces espèces, l’habitat (1) demeure relativement abondant et bien réparti en Nouvelle-Écosse et (2) est souvent situé dans les terres basses humides (une forme de relief résiliente aux changements climatiques). Ces résultats suggèrent qu’il subsiste des opportunités pour conserver l’habitat de reproduction de ces espèces malgré les changements de température et de régimes de précipitations. INTRODUCTION Species are currently declining at an unprecedented rate in modern history (Pimm et al. 1995, Ceballos et al. 2017), with habitat modification, fragmentation, and destruction among the most significant drivers (Millennium Ecosystem Assessment 2005, Crooks et al. 2017). Although many taxa are under threat from anthropogenic activities, migratory birds face unique risks due to a dependence on habitat that spans multiple continents and the significant biological stresses associated with migration itself (Weidensaul 2000). Of those in decline, 3 migratory forest passerines listed as at risk in Canada (see Schedule 1 of the Species at Risk Act; Government of Canada 2002) have been receiving considerable research and conservation attention (e.g., the International Rusty Blackbird Working Group 2013–2015, Ball et al. 2016, Environment Canada 2016 a, b) due to steep population declines: Rusty Blackbird (Euphagus carolinus; 85% decline from 1966 to 2003; Environment Canada 2015); Olive-sided Flycatcher (Contopus cooperi; 79% decline from 1968 to 2006; Environment Canada 2016a); and Canada Warbler (Cardellina canadensis; ~71% decline from 1970 to 2012; Environment Canada 2016b). These losses have been pronounced in the eastern portion of these species’ ranges (Government of Canada 2017). Moreover, climate-driven range shifts have been observed or predicted for these birds (McClure et al. 2012, National Audubon Society 2014, Stralberg et al. 2015). Maintaining and restoring functionally connected networks of high-quality habitat is considered one of the most effective responses to species declines (Noss 1983, Rubio and Saura 2012). However, due to increasing human population growth and the competing interests of many stakeholders, there are limits to the number of natural habitat areas that can be restored and/or conserved. Identifying areas to protect, maintain, or restore is therefore essential to achieving effective habitat conservation for these species. Climate change presents particular challenges to in situ land conservation efforts, as changing temperature and precipitation regimes are expected to result in currently suitable habitat becoming unsuitable over time (Anderson and Ferree 2010, Beier and Brost 2010). Increased wetland drying, insect induced tree mortality, and climate-driven range shifts have already been observed for passerines (McClure et al. 2012, Nogués-Bravo et al. 2012, Stralberg et al. 2015). Land conservation initiatives that protect critical habitat are more likely to be successful at stemming biodiversity loss if they target climate-resilient habitat (sometimes termed “climate refugia”), which has a stronger likelihood of remaining suitable over the long term (Anderson et al. 2012, Beier et al. 2015, Gill et al. 2015). Over the past few decades, species distribution models (SDMs) have emerged as useful tools for delineating high-value habitat (Franklin 1995, Guisan and Zimmermann 2000, Elith and Leathwick 2009). SDMs use occurrence data (i.e. point locations delineating where species have been observed) to generate mathematical representations of species’ distributions in environmental space (i.e. how species respond to environmental covariates), and these are in turn used to predict species’ distributions in geographic space (i.e. maps of relative habitat suitability; Elith and Leathwick 2009). Previous studies have published Canada-wide SDMs for the Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler (e.g., Hache et al. 2014, Stralberg et al. 2015) and a regional SDM for the Canada Warbler in Alberta (Ball et al. 2016). However, these models were completed at a resolution of 100 ha, which is too coarse to guide conservation and acquisition efforts that target individual parcels of land, which are often 10 ha or less in fragmented landscapes. Only one finer-scale model has been completed for the Olive-sided Flycatcher and Canada Warbler, in the Canadian Maritimes (6.25 ha resolution; Westwood et al. 2019). Neither of these regional models accounted for predicted future climate or climate-resilient habitat. Typically, SDM studies that attempt to locate climate refugia (e.g., Franklin 2009b, Stralberg et al. 2015) do so by (1) predicting suitable habitat under current climate conditions, (2) predicting suitable habitat under conditions based on common climate change models, and then (3) identifying overlapping areas between the two. However, the uncertainty associated with climate change predictions can be severe, leading some to question whether the noise exceeds the signal when such a strategy is used (Stralberg et al. 2015). While some SDM studies acknowledge this uncertainty (e.g., Sohl 2014), its effects are rarely tested. In this study, we used maximum entropy modeling (Maxent 3.3 software; Phillips et al. 2006) and a reverse stepwise elimination technique to build predictive, high-resolution, spatially explicit models of climate-resilient breeding habitat for these 3 species in Nova Scotia, Canada. However, rather than attempting to delineate future habitat suitability based on climate predictions, we employed proxy covariates representing enduring features (Anderson and Ferree 2010, Huettman and Gottschalk 2011, Beier et al. 2015). Enduring features are environmental processes and elements considered to be “ecologically resilient” in terms of mediating fundamental biotic and abiotic conditions related to a species’ niche (Holling 1986). These processes and features have previously been defined at global-, meso-, topo-, micro-, and nano-levels, and represent increasingly fine spatial and temporal variation in the delivery of water and energy (Mackay and Lindenmayer 2001). For an SDM at the scale of hundreds of kilometers, covariates are available at the topo-level (i.e. covariates that describe regional topography) and micro-level (i.e. covariates that describe characteristics of forest stands). Some previous researchers have discouraged the inclusion of abiotic factors such as topography in SDMs, as the influence of such features over species distributions tends to be indirect, especially for mobile vertebrate taxa (Franklin 2009a). Although topographic covariates have been used to help predict species distributions in earlier studies, such covariates have generally been employed as surrogates when mapped data representing important biotic features were unavailable. However, the utility of abiotic topography in identifying climate-resilient habitat has nonetheless been noted as certain topographic features (1) have the ability to promote “ecological memory,” whereby ecosystems retain similar structures and functions following disturbance (Holling 1992, Larkin et al. 2006) and (2) will generally be less affected by changing temperature and precipitation regimes than biotic features (Anderson et al. 2012). Therefore, we posit that models trained using appropriate topographic covariates (in conjunction with appropriate micro-level forest/habitat-type covariates) should be better able to identify climate-resilient habitat than models trained using forest covariates alone. Our objectives in this study were to (1) construct SDMs using a combination of topographic and forest covariates to identify potential climate-resilient habitat for Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler in Nova Scotia; (2) elucidate regional habitat associations of these species based on covariates selected according to reverse stepwise elimination; (3) determine whether topographic covariates were useful in predicting the distribution of each species; and (4) identify predicted areas of climate resilience for each species independently and for all 3 species as a suite. The climate resilience–based approach to species distribution modeling presented here should be of theoretical interest to conservation modelers and planners, and our model results should be of practical use to ongoing conservation efforts targeting the Rusty Blackbird, Canada Warbler, and Olive-sided Flycatcher in Nova Scotia. METHODS Study Area Nova Scotia (~45°N, ~63°W) is a maritime province in southeastern Canada, situated on the Atlantic migratory flyway and containing the easternmost breeding habitat of the Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler. The province is characterized by a modified continental climate and exhibits a wide (though not extreme) temperature range, ample precipitation, and great variability in daily weather conditions (Nova Scotia Museum of Natural History 1996b). With a total area of 52,939 km2 (Statistics Canada 2017a), Nova Scotia contains a diverse array of landscapes and an abundance of wetlands, lakes, and rivers (Nova Scotia Museum of Natural History 1996a). Land use/land cover is mixed and includes intact and fragmented forests, coastal barrens, agricultural areas, 2 cities (Halifax and Sydney), and a number of towns (Agriculture and Agri-food Canada 2015, Statistics Canada 2017b). Topography is gently rolling, with elevations ranging from 0 to 520 m above sea level, whereby the highest elevations occur in the northeast. Nova Scotia is situated within the Acadian Forest Ecozone and is characterized by mixed forest species, although conifers dominate in areas where drainage is impeded (Rowe 1972, Neily et al. 2005). Species Occurrence Data Occurrence data for the Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler were obtained from the Atlantic Canada Conservation Data Center. Occurrences primarily comprised records from the Maritime Breeding Bird Atlas (MBBA) database (2006–2010 surveys) but also included observations made by other individuals and research groups. Nearly all observations were made between 2005 and 2013; for Rusty Blackbird, there were 2 additional observations, from 1998 and 2001. Corresponding absence data was not available for these species. To clean the data, we removed occurrences where the bird was not guaranteed to be less than 150 m away from recorded coordinates. To reduce spatial autocorrelation, we applied a spatial distance filter of 1 km (Franklin and Miller 2009); when the distance between points was below this threshold, the point closest to a road was removed. After filtering, 136 Rusty Blackbird observations, 502 Olive-sided Flycatcher observations, and 312 Canada Warbler observations were available for modeling (Figure 1). FIGURE 1. Open in new tabDownload slide Kernel density maps illustrating the relative density of presence points in species datasets used in this study, where darker areas denote higher point densities. Minimum convex polygons that enclose all points are also shown (black outlines on map). Note that (1) kernel density maps were used as bias grids in Maxent modeling and (2) actual occurrence locations are not displayed here due to the sensitive nature of species at risk data. (Inset) Map of Canada with the study area (the province of Nova Scotia) shown in black. FIGURE 1. Open in new tabDownload slide Kernel density maps illustrating the relative density of presence points in species datasets used in this study, where darker areas denote higher point densities. Minimum convex polygons that enclose all points are also shown (black outlines on map). Note that (1) kernel density maps were used as bias grids in Maxent modeling and (2) actual occurrence locations are not displayed here due to the sensitive nature of species at risk data. (Inset) Map of Canada with the study area (the province of Nova Scotia) shown in black. Environmental Data Based on available literature, covariates used in other SDMs (Haché et al. 2014, Westwood et al. 2019), and personal communications with regional experts, we selected 11 covariate layers to include in reverse stepwise elimination (Table 1).These covariates represent environmental processes and features at the topo- and micro-levels (Holling 1986, Mackay and Lidenmayer 2001) and reflect known ecological preferences of the target bird species. In the study region, all 3 species use wet forest habitat (Westwood 2016). The Olive-sided Flycatcher uses edges of predominantly coniferous forests alongside gaps created by wetlands, recent burns, or clearcuts (Altman and Sallabanks 2012). The Canada Warbler uses forested wetlands and wet, shrubby mixedwood forests (Reitsma et al. 2009), and the Rusty Blackbird is a wetland obligate (Avery 2013). TABLE 1. List of covariates included in reverse stepwise elimination for the 3 study species. DEM = Digital elevation model; FID = Forest inventory database; RUBL = Rusty Blackbird; OSFL = Olive-sided Flycatcher; CAWA = Canada Warbler. All base datasets were obtained from the Nova Scotia Department of Natural Resources (NSDNR). Covariate name Base dataset(s) Considered for birds Topographic position index DEM RUBL, OSFL, CAWA Landscape complexity index DEM; wetland inventory RUBL, OSFL, CAWA Depth to water table Depth to water table RUBL, OSFL, CAWA Canopy height AVG FID RUBL, OSFL, CAWA Canopy height STD FID RUBL, OSFL, CAWA Canopy closure AVG FID RUBL, OSFL, CAWA Canopy closure STD FID RUBL, OSFL, CAWA Distance to coniferous stand FID RUBL, OSFL, CAWA Distance to deciduous stand FID CAWA Distance to all height stand FID RUBL, OSFL, CAWA Distance to stand with dead material FID RUBL, OSFL, CAWA Covariate name Base dataset(s) Considered for birds Topographic position index DEM RUBL, OSFL, CAWA Landscape complexity index DEM; wetland inventory RUBL, OSFL, CAWA Depth to water table Depth to water table RUBL, OSFL, CAWA Canopy height AVG FID RUBL, OSFL, CAWA Canopy height STD FID RUBL, OSFL, CAWA Canopy closure AVG FID RUBL, OSFL, CAWA Canopy closure STD FID RUBL, OSFL, CAWA Distance to coniferous stand FID RUBL, OSFL, CAWA Distance to deciduous stand FID CAWA Distance to all height stand FID RUBL, OSFL, CAWA Distance to stand with dead material FID RUBL, OSFL, CAWA Open in new tab TABLE 1. List of covariates included in reverse stepwise elimination for the 3 study species. DEM = Digital elevation model; FID = Forest inventory database; RUBL = Rusty Blackbird; OSFL = Olive-sided Flycatcher; CAWA = Canada Warbler. All base datasets were obtained from the Nova Scotia Department of Natural Resources (NSDNR). Covariate name Base dataset(s) Considered for birds Topographic position index DEM RUBL, OSFL, CAWA Landscape complexity index DEM; wetland inventory RUBL, OSFL, CAWA Depth to water table Depth to water table RUBL, OSFL, CAWA Canopy height AVG FID RUBL, OSFL, CAWA Canopy height STD FID RUBL, OSFL, CAWA Canopy closure AVG FID RUBL, OSFL, CAWA Canopy closure STD FID RUBL, OSFL, CAWA Distance to coniferous stand FID RUBL, OSFL, CAWA Distance to deciduous stand FID CAWA Distance to all height stand FID RUBL, OSFL, CAWA Distance to stand with dead material FID RUBL, OSFL, CAWA Covariate name Base dataset(s) Considered for birds Topographic position index DEM RUBL, OSFL, CAWA Landscape complexity index DEM; wetland inventory RUBL, OSFL, CAWA Depth to water table Depth to water table RUBL, OSFL, CAWA Canopy height AVG FID RUBL, OSFL, CAWA Canopy height STD FID RUBL, OSFL, CAWA Canopy closure AVG FID RUBL, OSFL, CAWA Canopy closure STD FID RUBL, OSFL, CAWA Distance to coniferous stand FID RUBL, OSFL, CAWA Distance to deciduous stand FID CAWA Distance to all height stand FID RUBL, OSFL, CAWA Distance to stand with dead material FID RUBL, OSFL, CAWA Open in new tab All covariate layers were prepared as a stack of ASCII rasters with the same projection (NAD 1983 UTM Zone 20), spatial extent, and cell size (150 m, to match the coarsest resolution of species occurrence data) using ArcGIS 10.3.1 (Environmental Systems Research Institute [ESRI], Redlands, California, USA). To reduce effects of multicollinearity on model outputs, we also ensured that no pair of covariates had a Spearman correlation coefficient larger than |0.6| using SPSS 21 (IBM 2013). Many of the environmental datasets used in this study are maintained and distributed by the Nova Scotia Department of Natural Resources (NSDNR); all of these NSDNR datasets (the use of which are described over subsequent paragraphs) are publicly available through Geonova (https://geonova.novascotia.ca). Topographic covariates. Regional topography (1) creates microclimates with variable temperature and moisture regimes (Anderson et al. 2012) and (2) regulates the accumulation of water and soil nutrients on the landscape. These phenomena influence ecosystem vegetation (Mackay and Lindenmayer 2001) and the ecological resilience of an area (Anderson et al. 2012). Of our 11 candidate covariates, 3 represented topographic features, including topographic position (Topographic Position Index), landscape complexity (Landscape complexity), and a depth to water table (Depth to water table). Each of these topographic covariates were defined according to regional indices specific to Nova Scotia. The topographic position index classifies landform position relative to the surrounding “neighborhood” (i.e. local topographic position). Topographic position affects many biophysical processes, such as soil erosion and deposition, wind exposure, cold air drainage, and hydrological balance (Weiss 2001). Using methods developed by Weiss (2001) and modified by Cooley (2014), we created a topographic position index that classified landform position based on Nova Scotia’s Enhanced Digital Elevation Model (DEM), which is distributed by the Nova Scotia Department of Natural Resources (NSNDR) and was produced in 2006. Specifically, this index classified 5 topographic positions (i.e. valleys, low slopes, mid slopes, up slopes, and ridges; see Appendix). Anderson et al. (2012) define landscape complexity as the number of microclimates present in an area, which is in turn a function of landform variety, wetland density, and elevation range. More complex landscapes typically support greater biodiversity levels and promote climate resilience (Anderson et al. 2012). We built a 150-m2 index delineating relative landscape complexity across Nova Scotia using methodology adapted from Anderson et al. (2012), the provincial DEM, and the provincial wetland inventory (which is maintained and distributed by NSDNR). (See Appendix for a detailed description of GIS methods.) The depth to water table is based on a DEM and hydrographic data (NSDNR; Murphy et al. 2007). This index describes where water is likely to flow and/or accumulate on a landscape. Westwood et al. (2019) found depth to water table to be a strong predictor of distribution for these species. Forest covariates. Forest habitat covariates are believed to have a more direct influence on the suitability of songbird habitat and on the diversity of songbird communities than topography (Franklin 2009a). We included 8 covariates that described characteristics of forest stands, including vertical and horizontal distribution of canopy elements or patch types. Forestry data were obtained from the province-wide Forest Inventory Database (FID; NSDNR). This layer is regularly updated through aerial photograph interpretation and describes land use as well as the structure and composition of vegetation. FID data used in this study were from 1988–2012, with the majority collected between 2003 and 2012. To prepare forest covariates, we converted the original FID polygon layer into raster layers and averaged values within each 150-m cell to characterize average canopy height (Canopy height AVG) and average canopy density (Canopy closure AVG). To characterize heterogeneity of canopy height (Canopy height STD) and canopy density (Canopy closure STD), we calculated the standard deviation of relevant polygon values within those same cells. To delineate availability and distribution of cover types, we measured the Euclidean distance from each cell to the nearest coniferous and deciduous stand (Distance to coniferous stand and Distance to deciduous stand, respectively). Finally, to delineate the availability and distribution of key patch types, we measured the distance between each cell and (1) the nearest all height stand (Distance to all height stand; i.e. stands in which 3 or more distinct canopy layers can be distinguished from aerial imagery; Pannozzo and Colman 2008) and (2) the nearest stand with dead material (Distance to stand with dead material). Maximum Entropy Modeling We used Maxent 3.3 software (Phillips et al. 2006) to model relative habitat suitability for the Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler in Nova Scotia. Maxent is a widely used machine learning SDM method that estimates the relative probability of species presence by comparing environmental conditions at occurrence points to those at 10,000 background points (i.e. locations where the species was not observed, with 10,000 being the standard established by Phillips and Dudik 2008 and repeated in many studies, including Elith et al. 2010a, Merow et al. 2013). Although methods that use presence–absence data are generally considered to be more accurate than presence-only methods, absence data were not available for our target species; therefore, our modeling efforts were limited to presence-only approaches. Furthermore, presence-only methods are most appropriate when data are collected using different protocols (Franklin 2009a), as was the case with our species datasets. Maxent ranks among the top performing presence-only modelling approaches (Elith et al. 2006) and has the additional advantages of being easy to use and relatively robust to small sample sizes and/or spatial errors in occurrence data (Elith et al. 2006, Merow et al. 2013). In building models, we applied a regularization penalty of 1.5 to reduce overfitting, after we noticed that the default regularization penalty of 1 led to unrealistic response curves. All other parameter settings were assigned default values. Correcting for Sample Bias Bias in sampling effort can greatly reduce model accuracy. When sample bias is present, environmental covariates may be assigned importance because they are typical of intensely surveyed areas, not because they represent a real biological relationship (Baldwin 2009, Phillips et al. 2009). This is particularly problematic for presence-only models such as Maxent, as presence-only datasets almost invariably comprise a collection of undesigned, opportunistic, or purposive observations obtained from multiple sources of varying integrity (Franklin 2009a). To compensate for sample bias in bird datasets, in addition to spatial filtering, we created “bias grids,” which modify the location and frequency of background data collection in order to characterize the background sample with similar spatial bias as that which affects occurrence data (Phillips and Dudik 2008). Thus, while the total number of background points (i.e. 10,000) does not change, when a bias grid is used, background points are sampled at a greater density around clusters of presence points (whereas when no bias grid is used, background points are evenly sampled throughout the entire study area). Our recent study found that applying a spatial filter in conjunction with a bias grid was among the most effective methods of reducing the effects of sample bias in Maxent compared to 5 other commonly used bias correction strategies (Bale 2017). We created bias grids by generating a kernel density map of occurrence points for each species using the “Kernel Density with barriers” tool included in the Hawth’s tools extension for ArcGIS (Beyer 2004). Following Elith et al. (2010b) and Fourcade et al. (2014), we adopted a kernel radius of 10 km and normalized output kernel density values between 1 and 20 (Figure 1). Reverse Stepwise Elimination We applied a reverse stepwise elimination technique to identify the most parsimonious subsets of covariates to be included in final models (Hooftman et al. 2015, Wang et al. 2018). For this, we first ran initial models for each species using all covariates (see Table 1), with one exception: due to differing ecological associations, distance to deciduous stand was not included in initials models for the Rusty Blackbird and Olive-sided Flycatcher. Following initial model runs, the covariate that contributed the least predictive power according to permutation importance score (Yost et al. 2008, Baldwin 2009) was identified, and a new model was then run without this covariate. This procedure was repeated until a single covariate remained, yielding a set of n candidate models, where n = the number of covariates included in the initial run. Of these, the most parsimonious model was identified using Akaike’s Information Criterion (AICc) (corrected for small sample sizes), calculated using a Perl script developed for Maxent by Warren and Seifert (2011). Models with the lowest AICc score were adopted as “final models” for each species. Model Evaluation For each species, we evaluated the fit of the final model using Area under the Receiver Operating Curve (AUC) statistics and by comparing expected vs. observed omission rates. AUC is a ranked approach that provides a measure of the likelihood that a randomly selected presence point has a higher suitability score than a randomly selected absence or, in the case of Maxent, background point (Elith et al. 2006, Fourcade et al. 2014). AUC scores range from 0 (no power to discriminate between presence and absence/background) to 1 (perfect discriminatory power). A score of 0.5 indicates that model predictions are no better than random. We reported mean AUC values for 10 cross-validated runs. For each of these runs, we used 90% of occurrence points to train the model and 10% of occurrence points to test the model. Therefore, 2 AUC values were generated for each final model, one based on training data (AUCtrain) and one based on test data (AUCtest). Although AUC is widely used in SDM studies, this metric has been criticized for tending to reward overfit models (Lobo et al. 2008), which often predict training data well and test data poorly (Warren and Seifert 2011; Bale et al. in prep). Therefore, in evaluating model reliability, we also (1) considered the difference between AUCtrain and AUCtest scores (AUCdiff hereafter) and (2) compared expected versus observed omission rates. In comparing expected and observed omission rates, we adopted 2 thresholds: the lowest presence threshold (i.e. LPT) and the 10% presence threshold (10PT) (Pearson et al. 2007). LPT refers to the maximum suitability score for which no presence locations were incorrectly classified as “background,” and 10PT refers to the suitability score at which 10% of presence locations were incorrectly classified as background. For each species, we calculated the LPT and 10PT values of training data (i.e. expected omission rates) and determined how many points were excluded when these thresholds were applied on test data (i.e. observed omission rates). Observed omission rates that are close to 0% for LPT and 10% for 10PT indicate that the model is well calibrated. Creating Binary Suitability Surfaces Results output by Maxent software include a continuous surface of ranked habitat suitability values as well as threshold values that can be used to convert this continuous surface into a binary one that distinguishes among only 2 categories: suitable habitat and unsuitable habitat. To estimate how much of the Nova Scotia landmass is predicted to contain suitable habitat for each of the 3 study species, we applied the maximum training sensitivity plus specificity (MaxSS) threshold. MaxSS is determined by optimizing sensitivity and specificity values and was identified by Liu et al. (2013) as the most robust threshold among the ones included in Maxent outputs. We subsequently combined all 3 binary suitability surfaces into a single map to identify areas of Nova Scotia that may contain suitable habitat for all 3 species. To further assess the utility of topographic covariates in species distribution modelling, we (1) calculated the percent area of suitable habitat (for all 3 species) that was composed of topographic features of interest (i.e. topographic features that showed good predictive power in our models) and (2) compared those values to the percent area that the topographic features of interest comprised across the entire province of Nova Scotia. RESULTS Final models for each species included 4–6 covariates representing both topographic and forest features. (Response curves as well as percent contribution and permutation importance scores for each covariate are listed in Table 2.) Of topographic covariates, the topographic position index and especially the depth to water table layer showed good predictive power in our models. For all species, while AUCtrain and AUCtest values were fairly low, AUCdiff values were also low, and observed omission rates were similar to expected omission rates (Table 3), which indicates that overfitting did not significantly confound our SDMs. Furthermore, for all species, maps of relative habitat suitability indicated that, at a resolution of 150 m2, suitable habitat remains fairly abundant and well distributed across Nova Scotia (Figures 2 and 3, Supplementary Material Figures S1 and S2). When each of the binary habitat suitability surfaces (created using the MaxSS threshold) were compared in an overlay, ~22% of the Nova Scotia landmass was predicted to be suitable for all 3 species (Figure 4, Supplemental Material Figure S3). Of areas suitable for all species, 43% was classified as “valley” by the topographic position index, and 66% was classified as either “valley” or “low-slope.” Conversely, for all of Nova Scotia, these values were 7% (valleys) and 34% (valleys or low-slopes), respectively (Table 4). Moreover, 49% of areas predicted to be suitable for all species was classified as “wet” by the depth to water table index (i.e. the distance between the water table and the earth surface was ≤1 m; Murphy et al. 2007). Conversely, when the depth to water table index was applied across the entire province of Nova Scotia, only 8% of the provincial landmass was classified as “wet” (Table 4). TABLE 2. Covariates included in final models as well as associated percent (%) contribution scores, permutation importance scores, and response curves for the (A) Rusty Blackbird, (B) Olive-sided Flycatcher, and (C) Canada Warbler. Note that the y axis of all response all curves represents relative habitat suitability, wherein suitability increases from the bottom to the top of the axis. The x axis of all response curves except TPI begins at 0 (left side of figure), and values (e.g., height, distance) increase toward the right. For TPI, 1 = valley, 2 = low slope, 3 = mid slope, 4 = upper slope, and 5 = ridgetop. Covariate % contribution Permutation importance Response curve A. Rusty Blackbird Depth to water table 62.4% 43% Canopy height AVG 19.2% 30% Canopy height STD 12.5% 18.2% Distance to coniferous stand 7.1% 8.7% B. Olive-sided Flycatcher Canopy height AVG 24.2% 28.7% Canopy height STD 29.5% 25.9% Distance to coniferous stand 19.5% 21.3% Topographic Position Index 20.9% 19.3% Distance to stand with dead material 6% 4.8% C. Canada Warbler Depth to water table 36.5% 36.8% Distance to coniferous stand 18.7% 17.1% Canopy height STD 18.2% 14.6% Distance to deciduous stand 8% 12.2% Distance to stand with dead material 11.2% 10.6% Landscape complexity 7.3% 8.7% Covariate % contribution Permutation importance Response curve A. Rusty Blackbird Depth to water table 62.4% 43% Canopy height AVG 19.2% 30% Canopy height STD 12.5% 18.2% Distance to coniferous stand 7.1% 8.7% B. Olive-sided Flycatcher Canopy height AVG 24.2% 28.7% Canopy height STD 29.5% 25.9% Distance to coniferous stand 19.5% 21.3% Topographic Position Index 20.9% 19.3% Distance to stand with dead material 6% 4.8% C. Canada Warbler Depth to water table 36.5% 36.8% Distance to coniferous stand 18.7% 17.1% Canopy height STD 18.2% 14.6% Distance to deciduous stand 8% 12.2% Distance to stand with dead material 11.2% 10.6% Landscape complexity 7.3% 8.7% Open in new tab TABLE 2. Covariates included in final models as well as associated percent (%) contribution scores, permutation importance scores, and response curves for the (A) Rusty Blackbird, (B) Olive-sided Flycatcher, and (C) Canada Warbler. Note that the y axis of all response all curves represents relative habitat suitability, wherein suitability increases from the bottom to the top of the axis. The x axis of all response curves except TPI begins at 0 (left side of figure), and values (e.g., height, distance) increase toward the right. For TPI, 1 = valley, 2 = low slope, 3 = mid slope, 4 = upper slope, and 5 = ridgetop. Covariate % contribution Permutation importance Response curve A. Rusty Blackbird Depth to water table 62.4% 43% Canopy height AVG 19.2% 30% Canopy height STD 12.5% 18.2% Distance to coniferous stand 7.1% 8.7% B. Olive-sided Flycatcher Canopy height AVG 24.2% 28.7% Canopy height STD 29.5% 25.9% Distance to coniferous stand 19.5% 21.3% Topographic Position Index 20.9% 19.3% Distance to stand with dead material 6% 4.8% C. Canada Warbler Depth to water table 36.5% 36.8% Distance to coniferous stand 18.7% 17.1% Canopy height STD 18.2% 14.6% Distance to deciduous stand 8% 12.2% Distance to stand with dead material 11.2% 10.6% Landscape complexity 7.3% 8.7% Covariate % contribution Permutation importance Response curve A. Rusty Blackbird Depth to water table 62.4% 43% Canopy height AVG 19.2% 30% Canopy height STD 12.5% 18.2% Distance to coniferous stand 7.1% 8.7% B. Olive-sided Flycatcher Canopy height AVG 24.2% 28.7% Canopy height STD 29.5% 25.9% Distance to coniferous stand 19.5% 21.3% Topographic Position Index 20.9% 19.3% Distance to stand with dead material 6% 4.8% C. Canada Warbler Depth to water table 36.5% 36.8% Distance to coniferous stand 18.7% 17.1% Canopy height STD 18.2% 14.6% Distance to deciduous stand 8% 12.2% Distance to stand with dead material 11.2% 10.6% Landscape complexity 7.3% 8.7% Open in new tab TABLE 3. Evaluation metrics for Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler models. Note that, for the lowest presence threshold (LPT) omission rate, values closer to 0% indicate a better calibrated model and, for the 10% presence threshold (10PT) omission rate, values closer to 10% indicate a better calibrated model. AUC = Area under the curve. Bird AUCtrain AUCtest AUCdiff Observed LPT omission rate Observed 10PT omission rate Rusty Blackbird 0.697 0.654 0.043 3.2% 14.23% Olive-sided Flycatcher 0.685 0.668 0.018 0.39% 11.98% Canada Warbler 0.725 0.692 0.034 0.64% 11.39% Bird AUCtrain AUCtest AUCdiff Observed LPT omission rate Observed 10PT omission rate Rusty Blackbird 0.697 0.654 0.043 3.2% 14.23% Olive-sided Flycatcher 0.685 0.668 0.018 0.39% 11.98% Canada Warbler 0.725 0.692 0.034 0.64% 11.39% Open in new tab TABLE 3. Evaluation metrics for Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler models. Note that, for the lowest presence threshold (LPT) omission rate, values closer to 0% indicate a better calibrated model and, for the 10% presence threshold (10PT) omission rate, values closer to 10% indicate a better calibrated model. AUC = Area under the curve. Bird AUCtrain AUCtest AUCdiff Observed LPT omission rate Observed 10PT omission rate Rusty Blackbird 0.697 0.654 0.043 3.2% 14.23% Olive-sided Flycatcher 0.685 0.668 0.018 0.39% 11.98% Canada Warbler 0.725 0.692 0.034 0.64% 11.39% Bird AUCtrain AUCtest AUCdiff Observed LPT omission rate Observed 10PT omission rate Rusty Blackbird 0.697 0.654 0.043 3.2% 14.23% Olive-sided Flycatcher 0.685 0.668 0.018 0.39% 11.98% Canada Warbler 0.725 0.692 0.034 0.64% 11.39% Open in new tab TABLE 4. Percentages of (1) habitat that was identified as being suitable for all birds and (2) the entire province of Nova Scotia that were classified as valley or valley/low slope by the Topographic Position Index (TPI) and as wet by the Depth to water table index (D2W). % classified as valley by “TPI” % classified as valley or low slope by “TPI” % classified as wet area by “D2W” Habitat identified as suitable for all birds 43% 66% 49% All of NS 7% 34% 8% % classified as valley by “TPI” % classified as valley or low slope by “TPI” % classified as wet area by “D2W” Habitat identified as suitable for all birds 43% 66% 49% All of NS 7% 34% 8% Open in new tab TABLE 4. Percentages of (1) habitat that was identified as being suitable for all birds and (2) the entire province of Nova Scotia that were classified as valley or valley/low slope by the Topographic Position Index (TPI) and as wet by the Depth to water table index (D2W). % classified as valley by “TPI” % classified as valley or low slope by “TPI” % classified as wet area by “D2W” Habitat identified as suitable for all birds 43% 66% 49% All of NS 7% 34% 8% % classified as valley by “TPI” % classified as valley or low slope by “TPI” % classified as wet area by “D2W” Habitat identified as suitable for all birds 43% 66% 49% All of NS 7% 34% 8% Open in new tab FIGURE 2. Open in new tabDownload slide Heat maps delineating relative habitat suitability for the Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler, where darker areas denote more suitable habitat and lighter areas denote less suitable habitat. (Inset) Map of Canada with the study area (the province of Nova Scotia) shown in black. FIGURE 2. Open in new tabDownload slide Heat maps delineating relative habitat suitability for the Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler, where darker areas denote more suitable habitat and lighter areas denote less suitable habitat. (Inset) Map of Canada with the study area (the province of Nova Scotia) shown in black. FIGURE 3. Open in new tabDownload slide Binary habitat suitability maps for the Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler (created by applying the MaxSS threshold to the relative habitat suitability maps shown in Figure 2), where black denotes suitable habitat areas, and white denotes unsuitable habitat areas. FIGURE 3. Open in new tabDownload slide Binary habitat suitability maps for the Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler (created by applying the MaxSS threshold to the relative habitat suitability maps shown in Figure 2), where black denotes suitable habitat areas, and white denotes unsuitable habitat areas. FIGURE 4. Open in new tabDownload slide Map showing areas in Nova Scotia that contain suitable habitat for all 3 target species. These areas were delineated by overlaying the binary suitability surfaces created for each individual species (i.e. binary maps shown in Figure 3) and identifying the intersecting areas of suitable habitat. FIGURE 4. Open in new tabDownload slide Map showing areas in Nova Scotia that contain suitable habitat for all 3 target species. These areas were delineated by overlaying the binary suitability surfaces created for each individual species (i.e. binary maps shown in Figure 3) and identifying the intersecting areas of suitable habitat. (Note that a geodatabase containing GIS files that delineate relative habitat suitability and binary habitat suitability for each of the target species is available online as Supplementary Material data. This geodatabase also contains a file that delineates areas predicted to contain suitable habitat for all 3 species.) Rusty Blackbird The final Rusty Blackbird model included the following covariates (listed in order of permutation importance scores): depth to water table, canopy height AVG, canopy height STD, and distance to coniferous stand (Table 2A). Response curves showed that relative habitat suitability steeply declined as depth to water table increased. In other words, habitat suitability rapidly decreased with a loss of soil moisture for this species. Note that depth to water table had greater importance to the Rusty Blackbird model than did any other covariate in any other model. Response curves for canopy height AVG and canopy height STD showed a linear decrease and a linear increase, respectively, revealing that Rusty Blackbirds tend to prefer areas characterized by lower average canopy height but greater canopy heterogeneity. In addition, habitat was more suitable near coniferous stands (Table 2A). The binary prediction surface indicated that 43% of Nova Scotia contained suitable Rusty Blackbird habitat (Figure 4A, Supplemental Material Figure 3A). However, model evaluation metrics indicated that the model for this species was more poorly fit (AUCtest: 0.654, observed omission rate at 10PT: 3.2% higher than expected; Table 3) than were models for the other species. Olive-sided Flycatcher The 5 covariates selected in the Olive-sided Flycatcher model (from greatest to least importance) were canopy height AVG, canopy height STD, distance to coniferous stand, topographic position index, and distance to stand with dead material (Table 2B). In contrast to the Rusty Blackbird, relative habitat suitability improved for the Olive-sided Flycatcher as average canopy height increased, but canopy heterogeneity decreased. The response curve for distance to coniferous stand further indicated that habitat became less suitable for Olive-sided Flycatchers away from conifer dominated areas (Table 2B), and this decrease was more pronounced than that observed for the Rusty Blackbird. Results for topographic position index indicated that the Olive-sided Flycatcher was more likely to occur in valleys and low-slope areas. Finally, relative habitat suitability also decreased away from stands that contained dead material, although this covariate had an importance score of <5%. The Olive-sided Flycatcher model (AUCtest score: 0.667, observed omission rate at 10PT: 1.98% higher than expected; Table 3) predicted that almost half of Nova Scotia remains suitable for this species (49% of the provincial landmass; Figure 4B, Supplemental Material Figure 3B). Canada Warbler The Canada Warbler model contained 6 covariates (from greatest to least importance): depth to water table, distance to coniferous stand, canopy height STD, distance to deciduous stand, distance to stand with dead material, and landscape complexity (Table 2C). As with the Rusty Blackbird, suitability scores decreased as depth to water table values increased, although this trend was less pronounced for the Canada Warbler than for the Rusty Blackbird. (Depth to water table also received a lower importance score in the Canada Warbler model than in the Rusty Blackbird model, although it was selected as the most important covariate in both models.) The Canada Warbler further showed a similar response to distance to coniferous stand as the other 2 species, wherein relative habitat suitability decreased away from conifer dominated areas. Response curves for the covariates canopy height STD, distance to deciduous stand, and distance to stand with dead material revealed that relative habitat suitability increased in areas characterized by greater variation in canopy height but decreased away from areas dominated by deciduous trees and/or woody debris. Finally, the response curve for landscape complexity indicated that relative habitat suitability increased in more topographically complex areas (Table 2C). Compared to models for the other 2 species, the Canada Warbler model showed the best fit (AUCtest score: 0.692, observed omission rate at 10PT: 1.39% higher than expected; Table 3), but predicted the smallest amount of suitable habitat (35% of the provincial landmass; Figure 4C, Supplementary Material Figure 3B). DISCUSSION This is the first study to develop provincial SDMs for the Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler in Nova Scotia. At first glance, the relatively low AUCtest scores (0.654–0.692) for the SDMs appear to suggest that their predictive power was only marginally better than random. However, although AUC is among the most widely used model evaluation metrics, many previous researchers have reported that it is a poor evaluator of SDMs (Lobo et al. 2008, Gonzalez et al. 2011), and this is especially true for presence-only models (Lobo et al. 2008, Van Proosdij et al. 2015). Furthermore, while AUC helps evaluate a model’s discriminatory power, it provides no indication of goodness-of-fit and tends to reward overly complex models. Multiple factors are known to artificially inflate AUC scores, including species prevalence across the study landscape, the number of occurrence points in species datasets, and sample bias (Lobo et al. 2008, Van Proosdij et al. 2015). Indeed, in other research (Bale 2017), we found that the models which yielded the highest AUCtest scores were those that employed the species datasets most severely affected by sample bias; a factor for which we controlled. Conversely, a comparison of expected vs. observed omission rates indicated that Olive-sided Flycatcher and Canada Warbler models were reasonably well fit. Observed omission rates for the Rusty Blackbird model showed larger deviations from expected omission rates than did observed omission rates for the other species considered in this study, indicating that, among the 3 study species, the Rusty Blackbird model had the poorest fit. This was also the case in species abundance models generated by Westwood et al. (2019), who used a similar dataset and postulated that low sample sizes were responsible for modelling challenges. In our case, only 136 Rusty Blackbird occurrences were available for model building, compared to 502 occurrences and 312 occurrences for the Olive-sided Flycatcher and Canada Warbler, respectively. We therefore suspect that our model overpredicted the amount of suitable Rusty Blackbird habitat in Nova Scotia (43%) due to a smaller sample size and the incorporation of fewer covariates. Species-Specific Habitat Associations For all species, covariates selected through reverse stepwise elimination generally corresponded well with known habitat preferences, indicating good ecological realism of the SDMs. The Rusty Blackbird is a wetland obligate that is dependent on shallow water to meet its foraging needs (Powell et al. 2010, 2014), which explains why depth to water table was (1) assigned a highest permutation importance score than all other covariates in all models and (2) why relative habitat suitability was found to rapidly decrease as depth to water table values increased. Rusty Blackbird nesting preferences (i.e. dense patches of stunted conifers surrounded by sparse canopy closure; Matsuoka et al. 2010) were also reflected in model results (i.e. associations with forest structure variables that represent patchy habitat and coniferous stands). We found that the most suitable Olive-sided Flycatcher habitat was also characterized by patchy vegetation dominated by coniferous trees, which has been noted in other parts of this species’ range as well (e.g., in Oregon, McGarigal and McComb 1995; in California, Brandy 2001). Although the Olive-sided Flycatcher has been recognized as an edge specialist (McGarigal and McComb 1995) that commonly nests in emergent trees along forest edges, we did not directly test for an edge association in this study. Nonetheless, our Olive-sided Flycatcher model showed that relative habitat suitability increased as distance to stand with dead material decreased. This finding may have resulted from this species’ preference for post-disturbance, early-seral habitat with higher snag densities, a habitat type that is associated with forest edges (Robertson and Hutto 2007). Results for topographic position index predicted that Olive-sided Flycatcher habitat quality should be higher in valleys and on low-slopes, areas that favor the formation of hydrological landforms, such as wetlands and watercourses (Mackay and Lindenmayer 2001). This corresponds with Westwood et al. (2019)who found that, for this species, an interaction term between forest cover and wetness was the best predictor of habitat suitability; whereas, for the Canada Warbler, habitat suitability was better predicted by wetness alone. (In our model, depth to water table was not selected in the final Olive-sided Flycatcher model, while forest cover and structure variables were selected.) Previous studies on Canada Warbler habitat in the eastern portion of this species’ range have identified vegetated wetlands and moist forests as important habitat types for this species (e.g., Lambert and Faccio 2005, Goodnow and Reitsma 2011, Westwood et al. 2017), although the Canada Warbler is associated with mesic deciduous forest habitats in western North America (Ball et al. 2016, Hunt et al. 2017). Our study, in which (1) depth to water table was the most important covariate in the Canada Warbler model and (2) wetter areas were associated with higher habitat suitability values, confirmed findings of previous studies in eastern North America. Westwood et al. (2019) found that, at a broad scale, landscape complexity was associated with higher predicted population density for Canada Warbler. Despite using a different dataset to characterize complexity in the current study, we also found that more complex landscapes positively impacted habitat suitability for Canada Warbler. This association is also likely due to the species’ preference for wet habitat, as areas with higher landscape complexity tend to have greater wetland density (Anderson et al. 2012). (Indeed, a wetland density index was 1 of 3 input layers used to create the landscape complexity index; see Appendix.) Results for forest structure covariates further revealed that suitable Canada Warbler habitat was positively associated with high vertical complexity and a denser understory layer, confirming previous findings (e.g., Reitsma et al. 2009, Goodnow and Reitsma 2011; Becker et al. 2012). Highly suitable Canada Warbler habitat was also associated with proximity to coniferous and deciduous stands, consistent with the species’ preference for mixed wood stands in the eastern part of its range (Reitsma et al. 2009, Becker et al. 2012). However, the higher permutation score associated with distance to coniferous stand may reflect the predominance of this cover type in the province of Nova Scotia, rather than a real ecological preference for cover type, as the Canada Warbler is known to nest in red maple (Acer rubrum) floodplains in this region (Westwood 2016). Benefits of Using Topographic Covariates to Improve Climate Resilience of Model Predictions While covariates related to forest type and structure are typically considered to have a more direct influence over species distributions, our models demonstrated that topographic features can also have predictive value in SDMs. In this study, not only did reverse stepwise elimination identify both topographic and forest covariates as being important to relative habitat suitability for all 3 species at the resolution and scale considered, topographic covariates showed stronger explanatory power than did many forest covariates, particularly for the Rusty Blackbird and Canada Warbler, likely due to their relationship with wet habitat. The stability of topography is especially important in an era of climate change, as shifts in temperature and precipitation regimes are expected to significantly alter North American forest ecosystems and species assemblages, including in Nova Scotia (Coristine and Kerr 2011, Robillard et al. 2015, Stralberg et al. 2015). Unlike forest features, topography remains relatively unaffected by a changing climate, and furthermore, topographic processes help regulate forest and other micro-level habitat characteristics (Holling 1986, Mackay and Lindenmayer 2001). For example, topography controls hydrological conditions, whereby moisture and nutrients (both of which are needed to support high-quality habitat for our target species) accumulate in concave lowlands in response to gravitational potential energy gradients (Mackay and Lindenmayer 2001). Thus, for obligate and facultative wetland species such as the Rusty Blackbird and Canada Warbler, habitat located in these areas may be more likely to remain suitable into the future as climate changes than habitat found in other landform types. Models that employ topographic covariates may also offer practical advantages related to model accuracy and utility, as characteristics of forest vegetation often change (e.g., through silviculture practices or the natural processes of disturbance and succession) more quickly than GIS layers representing them can be updated. Conversely, changes in topography often occur at geologic timescales, and therefore topographic GIS layers can be expected to remain relatively accurate. Nonetheless, we acknowledge that enduring topographic features are unlikely to be universally useful to SDMs; rather, the predictive power of topographic covariates, like that of all covariates, will undoubtedly vary according to species-specific habitat requirements. For some species, topographic features may not show any predictive power at all. In the current study, the topographic covariate depth to water table showed the strongest predictive power for Rusty Blackbird and Canada Warbler models. Conversely, for the Olive-sided Flycatcher model, although topographic position index was selected, forest covariates were more explanatory overall. This is likely due to the strong association between the Olive-sided Flycatcher and forest cover and structure, observed both in our study and elsewhere (Spies et al. 2007, Azeria et al. 2011, Westwood et al. 2019). We also acknowledge that some topographic features will be less useful in predicting climate resilience than others. As Gill et al. (2015) pointed out, conservation plans that incorporate topographic landforms should consider the persistence of these landforms across space and time, as various landform types form, move, and dissolve at not only different spatial scales, but at different time scales as well (Gill et al. 2015). In this study, we only considered topographic covariates that (1) could reasonably be expected to have ecological relevance to the distribution of bird habitat and (2) represented features created over millennia or eons (a characteristic of climate-resilient features; Dobrowski et al. 2012, Gill et al. 2015). Implications of Covariate-Specific Results In this study, all topographic covariates (i.e. depth to water table, landscape complexity, and topographic position index) showed predictive power, although the strength of this power varied, and no covariate was selected in all models. Implications of covariate-specific findings are discussed below. Depth to water table. Although climate change is likely to impact absolute depth to water table values (Kumar 2012), the depth to water table layer itself was derived from physiographic features. The topographic features used to predict water flow and levels are expected to be relatively enduring, and even if absolute wetness changes, relative wetness is unlikely to change (e.g., an area currently predicted to be drier is unlikely to become a wet area, and vice versa). Landscape complexity. While landscape complexity was selected in the Canada Warbler model, the conservation of topographically complex areas should benefit not only Canada Warbler but many other species as well. Topographic complexity creates steep climatic gradients that parse regional climate patterns into variable local-scale microclimates (Dobrowski et al. 2012). This in turn creates a diverse array of habitats and provides species with increased opportunities to move as climate changes. Indeed, it is not a coincidence that 25% of global terrestrial biodiversity is found in complex, mountainous landscapes (Dobrowski et al. 2012). Topographic position index. The Olive-sided Flycatcher model developed in this study indicated that, in Nova Scotia, this species tends to occur in low topographic positions, such as valleys or along slope bottoms. Our Rusty Blackbird and Canada Warbler models confirmed that these species are highly dependent on wet areas, which also occur in low topographic positions and topographically complex areas (Anderson et al. 2012). In many cases, low topographic positions located in topographically complex landscapes should be particularly useful for predicting climate-resilient habitat. Consider the fact that cold air pooling is common and widespread in complex, mountainous landscapes (Dobrowski et al. 2012), where cold air pools in low topographic positions. This phenomenon acts to decouple the local in situ climate from the regional climate, thereby providing a buffer against climate change. Cold air pooling can also shelter affected areas from regional advective influences and lower minimum temperatures (which have increased almost twice as fast as maximum temperatures over the past century; IPCC 2007). Implications for Conservation of Target Species Our results suggest that (1) Nova Scotian forests retain a high proportion of breeding habitat for the 3 species at risk considered in this study and that (2) this habitat is well distributed throughout the province, providing opportunities for conservation (Figures 3 and 4). Although the Olive-sided Flycatcher was predicted to have a greater amount of suitable habitat than the other 2 species (consistent with earlier research by Westwood et al. 2019), 22% of the Nova Scotia landmass was predicted to be suitable for all 3 species. Therefore, the Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler may have the potential to be conserved as a suite in this region. These findings corroborate those of other recent studies that used field measurements (Westwood et al. 2016) and a different type of SDM (a hierarchical species abundance model; Westwood et al. 2019) to show that the Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler share similar habitat preferences in this region. A spatial prioritization of conservation areas has been completed in this region for the Canada Warbler (Westwood 2017), but not for the other 2 species. The overlapping areas indicated in our study could be prioritized in ongoing conservation planning efforts for these at-risk species in a context of climate change. Topographic features, particularly those related to hydrologic control, showed strong predictive power in our models. Furthermore, almost half of the habitat predicted to be suitable for all 3 species (i.e. areas highlighted in Figure 4) was classified as a wet area by the depth to water table index. A plurality of habitat predicted to be suitable for all 3 species was classified as a valley by the topographic position index, and a majority was classified as either a valley or a low-slope by this same index. However, wet areas and low topographic positions were much less prevalent across Nova Scotia as a whole (Table 4). Previous studies that observed greater productivity and passerine species richness in wet areas hypothesized that this correlation can be explained by (1) increased moisture and nutrient accumulation (Neave et al. 1996) and (2) increased vertical complexity due to the juxtaposition of multiple habitats in riparian forest ecotones (LaRue et al. 1995). LaRue et al. (1995) also reported that many bird species that typically prefer drier environments may also be found in lowland riparian sites; however, the reverse was not true for species that prefer wetter habitat (LaRue et al. 1995). Thus, engaging in conservation planning while considering topographic variables related to hydrological control may not only be important for conserving climate-resilient habitat for the 3 species considered in our study, but a greater suite of passerines as well. Limitations Although there is theoretical and modeled evidence to suggest abiotic topography may be useful in identifying climate-resilient habitat, the utility of our resilience-based approach for conserving and recovering species has not yet been tested. Robust testing would necessitate long-term empirical studies to monitor the persistence of these species in habitats predicted to be climate resilient. Furthermore, while habitat delineated using a combination of topographic and forest covariates should be more likely to persist in the face of climate change than habitat delineated using forest covariates alone, additional steps could be taken to further prioritize suitable habitat according to climate resilience. For example, habitat patches could be scored according to “landscape resilience,” which is defined according to geophysical diversity, topoclimate diversity, and permeability (see Anderson et al. 2012, 2016). In addition, it is possible that model reliability was affected by classification errors in underlying datasets, which may be considerable in some parts of the FID (Westwood 2016). To minimize confounding effects of classification errors, we defined cover types generally (i.e. coniferous or deciduous), rather than by dominant tree species. In addition, although a temporal disconnect between FID update year and year of observation may have introduced a source of error, most components of the FID were last updated within the range of species observations used in this study (2005–2013). Nonetheless, as with all SDMs, results should be treated with caution until adequate ground truthing is conducted (see Westwood et al. 2019). In terms of practical limitations, while an assumption of Maxent modeling is that detectability during sampling does not vary with the covariates that determine probability (Yackulic et al. 2013), we did not attempt to control for detectability as related to habitat type in this study. However, most similar studies have not done so either. Unfortunately, available information about detectability for our study species in this region was inadequate to reliably investigate the impact of habitat type on detectability. Finally, although we predicted suitable and potentially climate-resilient habitat for these species in Nova Scotia, it should be noted that habitat loss on the breeding grounds may not be the primary threat to recovery for these species, with other impacts such as loss of wintering habitat, mercury contamination, and parasites potentially posing more imminent threats (Edmonds et al. 2010, Greenberg et al. 2011, Savard et al. 2011). Although high-quality breeding habitat remains essential to the persistence and recovery of these species, conservation efforts should nonetheless consider full-life-cycle modelling where possible (Kearney and Porter 2009, Zurrell 2017). CONCLUSIONS As human demands on the landscape and the effects of climate change continue to increase, the ability to predict suitable habitat that is also resilient to changing temperature and moisture regimes is becoming more critical. In this study, we used Maxent modeling techniques and an intersection of topographic and forest covariates to develop predictive, high-resolution, spatially explicit habitat models for the Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler. Due to the enduring nature of topographic features and the influence that these features have over the structure and function of forest habitat, the areas of suitable habitat that we identified are likely to be more climate resilient than areas predicted to be suitable using forest covariates alone. Nonetheless, future research could expand and improve on these findings by further prioritizing suitable habitat patches using other metrics of resilience. Our models can be directly used to benefit the conservation of these at-risk species by supporting the identification of areas for land acquisition and/or habitat management. Moreover, if confirmed through longer-term monitoring and field studies, our resilience approach to species distribution modeling could be applied to other terrestrial vertebrate species to predict key areas of climate-resilient habitat. ACKNOWLEDGMENTS The authors would like to express their sincere appreciation to Jennifer Strang and Raymond Jahncke of the Dalhousie GIS Center, without whose guidance this research would not have been possible. The authors also thank Dr. Cindy Staicer, who helped with the interpretation of bird models and provided insight into songbird habitat selection strategies; and Clara Ferrari, whose preliminary research into breeding habitat in the Southwest Nova Biosphere Reserve helped inform modeling methodology and the selection of candidate covariates. We also acknowledge the Nova Scotia Department of Natural Resources, the Nova Scotia Geomatics Centre, and the Atlantic Canada Data Conservation Centre (ACCDC), which provided the GIS data used in this research. We are also grateful to the Maritime Breeding Bird Atlas (MBBA, which supplied many of the observations contained in the ACCDC bird dataset), the thousands of volunteers who gathered data contained in the atlas, and the many MBBA sponsors (Bird Studies Canada, Environment Canada, the Canadian Wildlife Service, the New Brunswick Department of Natural Resources, and the Prince Edward Island Department of Agriculture and Forestry). Funding statement: This work was supported by K. Beazley’s Social Sciences and Humanities Research Council of Canada (SSSHRC) research grant. Ethics statement: This research was conducted using preexisting data and did not require ethics approval. Author contributions: S. Bale developed the Maxent models, completed statistical analyses, and wrote the manuscript. K. Beazley supervised the research and provided theoretical and conceptual direction as well as editorial support for the manuscript. A. Westwood provided editorial support for the manuscript as well as guidance pertaining to species’ biology and also helped with the interpretation of results. P. Bush provided conceptual direction, helped edit the manuscript, and offered expertise pertaining to the Acadian forest ecosystem. Conflict of interest statement: None of the authors have any conflicts of interest to declare. Data depository: Analyses reported in this article can be reproduced using the data provided by Bale et al. (2019). LITERATURE CITED Agriculture and Agri-Food Canada ( 2015 ). Land use 2010 [Raster GIS dataset] . Agriculture and Agri-Food Canada , Ottawa, ON, Canada . https://open.canada.ca/data/en/dataset/9e1efe92-e5a3-4f70-b313-68fb1283eadf Google Preview WorldCat COPAC Altman , B. , and R. Sallabanks ( 2012 ). Olive-sided Flycatcher (Contopus cooperi) . In The Birds of North America Online ( A. Poole , Editor). Cornell Lab of Ornithology , Ithaca, NY, USA . https://birdsna.org/Species-Account/bna/species/olsfly Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Anderson , M. G. , and C. E. Ferree ( 2010 ). Conserving the stage: Climate change and the geophysical underpinnings of species diversity . PLOS One 5 : e11554 . Google Scholar Crossref Search ADS PubMed WorldCat Anderson , M. G. , A. Barnett , M. Clark , C. Ferree , A. Olivero Sheldon , and J. Prince ( 2016 ). Resilient sites for terrestrial conservation in eastern North America . The Nature Conservancy, Eastern Conservation Science , Boston, MA, USA . Google Preview WorldCat COPAC Anderson , M. G. , M. Clark , and A. O. Sheldon ( 2012 ). Resilient sites for terrestrial conservation in the northeast and mid-Atlantic region . 2010 Edition. The Nature Conservancy Eastern Conservation Science Technical Report , Boston, MA, USA . Google Preview WorldCat COPAC Avery , M. L . ( 2013 ). Rusty Blackbird (Euphagus carolinus) . In The Birds of North America Online ( A. Poole , Editor). Cornell Lab of Ornithology , Ithaca, NY, USA . https://birdsna.org/SpeciesAccount/bna/species/rusbla Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Azeria , E. T. , J. Ibarzabal , C. Hebert , J. Boucher , L. Imbeau , and J. P. L. Savard ( 2011 ). Differential response of bird functional traits to post-fire salvage logging in a boreal forest ecosystem . Acta Oecologica 37 : 220 – 229 . Google Scholar Crossref Search ADS WorldCat Baldwin , R. A . ( 2009 ). Use of maximum entropy modeling in wildlife research . Entropy 11 : 854 – 866 . Google Scholar Crossref Search ADS WorldCat Bale , S. ( 2017 ). Building an arc in the Anthropocene . M.S. thesis, Dalhousie University , Halifax, NS, Canada . https://dalspace.library.dal.ca/handle/10222/73520 Google Preview WorldCat COPAC Bale , S. , K. F. Beazley , A. Westwood , and P. Bush ( 2019 ). Data from: The benefits of using topographic features to predict climate-resilient habitat for migratory forest landbirds: An example for the Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler . The Condor: Ornithological Applications 122 : 1 – 19 . doi:10.5061/dryad.jwstqjq50 WorldCat Ball , J. R. , P. Solymos , K. A. Schmiegelow , S. Hache , J. Schieck , and E. Bayne ( 2016 ). Regional habitat needs of a nationally listed species, Canada Warbler (Cardellina canadensis), in Alberta, Canada . Avian Conservation and Ecology 11 : 10 . Google Scholar Crossref Search ADS WorldCat Becker , D. A. , P. B. Wood , and P. D. Keyser ( 2012 ). Canada Warbler use of harvested stands following timber management in the southern portion of their range . Forest Ecology and Management 276 : 1 – 9 . Google Scholar Crossref Search ADS WorldCat Beier , P. , and B. Brost ( 2010 ). Use of land facets to plan for climate change: Conserving the arenas, not the actors . Conservation Biology 24 : 701 – 710 . Google Scholar Crossref Search ADS PubMed WorldCat Beier , P. , P. Sutcliffe , J. Hjort , D. P. Faith , R. L. Pressey , and F. Albuquerque ( 2015 ). A review of selection-based tests of abiotic surrogates for species representation . Conservation Biology 29 : 668 – 679 . Google Scholar Crossref Search ADS PubMed WorldCat Beyer , H. L . ( 2004 ). Hawth’s analysis tools for ArcGIS [software] . http://www.spatialecology.com/htools/tooldesc.php Brandy , P. M . ( 2001 ). A hierarchical analysis of Olive-sided Flycatcher habitat use in a managed landscape . M. Sc. Thesis . Humboldt State University , Arcata, California, USA . Ceballos , G. , P. R. Ehrlich , and R. Dirzo ( 2017 ). Biological annihilation via the ongoing sixth mass extinction signaled by vertebrate population losses and declines . Proceedings of the National Academy of Sciences USA 114 : E6089 – E6096 . Google Scholar Crossref Search ADS WorldCat Cooley , S. ( 2014 ). Terrain roughness – 13 ways . http://gis4geomorphology.com/roughness-topographic-position Coristine , L. E. , and J. T. Kerr ( 2011 ). Habitat loss, climate change, and emerging conservation challenges in Canada . Canadian Journal of Zoology 89 : 435 – 451 . Google Scholar Crossref Search ADS WorldCat Crooks , K. R. , C. L. Burdett , D. M. Theobald , S. R. B. King , M. Di Marco , C. Rondinini , and L. Boitani ( 2017 ). Quantification of habitat fragmentation reveals extinction risk in terrestrial mammals . Proceedings of the National Academy of Sciences USA 114 : 7635 – 7640 . Google Scholar Crossref Search ADS WorldCat Dobrowski , S. Z . ( 2011 ). A climatic basis for microrefugia: The influence of terrain on climate . Global Change Biology 17 : 1022 – 1035 . Google Scholar Crossref Search ADS WorldCat Dobrowski , S. Z. , J. Abatzoglou , A. K. Swanson , J. A. Greenberg , A. R. Mynsberge , Z. A. Holden , and M. K. Schwartz ( 2012 ). The climate velocity of the contiguous United States during the 20th century . Global Change Biology 19 : 241 – 251 . Google Scholar Crossref Search ADS PubMed WorldCat Edmonds , S. T. , D. C. Evers , D. A. Cristol , C. Mettke-Hofmann , L. L. Powell , A. J. McGann , J. W. Armiger , O. P. Lane , D. F. Tessler , P. Newell , K. Heyden , and N. J. O’Driscoll ( 2010 ). Geographic and seasonal variation in mercury exposure of the declining Rusty Blackbird . The Condor 112 : 789 – 799 . Google Scholar Crossref Search ADS WorldCat Elith , J. , and J. R. Leathwick ( 2009 ). Species distribution models: Ecological explanation and prediction across space and time . Annual Review of Ecology, Evolution, and Systematics 40 : 677 – 697 . Google Scholar Crossref Search ADS WorldCat Elith , J. , C. H. Graham , R. P. Anderson , M. Dudik , S. Ferrier , A. Guisan , R. J. Hijmans , F. Huettmann , J. R. Leathwick , A. Lehmann , A. Li , et al. . ( 2006 ). Novel methods improve prediction of species’ distributions from occurrence data . Ecography 29 : 129 – 151 . Google Scholar Crossref Search ADS WorldCat Elith , J. , S. J. Phillips , T. Hastie , M. Dudik , Y. E. Chee , and C. J. Yates . ( 2010a ). A statistical explanation of Maxent for ecologists . Biodiversity Research 17 : 1 – 10 . WorldCat Elith , J. , M. Kearney , and S. Phillips . ( 2010b ). The art of modelling range-shifting species . Methods in Ecology and Evolution 1 : 330 – 342 . Google Scholar Crossref Search ADS WorldCat Environment Canada ( 2015 ). Management Plan for the Rusty Blackbird (Euphagus carolinus) in Canada . Species at Risk Act Management Plan Series. Environment Canada , Ottawa, ON, Canada . Google Preview WorldCat COPAC Environment Canada ( 2016a ). Recovery strategy for the Olive-sided Flycatcher (Contopus cooperi). Species at Risk Act Recovery Strategy Series Report. http://publications.gc.ca/collections/collection_2016/eccc/En3-4-233-2016eng.pdf Environment Canada ( 2016b ). Recovery strategy for the Canada Warbler (Cardellina canadensis) in Canada. Species at Risk Act Recovery Strategy Series Report. Environment Canada , Ottawa, ON, Canada . COPAC Fourcade , Y. , J. O. Engler , D. Rödder , and J. Secondi ( 2014 ). Mapping species distributions with MAXENT using a geographically biased sample of presence data: A performance assessment of methods for correcting sampling bias . PLOS One 9 : e97122 . Google Scholar Crossref Search ADS PubMed WorldCat Franklin , J. ( 1995 ). Predictive vegetation mapping: Geographic modelling of biospatial patterns in relation to environmental gradients . Progress in Physical Geography 19 : 474 – 499 . Google Scholar Crossref Search ADS WorldCat Franklin , J. ( 2009a ). Data for species distribution models: The biological data . In Mapping Species Distributions: Spatial Inference and Predictions ( M. Usher , D. Saunders , R. Peet , and A. Dobson , Editors). Cambridge University Press , Cambridge, UK . pp. 55 – 75 . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Franklin , J. ( 2009b ). Ecological understanding of species distributions . In Mapping Species Distributions: Spatial Inference and Predictions ( M. Usher , D. Saunders , R. Peet , and A. Dobson , Editors). Cambridge University Press , Cambridge, UK . pp. 34 – 54 . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Franklin , J. , and J. Miller ( 2009 ). Statistical models – Modern regression . In Mapping Species Distributions: Spatial Inference and Predictions ( M. Usher , D. Saunders , R. Peet , and A. Dobson , Editors). Cambridge University Press , Cambridge, UK . pp. 114 – 153 . Google Preview WorldCat COPAC Gill , J. L. , J. L. Blois , B. Benito , S. Dobrowski , M. L. Hunter , Jr. , and J. L. McGuire ( 2015 ). A 2.5-million-year perspective on coarse-filter strategies for conserving nature’s stage . Conservation Biology 29 : 640 – 648 . Google Scholar Crossref Search ADS PubMed WorldCat Gonzalez , S. C. , J. A. Soto-Centeno , and D. L. Reed ( 2011 ). Population distribution models: Species distributions are better modeled using biologically relevant data partitions . BMC Ecology 11(1) : 20 . Google Scholar Crossref Search ADS WorldCat Goodnow , M. L. , and L. R. Reitsma ( 2011 ). Nest-site selection in the Canada Warbler (Wilsonia canadensis) in central New Hampshire . Canadian Journal of Zoology 89 : 1172 – 1177 . Google Scholar Crossref Search ADS WorldCat Government of Canada ( 2002 ). A to Z species index . Government of Canada , Ottawa, ON, Canada . https://wildlifespecies.canada.ca/species-risk-registry/sar/index/default_e.cfm Google Preview WorldCat COPAC Government of Canada ( 2017 ). Breeding Bird Survey results . Government of Canada , Ottawa, ON, Canada . https://wildlifespecies.canada.ca/breeding-bird-survey-results/ Google Preview WorldCat COPAC Greenberg , R. , D. W. Demarest , S. M. Matsuoka , C. Mettke-Hofmann , D. Evers , P. B. Hamel , J. Luscier , L. L. Powell , D. Shaw , M. L. Avery , et al. . ( 2011 ). Understanding declines in Rusty Blackbirds . In Boreal Birds of North America: A Hemispheric View of Their Conservation Links and Significance ( J. V. Wells , Editor). University of California Press , Berkeley, CA, USA . pp. 107 – 126 . Google Preview WorldCat COPAC Guisan , A. , and Z. Zimmermann ( 2000 ). Predictive habitat distribution models in ecology . Ecological Modelling 135 : 147 – 186 . Google Scholar Crossref Search ADS WorldCat Haché , S. , P. Solymos , T. Fontaine , E. Bayne , S. Cumming , F. Schmiegelow , and D. Stralberg ( 2014 ). Critical habitat of Olive-sided Flycatcher, Canada Warbler, and Common Nighthawk in Canada. Report 1. Project No. K4B20-13–0367. Technical Report for Environment Canada . http://www.borealbirds.ca/files/Technical_Reports/Hacheetal2014.pdf COPAC Holling , C. S . ( 1986 ). The resilience of terrestrial ecosystems . In Foundations of Ecological Resilience ( L. H. Gunderson , C. R. Allen , and C. S. Holling , Editors). Island Press , Washington, DC, USA . pp. 67 – 118 . Google Preview WorldCat COPAC Holling , C. S . ( 1992 ). Cross-scale morphology, geometry, and dynamics of ecosystems . Ecological Monographs 62 : 447 – 502 . Google Scholar Crossref Search ADS WorldCat Hooftman , D. A. P. , B. Edwards , and J. M. Bullock ( 2015 ). Reductions in connectivity and habitat quality drive local extinctions in a plant diversity hotspot . Ecography 29 : 583 – 592 . WorldCat Huettman , F. , and T. Gottschalk ( 2011 ). Simplicity, model fit, complexity and uncertainty in spatial prediction models applied over time: We are quite sure, aren’t we? In Predictive Species and Habitat Modeling in Landscape Ecology ( D. C. Ashton , Y. Wiersma , and F. Huettman , Editors). Springer-Verlag , New York, NY, USA. pp. 189 – 208 . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Hunt , A. R. , E. M. Bayne , and S. Hache ( 2017 ). Forestry and conspecifics influence Canada Warbler (Cardellina canadensis) habitat use and reproductive activity in boreal Alberta, Canada . The Condor: Ornithological Applications 119 : 832 – 847 . WorldCat IBM ( 2013 ). Statistical package for the social sciences . IBM , Armonk, NY, USA . Google Preview WorldCat COPAC [IPCC] Intergovernmental Panel on Climate Change ( 2007 ). Climate change 2007: The physical science basis . In Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change ( S. Solomon , D. Qin , M. Manning , Z. Chen , M. Marquis , K. B. Avery , M. Tignor , and H. L. Miller , Editors). Cambridge University Press , UK, and New York, NY, USA . Google Preview WorldCat COPAC International Rusty Blackbird Working Group ( 2013–2015 ). Rusty Blackbird papers. Rustyblackbird.org. http://rustyblackbird.org/research/resources/rusty-papers/ Kearney , M. , and W. Porter ( 2009 ). Mechanistic niche modelling: Combining physiological and spatial data to predict species’ ranges . Ecology Letters 12 : 334 – 350 . Google Scholar Crossref Search ADS PubMed WorldCat Kumar , C. P . ( 2012 ). Climate change and its impact on groundwater resources . International Journal of Engineering and Science 1 : 43 – 60 . WorldCat Lambert , J. D. , and S. D. Faccio ( 2005 ). Canada Warbler: Population status, habitat use, and stewardship guidelines for northeastern forests. Report No. 05-4. Vermont Institute of Natural Science Technical Report. http://vtecostudies.org/wp-content/uploads/2014/08/vce-cawa-lambert.pdf Larkin , D. J. , G. Vivian-Smith , and J. B. Zedler ( 2006 ). Topographic heterogeneity: Theory and ecological restoration . In Foundations of Restoration Ecology ( D. A. Falk , M. A. Palmer , and J. B. Zedler , Editors). Island Press , Washington, DC, USA . pp. 142 – 164 . Google Preview WorldCat COPAC LaRue , P. , L. Belanger , and J. Huot ( 1995 ). Riparian edge effects on boreal balsam fir bird communities . Canadian Journal of Forest Research 25 : 555 – 566 . Google Scholar Crossref Search ADS WorldCat Liu , C. , M. White , and G. Newell ( 2013 ). Selecting thresholds for the prediction of species occurrence with presence-only data . Journal of Biogeography 40 : 778 – 789 . Google Scholar Crossref Search ADS WorldCat Lobo , J. M. , A. Jimenez-Valverde , and R. Real ( 2008 ). AUC: A misleading measure of the performance of predictive distribution models . Global Ecology and Biogeography 17 : 145 – 151 . Google Scholar Crossref Search ADS WorldCat Mackay , B. G. , and D. B. Lindenmayer ( 2001 ). Towards a hierarchical framework for modeling the spatial distribution of animals . Journal of Biogeography 28 : 1147 – 1166 . Google Scholar Crossref Search ADS WorldCat Matsuoka , S. M. , D. Shaw , J. Sinclair , R. Johnson , N. Corcoran , P. Meyers , and N. Rojek ( 2010 ). Nesting ecology of the Rusty Blackbird in Alaska and Canada . The Condor 112 : 810 – 824 . Google Scholar Crossref Search ADS WorldCat McClure , C. J. , B. W. Rolek , K. McDonald , and G. E. Hill ( 2012 ). Climate change and the decline of a once common bird . Ecology and Evolution 2 : 370 – 378 . Google Scholar Crossref Search ADS PubMed WorldCat McGarigal , K. , and W. C. McComb ( 1995 ). Relationships between landscape structure and breeding birds in the Oregon Coast Range . Ecological Monographs 65 : 235 – 260 . Google Scholar Crossref Search ADS WorldCat Merow , C. , M. J. Smith , and J. A. Silander ( 2013 ). A practical guide to Maxent for modeling species’ distributions: What it does, and why inputs and settings matter . Ecography 36 : 1058 – 1069 . Google Scholar Crossref Search ADS WorldCat Millenium Ecosystem Assessment ( 2005 ). Ecosystems and Human Well-Being: Synthesis. Millenium Ecosystem Assessment Synthesis Report . Island Press , Washington, DC, USA . Google Preview WorldCat COPAC Murphy , P. N. , J. Ogilvie , K. Connor , and P. A. Arp ( 2007 ). Mapping wetlands: A comparison of two different approaches for New Brunswick, Canada . Wetlands 27 : 846 – 854 . Google Scholar Crossref Search ADS WorldCat National Audubon Society ( 2014 ). Audubon’s birds and climate change report: A primer for practitioners . Edition 1.2. National Audubon Society Conservation Science Report . http://climate.audubon.org/sites/default/files/AudubonBirds-Climate-Report-v1.2.pdf Google Preview WorldCat COPAC Neave , H. M. , R. B. Cunningham , T. W. , Norton , and H. A. Nix ( 1996 ). Biological inventory for conservation evaluation III: Relationships between birds, vegetation, and environmental attributes in southern Australia . Forest Ecology and Management 8 : 197 – 297 . Google Scholar Crossref Search ADS WorldCat Neily , P. D. , E. Quigley , B. Lawrence , B. Stewart , and T. Duke ( 2005 ). Ecological land classification for Nova Scotia . Revised edition 2005. Nova Scotia Department of Natural Resources, Renewable Resources Branch Report . https://novascotia.ca/natr/forestry/ecological/pdf/ELCrevised2.pdf Google Preview WorldCat COPAC Nogués-Bravo , D. , J. I. Lopez-Moreno , and S. M. Vicente-Serrano ( 2012 ). Climate change and its impact. In Mediterranean Mountain Environments ( I. Vogiatzakis , Editor). John Wiley & Sons, Ltd. , West Sussex, UK . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC North American Bird Conservation Initiative Canada ( 2012 ). The State of Canada’s Birds, 2012 . Environment Canada , Ottawa, ON, Canada. Google Preview WorldCat COPAC Noss , R . ( 1983 ). A regional landscape approach to maintain diversity . BioScience 33 : 700 – 706 . Google Scholar Crossref Search ADS WorldCat Nova Scotia Museum of Natural History ( 1996a ). Canadian setting. In Natural History of Nova Scotia , Volume 1 ( D. Davis and S. Brown , Editors). Nova Scotia Museum of Natural History , Halifax, NS, Canada . https://ojs.library.dal.ca/NSM/issue/view/349 Google Preview WorldCat COPAC Nova Scotia Museum of Natural History ( 1996b ). Climate. In Natural History of Nova Scotia , volume 1 ( D. Davis and S. Brown , Editors). Nova Scotia Museum of Natural History , Halifax, NS, Canada . https://ojs.library.dal.ca/NSM/issue/view/349 Google Preview WorldCat COPAC Pannozzo , L. , and R. Colman ( 2008 ). GPI forest headline indicators for Nova Scotia . GPI Atlantic Technical Report. http://www.gpiatlantic.org/pdf/forest/forestupdate.pdf Pearson , R. G. , C. T. Raxworthy , M. Nakamura , and A. T. Peterson ( 2007 ). Predicting species distributions from small numbers of occurrence records: A test case using cryptic geckos in Madagascar . Journal of Biogeography 34 : 102 – 117 . Google Scholar Crossref Search ADS WorldCat Phillips , S. J. , and M. Dudik ( 2008 ). Modeling of species distributions with Maxent. New extensions & a comprehensive evaluation . Ecography 31 : 161 – 175 . Google Scholar Crossref Search ADS WorldCat Phillips , S. J. , R. P. Anderson , and R. E. Schapire ( 2006 ). Maximum entropy modeling of species geographic distributions . Ecological Modelling 190 : 231 – 259 . Google Scholar Crossref Search ADS WorldCat Phillips , S. J. , M. Dudik , J. Elith , C. H. Graham , A. Lehmann , J. R. Leathwick , and J. S. Ferrier ( 2009 ). Sample selection bias and presence-only distribution models: Implications for background and pseudo-absence data . Ecological Applications 19 : 181 – 197 . Google Scholar Crossref Search ADS PubMed WorldCat Pimm , S. L. , G. J. Russell , J. L. Gittleman , and T. M. Brooks ( 1995 ). The future of biodiversity . Science 269 : 347 – 350 . Google Scholar Crossref Search ADS PubMed WorldCat Powell , L. L. , W. E. Glanz , and T. P. Hodgman ( 2014 ). Habitat occupancy of Rusty Blackbirds (Euphagus carolinus) breeding in northern New England, USA . The Condor: Ornithological Applications 116 : 122 – 133 . WorldCat Powell , L. L. , W. E. Glanz , T. P. Hodgman , C. Fisher , and J. Osenton ( 2010 ). Nest-site selection and nest survival of the Rusty Blackbird: Does timber management adjacent to wetlands create ecological traps? The Condor 112 : 800 – 809 . Google Scholar Crossref Search ADS WorldCat Reitsma , L. R. , M. Goodnow , M. T. Hallworth , and C. J. Conway ( 2009 ). Canada Warbler (Cardellina canadensis). In The Birds of North America Online ( A. Poole , Editor). Cornell Lab of Ornithology , Ithaca, NY, USA . https://birdsna.org/SpeciesAccount/bna/species/canwar Google Preview WorldCat COPAC Robertson , B. A. , and R. L. Hutto ( 2007 ). Is selectively harvested forest an ecological trap for Olive-sided Flycatchers? The Condor 109 : 109 – 121 . Google Scholar Crossref Search ADS WorldCat Robillard , C. M. , L. E. Coristine , R. N. Soares , and J. T. Kerr ( 2015 ). Facilitating climate-change-induced range shifts across continental land-use barriers . Conservation Biology 29 : 1586 – 1595 . Google Scholar Crossref Search ADS PubMed WorldCat Rowe , J. S . ( 1972 ). Forest regions of Canada . Department of the Environment, Canadian Forestry Service Publication Report No. 1300, Ottawa, ON, Canada . Google Preview WorldCat COPAC Rubio , L. , and S. Saura ( 2012 ). Assessing the importance of individual habitat patches as irreplaceable connecting elements: An analysis of simulated and real landscape data . Ecological Complexity 11 : 28 – 37 . Google Scholar Crossref Search ADS WorldCat Savard , J. P. L. , M. Cousineau , and B. Drolet ( 2011 ). Exploratory analysis of correlates of the abundance of Rusty Blackbirds (Euphagus carolinus) during fall migration . Ecoscience 18 : 402 – 408 . Google Scholar Crossref Search ADS WorldCat Sohl , T. L . ( 2014 ). The relative impacts of climate and land-use change on conterminous United States bird species from 2001 to 2075 . PLOS One 9 : e112251 . Google Scholar Crossref Search ADS PubMed WorldCat Spies , T. A. , B. C. McComb , R. S. Kennedy , M. T. McGrath , K. Olsen , and R. J. Pabst ( 2007 ). Potential effects of forest policies on terrestrial biodiversity in a multi-ownership province . Ecological Applications 17 : 48 – 65 . Google Scholar Crossref Search ADS PubMed WorldCat Statistics Canada ( 2017a ). Focus on geography series, 2016 census . Catalogue no. 98–404 X2016001. Statistics Canada , Ottawa, ON, Canada . https://www12.statcan.gc.ca/census-recensement/2016/as-sa/fogs-spg/Factspr-eng.cfm?Lang=Eng&GK=PR&GC=12&TOPIC=1 Google Preview WorldCat COPAC Statistics Canada ( 2017b ). Population and dwelling count highlight tables, 2016 census . Statistics Canada , Ottawa, ON, Canada . https://www12.statcan.gc.ca/census-recensement/2016/dp-pd/hlt-fst/pdpl/Table.cfm?Lang=Eng&T=801&SR=1&S=47&O=A&RPP=25 Google Preview WorldCat COPAC Stralberg , D. , S. M. Matsuoka , A. Hamann , E. M. Bayne , P. Sólymos , F. K. Schmiegelow , X. Wang , S. G. Cumming , and S. J. Song ( 2015 ). Projecting boreal bird responses to climate change: The signal exceeds the noise . Ecological Applications 25 : 52 – 69 . Google Scholar Crossref Search ADS PubMed WorldCat Van Proosdij , A. S. J. , M. S. M. Sosef , J. J. Wieringa , and N. Raes ( 2015 ). Minimum required number of specimen records to develop accurate species distribution models . Ecography 39 : 542 – 552 . Google Scholar Crossref Search ADS WorldCat Wang , L. , L. A. Kerr , N. R. Record , E. Bridger , B. Tupper , K. E. Mills , E. M. Armstrong , and A. J. Pershing ( 2018 ). Modeling marine pelagic fish species spatiotemporal distributions utilizing a maximum entropy approach . Fisheries Oceanography 27 : 571 – 586 . Google Scholar Crossref Search ADS WorldCat Warren , D. L. , and S. N. Seifert ( 2011 ). Ecological niche modeling in Maxent: The importance of model complexity and the performance of model selection criteria . Ecological Applications 21 : 335 – 342 . Google Scholar Crossref Search ADS PubMed WorldCat Weidensaul , S . ( 2000 ). Living on the Wind: Across the Hemisphere with Migratory Birds . North Point Press , Union Square West, New York, NY, USA . Google Preview WorldCat COPAC Weiss , A. D . ( 2001 ). Topographic Position and Landform Analysis . The Nature Conservancy , Northwest Division, Seattle, WA, USA . http://www.jennessent.com/downloads/tpi-poster-tnc_18x22.pdf Google Preview WorldCat COPAC Westwood , A. R . ( 2016 ). Conservation of three forest landbird species at risk: Characterizing and modelling habitat at multiple scales to guide management planning . Ph.D. thesis, Department of Biology, Dalhousie University, Nova Scotia, Canada. Google Preview WorldCat COPAC Westwood , A. R. , L. R. Reitsma , and D. Lambert ( 2017 ). Prioritizing areas for Canada Warbler conservation and management in the Atlantic northern forest of Canada . High Branch Conservation Services Technical Report , Hartland, VT, USA . Google Preview WorldCat COPAC Westwood , A. R. , C. Staicer , P. Solymos , T. Fontaine , E. M. Bayne , and D. Mazerolle ( 2019 ). Estimating the conservation value of protected areas in Maritime Canada for two species at risk: The Olive-sided Flycatcher (Contopus cooperi) and Canada Warbler (Cardellina canadensis) . Avian Conservation and Ecology 14(1):16. Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Yackulic , C. B. , R. Chandler , E. F. Zipkin , A. Royle , J. D. Nichols , E. H. Campbell Grant , and S. Veran ( 2013 ). Presence-only modelling with Maxent: When can we trust the inferences? Methods in Ecology and Evolution 4 : 236 – 243 . Google Scholar Crossref Search ADS WorldCat Yost , A. C. , S. L. Petersen , M. Gregg , and M. Miller ( 2008 ). Predictive modeling and mapping Sage Grouse (Centrocercus urophasianus) nesting habitat using maximum entropy and a long-term dataset from southern Oregon . Ecological Informatics 3 : 375 – 386 . Google Scholar Crossref Search ADS WorldCat Zurrell , D . ( 2017 ). Integrating demography, dispersal and interspecific interactions into bird distribution models . Journal of Avian Biology 48 : 1505 – 1516 . Google Scholar Crossref Search ADS WorldCat APPENDIX CREATION OF TOPOGRAPHIC POSITION INDEX For this study, we created a topographic position index that distinguished among 5 topographic positions (valleys, low slopes, mid slopes, up slopes, and ridges) using a methodology developed by Weiss (2001) and modified by Cooley (2014). This index was included as a candidate covariate in the development of Maxent models and was also used in the creation of the landscape complexity index. In brief, to create the topographic position index, we first obtained the 20-m provincial digital elevation model (DEM) (distributed by NSDNR; https://novascotia.ca/natr/meb/download/dp055.asp) as a base layer. We then ran the ArcMap Focal Statistics tool on the provincial DEM 3 times to generate layers that delineated the minimum elevation (DEMmin), maximum elevation (DEMmax), and mean elevation (DEMmean), respectively, in a 420-m2 neighborhood. A continuous TPI was subsequently calculated by inputting the following equation into the ArcMap “Raster Calculator” tool: (DEMmean – DEMmin) / (DEMmax – DEMmin). Following this, we increased the cell size of the continuous topographic position index from 20 m to 150 m (using the Resample tool) so that the resolution of the final topographic position index would be consistent with that of other environmental data layers. Finally, we applied the Reclassify tool to convert the continuous (150 m) index into a categorical one. For this, classes were defined according to standard deviation (STD) as follows: valleys (less than −1 STD), low slopes (−1 STD to −0.5 STD), mid slopes (−0.5 STD to 0.5 STD), up slopes (0.5 STD to 1 STD), and ridges (>1 STD) (Anderson et al. 2012). CREATION OF LANDSCAPE COMPLEXITY INDEX We derived landscape complexity scores according to 3 underlying indices and methods defined by Anderson et al. (2012): (1) landform variety, (2) elevation range, and (3) wetland density. Landform variety. Landforms describe natural surface features that are created by the topographic shape of the landscape (e.g., cliffs, valleys, slopes, coves). These features express local solar radiation and, even when climate is not considered, regulate species distributions edaphically through their control over erosion and deposition rates, soil texture and depth, and nutrient and moisture availability. A greater variety of landforms not only increases the likelihood of species’ persistence by providing a larger number of microclimates in an area, but also serves as a buffer against the effects of climate change (Dobrowski 2011, Anderson et al. 2012). To delineate landform variety, we used our 20-m topographic position index whose creation is described in the preceding paragraph. Originally, this index distinguished among 5 topographic positions: valleys, low slopes, mid slopes, up slopes, and ridgetops. However, the topographic position index employed by Anderson et al. (2012) (called a land position class index by those researchers) only distinguished among 4 topographic positions. Therefore, for the current research, we combined the up slopes and ridgetops from our original index into a single class. We opted to combine the top 2 topographic positions (i.e. as opposed to valleys and low slopes) because Nova Scotia is mostly flat and because a visual examination of the modified topographic position index overlaid on a hillshade DEM revealed that combining the higher topographic positions generally yielded a more accurate index than did combining the lower ones. We then used the DEM to create a 20-m slope map over Nova Scotia and combined this with the topographic position index to delineate 10 landform types, in accordance with Anderson et al. (2012). Specifically, these were cliff or steep slope (any area with a slope of 24–90°), slope crest (ridgetop/up slope with a slope of 6–24°), flat ridgetop (ridgetop/up slope with a slope of 0–6°), rounded ridge (mid slope with a slope of 6–24°), gentle hill (mid slope with a slope of 2–6°), hilltop (mid slope with a slope of 0–2°), lower sideslope (low slope with a slope of 6–24°), toe slope (low slope with a slope of 2–6°), flats (low slope with a slope of 0–2°), and cove or slope bottom (valley with a slope of 0–35°). Slopes were then further subdivided according to aspect (i.e. northeast or southwest), and flats were further subdivided by flow accumulation (i.e. wet or dry), for a total of 12 landforms. Subsequently, we performed a focal variety analysis to determine the number of landforms within a 27-ha circular neighborhood. (This number was selected so that the ratio of cell size to neighborhood size matched that used by Anderson et al. 2012). Elevation range. As climate changes, species ranges may shift, increase, or decrease in concert with elevation. This is particularly true in mountainous or hilly landscapes, where slopes can magnify the effects of a changing climate (Anderson et al. 2012). To determine local elevation ranges within Nova Scotia, we performed a focal range analysis on the same 27-ha circular neighborhood as that used in the aforementioned landform variety analysis. As elevation range values were highly skewed toward zero, we log-transformed this data prior to further analysis. Wetland density. A large portion of Nova Scotia is wet and flat as a result of past glaciations. In flat areas, Anderson et al. (2012) suggest that using landform variety and elevation range alone is insufficient to delineate landscape complexity. They therefore recommend that wetland density be used to infer micro-scale topographic diversity and freshwater accumulation patterns, positing that areas of high wetland density should have higher levels of topographic variation, and that small, isolated wetlands are at greater risk of shrinkage or disappearance. To create a wetland density index, we calculated the density of wetlands within 27-ha and 270-ha neighborhoods (with this size again selected to match the ratios used by Anderson et al. 2012) using the provincial wetland inventory (provided by NSDNR, obtained via the Dalhousie University GIS Centre). We then combined these fine- and coarse-scale wetland density values to obtain a single wetland density index, whereby the fine-scale values were given twice the weight of the coarse-scale values. Finally, the landform variety, elevation range, and wetland density indices were normalized by Z-score and combined (using the equation shown below) to obtain a final landscape complexity index (Anderson et al. 2012). However, in creating this index, note that wetland density values were only applied to relatively flat landforms (i.e. landforms with a slope of 0–6°: flats, flat hilltops, and flat ridgetops), where wetlands could reasonably be expected to occur. Landscape complexity: Flat areas = (2*Landform variety + 1*elevation range + 1*wetland density) / 4 Incline areas = (2*Landform variety + 1*elevation range) / 3 Copyright © American Ornithological Society 2020. All rights reserved. For permissions, e-mail: journals.permissions@oup.com. 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)
TI - The benefits of using topographic features to predict climate-resilient habitat for migratory forest landbirds: An example for the Rusty Blackbird, Olive-sided Flycatcher, and Canada Warbler
JF - Condor: Ornithological Applications
DO - 10.1093/condor/duz057
DA - 2020-03-10
UR - https://www.deepdyve.com/lp/oxford-university-press/the-benefits-of-using-topographic-features-to-predict-climate-X1QM6jAon2
SP - 1
VL - Advance Article
IS - 
DP - DeepDyve
ER -