The declining trend of Arctic September sea ice constitutes a significant change in the Arctic climate system. Large year- to-year variations are superimposed on this sea–ice trend, with the largest variability observed in the eastern Arctic Ocean. Knowledge of the processes important for this variability may lead to an improved understanding of seasonal and long-term changes. Previous studies suggest that transport of heat and moisture into the Arctic during spring enhances downward surface longwave radiation, thereby controlling the annual melt onset, setting the stage for the September ice minimum. In agreement with these studies, we find that years with a low September sea–ice concentration (SIC) are characterized by more persistent periods in spring with enhanced energy flux to the surface in forms of net longwave radiation plus turbulent fluxes, compared to years with a high SIC. Two main atmospheric circulation patterns related to these episodes are identified: one resembles the so-called Arctic dipole anomaly that promotes transport of heat and moisture from the North Pacific, whereas the other is characterized by negative geopotential height anomalies over the Arctic, favoring cyclonic flow from Siberia and the Kara Sea into the eastern Arctic Ocean. However, differences between years with low and high September SIC appear not to be due to different spring circulation patterns; instead it is the persistence and intensity of processes associated with these patterns that distinguish the two groups of anomalous years: Years with low September SIC feature episodes that are consistently stronger and more persistent than years with high SIC. Keywords Climate variability · Arctic sea ice · Self organizing maps (SOMs) · Atmospheric circulation · Atmospheric energy transport 1 Introduction The Arctic sea–ice extent has declined significantly during recent years, with the largest reduction in September (Serreze and Stroeve 2015; IPCC 2013). Superimposed on this trend is a large inter-annual variability; for example, the ice extent in Electronic supplementary material The online version of this September 2013 and 2014 was about 1.7 million km higher article (https ://doi.org/10.1007/s0038 2-018-4279-z) contains supplementary material, which is available to authorized users. than the record minimum in 2012, when September sea ice was reduced to 3.6 million km . It has been suggested that * Marie-Luise Kapsch about half of the trend can be attributed to rising surface-air firstname.lastname@example.org temperatures, as a direct result of increasing atmospheric Department of Meteorology and Bolin Centre for Climate greenhouse gases (Kay et al. 2011; Burt et al. 2016). Other Research, Stockholm University, 10691 Stockholm, Sweden processes contributing to the recent sea–ice trend include Max-Planck Institute for Meteorology, Bundesstraße 53, indirect effects of global warming and internal variability of 20146 Hamburg, Germany the climate system (Kay et al. 2011; Döscher et al. 2014), for Department of Marine and Coastal Sciences, Rutgers example, cloud cover anomalies (Eastman and Warren 2010), University, 71 Dudley Road, New Brunswick, NJ 08901, variability of ocean currents (Comiso et al. 2008), and shifts USA and variability of atmospheric circulation patterns (e.g. Over- Department of Physics and Technology, University land et al. 2008; Ogi and Wallace 2012; Ding et al. 2017). In of Tromsø, Postbox 6050, Langnes, 9037 Tromsø, Norway Vol.:(0123456789) 1 3 M.-L. Kapsch et al. order to interpret trends and variability of sea ice it is hence 2 Data and methods important to also understand the atmospheric and oceanic variability in the Arctic. Here, we focus on the atmospheric 2.1 Reanalysis data variability. In this study we investigate the atmospheric processes dur- All data used in this study come from the ERA-Interim rea- ing years with anomalously low summer sea ice since 1979. nalysis by the European Center for Medium-Range Weather Recent studies have pointed to the specific importance of Forecast (ECMWF; Dee and Uppala 2009; Dee et al. 2011) atmospheric conditions during spring and early summer for the for 1979–2012, at a 0.5°x0.5° horizontal grid resolution. seasonal sea–ice evolution (Eastman and Warren 2010; Dev- Despite known shortcomings, ERA-Interim is acknowledged asthale et al. 2013; Kapsch et al. 2013, 2016; Cox et al. 2016). as one of the best reanalyzes in representing the Arctic cli- Using a 1-year observational data set, Persson (2012) found mate (e.g., Jakobson et al. 2012; Kapsch et al. 2013—Sup- that although melt onset in the Arctic does not occur before plementary Information; Lindsay et al. 2014). We use verti- May, significant surface warming associated with atmospheric cally integrated cloud water, near-surface air temperatures processes that increase the energy flux to the surface can occur and winds, and geopotential heights averaged daily from the months before the melt. Kapsch et al. (2013) showed that posi- 6-hourly ERA-Interim analyses. Daily averaged SICs are tive anomalies of clouds and atmospheric water vapor are pre- also taken from ERA-Interim; note, however, that SICs are sent over the part of the Arctic that exhibits the largest sea–ice not a model product in ERA-Interim, but are prescribed from variability in spring of years with anomalously low September independent analyses, based on space-borne sensors (Dee sea–ice concentrations (SICs). In years with relatively high et al. 2011). For radiation fluxes, evaporation, and precipi- SICs, the opposite situation prevails. Positive anomalies of tation we use accumulated values from 24-h ERA-Interim clouds and water vapor in spring lead to enhanced energy forecasts initiated at 00 UTC. flux to the surface, specifically in the form of more downward The atmospheric energy transport is calculated at longwave radiation (LWD), and an earlier melt onset (Mortin 6-hourly resolution, based on the model’s hybrid levels and et al. 2016; Lee et al. 2002). These cloud and water–vapor corrected for a mass-flux inconsistency in ERA-Interim anomalies are linked to anomalous moisture transport into the (Trenberth 1991; Graversen et al. 2007). The energy trans- Arctic, indicating a remote origin (Kapsch et al. 2013; Mortin port is separated into its dry-static and latent components: et al. 2016). Here, we investigate whether positive anomalies of the 1 1 = + c T + gz d (1) energy flux to the surface are associated with certain atmos- g 2 pheric circulation patterns that drive anomalous transport of moisture into the region of largest sea–ice variability (Kapsch et al. 2013). For the energy flux, we specifically focus on the net longwave radiation and turbulent fluxes (LWNT) during = Lq d, (2) spring, prior to melt onset. The circulation patterns are identi- fied based on a Self-Organizing-Maps (SOMs; e.g. Kohonen 1982, 2001; Skific and Francis 2012) algorithm. We seek where v(u,v) is the horizontal wind vector, c is the specific to answer two major questions: (1) are there specific spring heat capacity of moist air at constant pressure, T is the tem- atmospheric circulation regimes that favor atmospheric mois- perature, gz is the geopotential height, L is the latent heat of ture or heat transport into the Arctic in years with a low Sep- vaporization, q is the specific humidity, and η is the vertical tember SIC? and (2) are there systematic differences in the model hybrid coordinate. The transports are vertically inte- character, e.g., duration and strength, of spring atmospheric grated over the entire atmospheric column. Convergences of transport events in years with a low September SIC versus latent (ConLE) and dry-static energy (ConDE) are calculated years with a high SIC? In order to approach these questions, from horizontal divergence of in Eq. (1) and in we present a characterization of individual events, attempting Eq. (2). to provide a step forward in our understanding of important processes governing the large variability of the ice cover. 2.2 Determination of years with extreme sea–ice conditions Figure 1 shows the time evolution of the September SIC for 1979–2012, averaged over the area where September SIC variability is the largest (Fig. 2; Kapsch et al. 2013); this area is shown in Fig. 2, extending from the Beaufort Sea over the East Siberian and Laptev Seas towards the Kara Sea (105°E 1 3 Summers with low Arctic sea ice linked to persistence of spring atmospheric circulation patterns to 150°W and 74°N to 84°N). This area is hereafter referred to as the investigation area. Because the intention here is to explore the atmospheric dynamics for years with anoma- lously low summer sea–ice conditions, the SIC variability is separated from its trend. A 5-year running mean over the SICs from 1979 to 2014 provides a reference SIC climatol- ogy, yielding the SIC anomalies when subtracted from the original SIC data (Table 1). Note, that in the first and last 2 years of the time series, the 5-year reference climatology is a weighted mean over the adjacent years, constructed such that the anomalies add up to zero over the time series (i.e., 1 1 x ̄ = 2x + 2x + x ; x ̄ = [2x + x + x + x ]). 1 1 2 3 2 1 2 3 4 5 5 Applying a 5-year running mean as reference climatol- ogy allows for an investigation of the year-to-year variabil- ity without a priori specification of the characteristics of the underlying trend (e.g., whether it is linear or of higher Fig. 1 September sea–ice concentration (SIC) in percent for the order). Years with an anomalously low (high) SIC are deter- years 1979–2012. September SICs averaged over the investigation mined as years that fall below (above) one standard devia- area (black) are shown together with the 5-year running mean of the SIC time series (solid red). Dashed lines indicate where the SICs tion of the residual anomaly time series. This results in five depart ± 1 standard deviation from the running mean values. Black years with anomalously low (hereafter referred to as LIYs) dots mark years with anomalously low or high September SICs (see and 4 years with anomalously high SICs (similarly referred Sect. 2.2 for further details). Note that the running mean time series is to as HIYs; Fig. 1). Tests with a somewhat longer averaging based on SICs from 1979 to 2014 period than 5 years for the reference climatology change Fig. 2 Sea ice conditions during years with an anomalous low September sea–ice concentration (SIC). September SICs (color) and the ice-edge (SIC > 15%) for the respective 5-year climatology (black line) are shown. The black box indicates the investigation area 1 3 M.-L. Kapsch et al. Table 1 Characteristics of episodes with positive longwave net radiation plus turbulent flux anomalies (LWNT), the remaining days during spring (Mar 1 to melt onset, excluding LWNT episodes) and spring of years with a low/high September sea–ice concentration (LIYs/HIYs) Year Number of Duration LWNT anomaly LWNT anomaly LWNT anomaly September sea–ice LWNT epi- of episodes during episodes of during remaining during spring mul- concentration sodes [days] positive LWNT mul- spring days multi- tiplied by # spring anomaly [%] 8 —2 tiplied by duration plied by # of days days [10 Jm ] 8 −2 8 −2 of events [10 Jm ] [10 Jm ] LIYs 1990 14 62 40 − 2 48 − 10.2 1995 13 56 32 − 9 31 − 9.3 2003 9 32 10 − 7 18 − 11.5 2007 9 39 17 − 19 7 − 19.4 2012 8 33 8 − 19 − 10 − 18.2 HIYs 1992 6 21 5 − 32 − 16 9.8 1996 8 37 16 − 31 − 4 14.7 2001 5 14 1 − 31 − 29 19.1 2006 6 34 10 − 18 − 5 20.1 All years 1979–2012 8 34 10 − 19 − 3 − 0.53 All values are averages over the area indicated by the black box in Fig. 2. The last row shows the characteristics for all years between 1979 and 2012; note, the values are normalized by the number of years included in the time period. The last column presents the values for the Northern Hemisphere sea–ice concentration anomalies as shown in Fig. 3 Values in bold indicate statistical significance. A one-sided student’s t test was applied (α = 0.05), with the null-hypothesis that individual values are larger (LIYs) or smaller (HIYs) compared to the distribution of all years The values are normalized by the number of years the classification only slightly; the most extreme LIYs are Statistical significance of anomalies in LWNT and other identical. The five identified LIYs are 1990, 1995, 2003, variables presented in Fig. 3 and S1 are determined using a 2007 and 2012 (Fig. 1); Fig. 2 shows the September SIC for one-sided student’s t test (α = 0.05), with the null-hypothesis each of these. The four HIYs are 1992, 1996, 2001 and 2006. that the anomalies are not significantly different from the respective climatology, unless stated otherwise. 2.3 Episodes of an anomalous surface energy budget 2.4 Melt onset calculation Anomalies of each variable considered in the surface-energy The melt onset dates for LIYs and HIYs are given in Table 2. budget as well as the energy transports are obtained by The melt onset is determined as the first day in spring when removing the annual cycle. The annual cycle is calculated the 14-day running median of the near-surface air tem- for each grid point using daily resolution data. First, we cal- perature exceeds − 1 °C (Rigor et al. 2002; Persson 2012). culate a 5-year average for each day of a given year, using Melt onset is calculated for each grid point with more than the same method as applied for the SIC (Sect. 2.2). Second, 80% sea ice for at least 98% of the days in January through a 30-day running mean is applied on the annual cycle of each April (for details see Mortin et al. 2014). In the following year to suppress inter-monthly variability. The anomalies are we define the date of melt onset over the investigation area then obtained as the residual of the actual time series and the as the date when half of the grid boxes in the area have annual cycle climatology. Finally, the anomaly (the residual) experienced melt onset, hereafter referred to as median melt is averaged over the investigation area. The reanalysis data onset (Table 2). The spatial distribution of the melt onset used for this study comprise the years 1979–2012. The final date anomaly for each of the LIYs is displayed in Fig. 4. time series of the anomalies for all the identified LIYs and The anomalies for each year and grid point are relative to a HIYs are displayed in Fig. 3 and Figure S1 in the Supple- 5-year reference climatology, as described in Sect. 2.2. mentary Material, respectively. We only consider episodes of positive LWNT anomalies 2.5 Self‑organizing maps (SOMs) between the first of March and melt onset, defined below. If several individual episodes appear, such as for example in To identify atmospheric circulation patterns, a Self-Organ- late March and early April of 1990 (Fig. 3), each episode is izing Map (SOM) algorithm is used. The process of SOM treated individually. formation starts by creating an initial set of reference vectors 1 3 Summers with low Arctic sea ice linked to persistence of spring atmospheric circulation patterns Fig. 3 Atmospheric variables and anomalies for the five LIYs. Shown are anomalies of positive net longwave radiation plus turbulent fluxes (LWNT; red), net shortwave radiation (SWN; green) and convergence of latent (ConLE; blue) and dry-static energy (ConDE; cyan). The flux anomalies are positive if downward to the surface. SICs are shown for the respective year (thick black) together with the 5-year climatology (thin black). Days with significantly larger LWNT anomalies and smaller SICs are indicated with horizontal lines (α = 0.05; red and black, respectively). Units for LWNT and SWN are in −2 Wm , for ConQE and ConDE −1 −2 in 10 Wm and for SICs in %. A 14-day low-pass fft-filter is applied to all time series for visualization. Vertical lines indicate the begin of melt onset (BMO), median melt onset (MMO), and end of melt onset (EMO; see Table 2; Sect. 2.4), respectively 1 3 M.-L. Kapsch et al. Table 2 Melt dates for years with low (LIY) and high (HIY) Septem- that best describes a 3-dimensional data sample. In the train- ber sea–ice concentrations ing stage, these reference vectors are created by placing the first and second empirical orthogonal function (EOF)-based Years BMO [DOY] MMO [DOY] EMO [DOY] vectors in the diagonally opposite corners of the map matrix, LIYs 1990 128** (144) 146** (155) 157** (164) then interpolating the remaining space between them. Dur- 1995 143** (146) 152** (155) 156** (164) ing the training, the vectors adjust depending on the Euclid- 2003 148* (146) 152 (152) 157** (158) ean distance from the data sample; the closer the vectors are 2007 139 (140) 150 (149) 162* (157) to the data sample, the more they are adjusted. The training 2012 138** (141) 143** (149) 151** (154) is complete once the set of vectors is found that minimizes HIYs 1992 152* (144) 161* (154) 176* (164) the distances and best describes the given data. An important 1996 144** (147) 160* (156) 166* (163) feature of SOMs is that the initialization of a map, using 2001 151* (147) 159* (154) 162* (159) EOFs, does not precondition its final form, but rather speeds 2006 136** (141) 143** (150) 153** (159) up the training process compared with random initialization. Moreover, one advantages of SOMs over EOFs is that the Melt dates are averaged over the area specified in Fig. 2. Shown are the days of the year (DOY) for the median melt onset (MMO; see process of training allows for non-linear features of the com- Sect. 2.4) as well as the begin of melt onset (BMO) and end of melt plex climate system to be captured, thus providing a more onset (EMO). BMO is defined as the date on which melt onset has detailed and more realistic view of the data’s true nature occurred in all grid boxes of the area, and EMO as the date on which (Reusch et al. 2005). More details on the map training can melt onset has occurred in all grid boxes of the area. Please refer to Sect. 2.4 for details be found in Skific et al. (2009), Skific and Francis (2012) Statistically significant earlier (**) and later (*) melt onset; values in and Hewitson and Crane (2002). parenthesis display respective 5-year climatology The representation of reference vectors on a 2-dimen- sional array according to their similarity is a great advan- tage in atmospheric sciences, because to an experienced meteorologist, this representation is reminiscent of Fig. 4 Melt onset date anomalies during LIYs. The black box indicates the investigation area (see Sect. 2.4 for details) 1 3 Summers with low Arctic sea ice linked to persistence of spring atmospheric circulation patterns familiar atmospheric circulation features. The interpreta- 3 Results tion of the SOMs then becomes intuitive as well as quanti- tative, as one can observe dominant features or transitions 3.1 Sea–ice conditions during years of low between various states (Hewitson and Crane 2002). These September sea ice unique aspects of SOMs help to explain climate variability and reveal several inherent properties of the climate sys- Except for 2003, the September SIC for the entire Northern tem: first, they provide structures that are coherent, not a Hemisphere is anomalously low during all LIYs, which are set of discrete unrelated patterns (Sheridan and Lee 2011); identified with respect to the investigation area (Fig. 2). second, they reveal structures that can also be highly com- This indicates that the investigation area captures a large plex and non-linear. part of the Arctic-wide interannual sea–ice variability. In this study, the SOM training and analysis are based Most of the SIC anomalies during LIYs are centered in the on daily spatial anomalies of 850 hPa geopotential heights Laptev, East Siberian, Chukchi, and Beaufort Seas. During during spring (defined as March 1 to melt onset) of the LIYs, the September SIC anomalies in the investigation years 1979–2012. The use of geopotential anomalies, area are between − 9 and − 19% (Fig. 1; Table 1). Note that rather than absolute values, ensures that each SOM cluster owing to the trend of SIC in recent years (e.g., Serreze and represents purely circulation-related features rather than Stroeve 2015), the 5-year reference climatology of SIC for features tied to the absolute magnitude of the geopoten- 2007 and 2012 is significantly reduced relative to earlier tial. The daily anomalies are calculated by subtracting the periods along the coasts of the East Siberian, Chukchi and geopotential at each grid point from the domain-averaged Beaufort Seas (Fig. 2). For example, in September 2007 geopotential for each daily field. Before the SOM training most of the East Siberian Sea and parts of the Beaufort Sea process, geopotential height anomalies are re-gridded from were ice-free. Yet SIC anomalies in the investigation area the original 0.5° × 0.5° to a 250 km × 250 km Equal-Area still fall more than 18% below the 5-year reference values Scalable Earth Grid (EASE). Interpolation to an equal- for 2007 and 2012 (Table 1). area grid ensures equal weighting of the grid boxes in the All LIYs exhibit an earlier melt onset than the 5-year SOM training. Also, it reduces the size of the input data, reference climatology over the investigation area or parts which speeds up the processing. Further, all grid points thereof (Fig. 4). Earlier melt onset over the investigation over Greenland are removed, as the high topography leads area is most significant in 1990, occurring on average to permanent spatial anomalies in the geopotential fields about 9 days earlier (Table 2). In 2003 and 2007 the area- often unrelated to the dynamic anomalies in the flow field, averaged melt onset is not significantly different from the thereby creating unrealistic features in the SOMs. The 5-year mean climatology. In contrast, for HIYs the Sep- resulting set of vectors is displayed in a SOM matrix (here- tember SIC anomalies are 10–20% higher than the respec- after referred to as master SOM; see Skific and Francis tive 5-year reference climatology over the investigation 2012), defined as a 12-cluster matrix in order to minimize area (Table 1; Fig. 1). Also, the melt onset is significantly the size of the matrix while still capturing the essential later for all HIYs except for 2006 (Table 2). Hence these variability of the data set. results indicate that the melt onset plays a role for the In order to investigate the atmospheric conditions dur- September sea–ice conditions. ing episodes of enhanced LWNT, atmospheric variables are “mapped onto” each cluster in the master SOM. For the mapping, each day of the investigation period is assigned 3.2 Spring atmospheric conditions during years to the best-match SOM cluster, determined by finding the of low September sea ice SOM pattern with the smallest Euclidean distance of the 850 hPa geopotential height anomaly field. Geopotential As displayed in Fig. 3, spring and early summer of LIYs height fields (not anomalies) during spring of all years are are often characterized by episodes with anomalously pos- mapped onto the master SOM and displayed in Fig. 5 to itive LWNT over the investigation area. In the five LIYs, characterize each circulation pattern. Other mapped vari- between 8 and 14 episodes with enhanced LWNT occur, ables include anomalies of latent and dry-static energy trans- compared to 5 and 8 for HIYs; the average over all years port and convergence, cloud cover, radiation and winds. This is 8 (Table 1). The length of the individual events varies feature of SOM analysis allows for a thorough exploration of between 1 and 7 days (not shown). The LWNT episodes the characteristics of each cluster (Skific and Francis 2012). in LIYs are associated with an average LWNT anomaly −2 Further, atmospheric variables used to calculate LWNT epi- of approximately 2 GJm during a total of 5–9 weeks sodes during LIYs and HIYs are mapped onto the SOM throughout the entire spring period, compared to less than −2 clusters individually to identify differences and similarities 1 GJm for HIYs during 2–4 weeks. During this time of between the episodes in the respective years. 1 3 M.-L. Kapsch et al. Fig. 5 Circulation patterns during spring. Shown are 850 hPa geo- 1979–2012 (see Sect. 2.5). The number on the top right of each map potential heights. The patterns are clustered and organized accord- gives the relative frequency each of these patterns occurred through- ing to the Self-Organizing Map algorithm applied on geopotential out 1979–2012 (3240 days in total) height anomalies during spring (March 1 to melt onset) of the years the year, LWNT anomalies dominate the surface energy towards thin first-year ice. Thinner ice melts more easily and budget, as anomalies of shortwave net radiation (SWN) is more prone to wind forcing, hence less energy is needed contribute little to the surface energy budget owing to a to cause signic fi ant ice anomalies. Moreover, as the ice thins high surface albedo and, in the beginning of the season, and the SIC declines, surface temperatures increase, which low solar inclination. Changes in LWNT are associated leads to an increased upward longwave flux and reduced with variations in surface and near-surface air temperature, positive net longwave radiation. water–vapor content, and cloud cover. Variations in water Figure 6 shows the lag correlation between anomalies in vapor and clouds alter the emissivity of the atmosphere LWNT and the convergence of moisture advection (ConLE). and hence LWD at the surface. During LIYs, LWNT anomalies occur in coherence with Fewer LWNT episodes tend to occur over the LIYs significant positive anomalies of ConLE, although with a during the latter part of these years; the number and dura- lag in the order of one day; the estimated lag is sensitive tion of LWNT episodes are statistically different from the to the length of the anomalies. Extended episodes of posi- 1979–2012 average only for the years 1990, 1995 and 2007 tive anomalies of ConDE also prevail throughout spring (Table 1). One factor contributing to this decrease could of LIYs, but are less correlated to the LWNT anomalies in be related to a shortening of the spring season owing to a most of the years (not shown). This suggests that anoma- significantly earlier melt onset (Stroeve et al. 2014). Further, lies of LWNT are mostly affected by anomalies of ConLE, the LWNT anomalies during the LWNT episodes become rather than by ConDE. Only 2007 displays a significant weaker with time, while sea–ice concentration anomalies correlation between LWNT and ConDE; 0.35 for a 3-day become larger. This is consistent with an Arctic regime shift lag (not shown), indicating that ConDE anomalies played 1 3 Summers with low Arctic sea ice linked to persistence of spring atmospheric circulation patterns the circulation patterns during episodes of positive LWNT anomaly over the investigation area during LIYs, we map potential contributors to LWNT anomalies onto the mas- ter SOM. These are the surface energy fluxes, the conver - gence of moisture and heat, and the anomalies of integrated cloud water and atmospheric moisture. We do this only for spring days in LIYs with a positive LWNT anomaly over the investigation area; see Fig. 7. Note that by this construc- tion, anomalies in LWD and LWN are positive for all CPs (Fig. 7c, g). Also, CP7 did not occur throughout spring of any of the LIYs. In Fig. 8 we also explore the frequency of occurrence and the persistence of the CPs; only the most frequently occurring CPs will be discussed. Approximately 44% of the days during LWNT episodes in LIYs are associated with low geopotential heights cen- tered near the Russian coast with high values over the North Fig. 6 Correlations between anomalies of the net longwave radiation American side of the Arctic (CP1–CP4; Fig. 5). Such a pat- plus turbulent fluxes (LWNT) and the latent (solid) energy conver - tern is often referred to as the Arctic dipole anomaly (AD; gence as function of time lag. The correlations are calculated for the Overland and Wang 2005). The AD favors flow from the spring period (March 1 to melt onset) after averaging over the inves- tigation area (see Fig. 2). Values above/below the horizontal lines North Pacific across the central Arctic towards the Atlan- (black, dashed) are statistically significant correlations (α = 0.05) tic and is associated with northward transport of heat and moisture into the investigation area, causing ConLE and a larger role in the surface-energy budget during spring of ConDE anomalies to become positive (Fig. 7a, b). This 2007 than in other LIYs. In 2007, however, an early melt leads to an increase of clouds and anomalously high sur- onset occurred only in a small part of the investigation area face temperatures (not shown) over the investigation area, (Fig. 4), suggesting that melt-related processes during spring thereby creating positive anomalies of LWD and LWN at contributed less to the 2007 September sea–ice minimum the surface (Fig. 7c, f, g). Note, however, that certain varia- than did other processes. This is consistent with findings tions in the AD evident in CP1–CP4 may transport air from by Graversen et al. (2011), who suggested that latent and the cold Asian continent rather than the warm North Pacific dry-static energy transports into the Arctic during summer into the investigation area, leading to negative surface flux (June to August), rather than in spring, contributed to the anomalies. sea–ice anomaly in 2007. Zhang et al. (2008) and Kauker For clusters CP5–CP12 geopotential heights are also low et al. (2009) argued that ice dynamics may also have been along the Russian coast but, unlike AD patterns, the low important. Kauker et al. (2009) found that a large part of the geopotential heights extend far west across the Norwegian sea–ice reduction in 2007 can be attributed to an anoma- Sea, Fram Strait, Greenland, and the Canadian archipelago lously low ice-thickness in March, increased wind stress (Fig. 5). The detailed structure differs between these CPs, on ice in May and June, as well as relatively warm sea- as the spatial extent of the geopotential heights varies. The surface temperatures throughout September. A warming of associated circulation of most of these patterns results in a the sea–ice layer during spring owing to increased LWNT westerly flow into the investigation area, with a southwest- might, however, have pre-conditioned the ice and altered its erly component in patterns with low geopotential heights sensitivity to other atmospheric and oceanic forcing later that extend farther inland over Russia (Fig. 5). The westerly in the year. flow brings relatively moist air from Russia and the Kara Sea into the western part of the investigation area, where it leads 3.3 Atmospheric circulation associated with LWNT to positive anomalies in ConLE (Fig. 7a). Only for CP8 are episodes during LIYs the ConLE anomalies negative over the investigation area, while ConDE anomalies are very small or negative for most In this section the characteristic atmospheric circulation pat- of these patterns (Fig. 7b). The fact that almost all patterns terns (CPs) during episodes of enhanced LWNT during LIYs during LWNT episodes during LIYs are associated with are explored. The objective is to ascertain whether common positive ConLE anomalies supports earlier findings sug- atmospheric conditions prevailed during these episodes of gesting that moisture convergence triggers positive LWNT positive LWNT anomalies. Several distinct CPs prevail; anomalies (see Sect. 3.2). we refer to these patterns according to their position in the Figure 8 shows the timing in spring of the different pat- SOM matrix (CP1–CP12; Fig. 5). To specifically focus on terns (Fig. 8a) and their persistence (Fig. 8b). Note that CP9 1 3 M.-L. Kapsch et al. Fig. 7 Anomalies of (a) latent and (b) dry-static energy convergence, positive LWNT episodes in spring of LIYs for the circulation patterns (c) downward longwave radiation, net (d) shortwave radiation, (e) presented in Fig. 5. Values are averaged over the area indicated by the cloud water and (f) water vapor and (g) net longwave radiation during black box in Fig. 5 1 3 Summers with low Arctic sea ice linked to persistence of spring atmospheric circulation patterns Fig. 8 Frequencies of occur- rence and persistence of the circulation regimes presented in Fig. 5 during LIYs. Top: Fre- quencies of occurrence during episodes of enhanced longwave net radiation plus turbulent fluxes (LWNT; black) and remaining spring days (RSDs; red) as function of month. Bot- tom: Persistence of each pattern for LWNT (black) and RSDs (red) of LIYs. Horizontal lines indicate the median persistence, the edges of the boxes the 25th and 75th percentiles and whisk- ers the 5th and 95th percentiles and CP10 occur relatively early in spring of LIYs, likely consecutive days, giving them a longer time to act on the causing an earlier preconditioning of the surface (Fig. 8a). surface. Consequently, their impact is likely larger than for The most persistent circulation patterns during LWNT the other patterns. In contrast to these relatively persistent episodes in spring of LIYs are CP1, CP3, CP5, and CP10 patterns, CP2, CP6, CP7 and CP11 are transition patterns (Fig. 8b). Their median persistence exceeds 3 consecu- that occur only rarely during LIYs, and when they do occur tive days and most of these patterns can last more than 5 they are less persistent than other patterns. 1 3 M.-L. Kapsch et al. While similar circulation patterns do occur during LWNT (Fig. S3 in the Supplementary Material). Three of these pat- episodes and during the remaining spring days (hereafter terns also occur earlier in spring, resulting in an earlier pre- RSDs; days that fall outside the LWNT episodes in spring conditioning of the surface in LIYs and, hence, an earlier of LIYs), there are significant differences in frequencies and melt onset (Fig. 4; Tables 1, 2). Owing to the earlier melt persistence of the patterns (Fig. 8). Specifically, CP1, CP3 onset, feedback mechanisms have longer time to act on the and CP10 occur more frequently during LWNT episodes surface. Note that an earlier melt onset, leading to increased than during RSDs of LIYs; 45 and 24%, respectively. These absorption of solar radiation throughout the spring and sum- patterns are also more persistent during LWNT episodes mer season, was found to be one of the major drivers of than during the RSDs (Fig. 8b). In contrast, CP5 and CP12 differences in annual ice evolution (e.g., Wang et al. 2016). occur less frequently during LWNT episodes than during We also explore possible associations between the occur- RSDs; 17 and 29%, respectively. Note, that ConLE and rence of the LWNT episodes and differing phases of the ConDE anomalies are very small or negative for CP9 and Arctic Oscillation (AO) by mapping the AO-index onto CP12, which suggests that the positive anomalies of LWNT the SOM matrix in Fig. 10. The AO index is defined as the that are assigned to these CPs are not associated with the leading principal component of the Northern Hemisphere convergence of moisture and dry-static energy. Further, sea-level pressure (Thompson and Wallace 1998). A posi- almost all patterns show larger positive anomalies of clouds, tive index (AO+) represents lower-than-average sea-level water vapor, and ConLE during LWNT episodes than during pressure over the Arctic. We find that CP4, CP6, CP7 and the RSDs (Fig. S2 in the Supplementary Material), indicat- CP9-CP11 are associated with a positive phase of the AO ing that the same patterns have a different impact on the during spring (Fig. 10a). During LWNT episodes in spring surface during the RSDs. of LIYs, 7 out of the 11 occurring circulation patterns are associated with AO+ (CP3, CP4 and CP8 CP12; Fig. 10b); 3.4 Differences in the atmospheric circulation during LWNT episodes in spring of HIYs only 3 patterns between LIYs and HIYs are associated with AO+ (CP7, CP9 and CP10; Fig. 10c). These results indicate that the patterns in the bottom part Differences in SIC, melt onset, and frequency and intensity of the SOM matrix are in general associated with AO+. of the LWNT episodes are evident for the LIYs and HIYs Further, a larger positive AO in CP3, CP4, CP9 and CP12 (Figs. 3, 9, and S1). The five LIYs are associated with 8–14 during LWNT episodes in LIYs rather than those in HIYs episodes of positive surface LWNT anomalies, resulting in suggests a link between the positive phase of the AO and −2 an average energy surplus 0.8–4.0 GJm over 32–62 days the frequency of occurrence of episodes of enhanced LWNT during spring (Table 1). In contrast, HIYs are character- during spring (see also Fig. 9). Note, however, that the AO ized by fewer and shorter positive LWNT episodes. The index is based on pan-Arctic sea-level pressure anomalies four identified HIYs exhibit between 5 and 8 events that are and is not restricted to the investigation area. Hence, this −2 associated with a smaller energy surplus of 0.1–1.6 GJm association between the circulation patterns and the AO and that span fewer days (20–49 days) as compared to LIYs. might only partly explain the processes behind the more Further, differences in the frequencies and timing of frequent occurrence of LWNT episodes during LIYs over the atmospheric circulation patterns between the two sets the investigation area. of years are found for the LWNT episodes in spring (see Fig. 9). Interestingly, LWNT episodes during LIYs are often characterized by less frequent occurrence of CP4 (an AD pattern) as compared to the HIY LWNT episodes, while 4 Discussion and conclusions CP9–CP12 occur more frequently (Fig. 9a, b). We find that AD-like circulation patterns (CP1–CP4) prevail on about Five years that exhibit significant negative September SIC 44% of the LWNT episodes in LIYs and 49% of the days anomalies over the Laptev, East Siberian and Beaufort Seas in HIYs (Fig. 9). CP9 to CP12 occur during 44 and 35% of during 1979–2012 are characterized by frequent and intense the LWNT episodes for LIYs and HIYs, respectively. The episodes of positive surface LWNT anomalies during spring, largest differences in frequencies are evident for CP4 and here dene fi d as March 1 to melt onset. In contrast, the spring CP9 (Fig. 9b). in years with anomalously high September SIC are char- Another characteristic feature of the spring circulation acterized by fewer and shorter positive LWNT episodes patterns in LIYs is the total duration of all LWNT episodes. (Table 1). Further, the timing of melt onset occurs signifi- Figure 9 shows that CP1, CP5, CP8, and CP10 have episodes cantly earlier for most of the LIYs (Table 2). This leads to of positive LWNT anomalies persisting longer in LIYs than a decrease of the surface albedo early in the season hereby for HIYs, with differences becoming more significant for the enhancing the ice-albedo feedback, leading to an increased longer-lasting episodes for all of these patterns except CP10 absorption of shortwave radiation throughout the rest of 1 3 Summers with low Arctic sea ice linked to persistence of spring atmospheric circulation patterns Fig. 9 Relative frequency of occurrence of the circulation patterns associated with LWNT episodes during LIYs and HIYs, the difference in the relative frequency between the two data sets as well as the persistence, as presented in Fig. 8b. Top: Relative frequency of occur- rence during LWNT episodes of LIYs (black) and HIYs (red) as function of month. The rela- tive frequency is defined as the absolute frequency divided by the total number of days com- prising LWNT episodes in LIYs and HIYs, respectively. Middle: Differences in the relative fre- quencies. Bottom: Persistence for LWNT episodes in LIYs (black) and HIYs (red) 1 3 M.-L. Kapsch et al. Fig. 10 Arctic Oscillation (AO) index mapped onto the circulation regimes of the master SOM. Left: AO index for (a) all spring days between 1979 and 2012, and LWNT episodes during springs of years with (b) low and (c) high September sea–ice concentra- tion (LIYs/HIYs). The AO index was downloaded from NOAA Center for Weather and Climate Prediction, Climate Prediction Center (http://www. cpc.ncep.noaa.gov/produ cts/ preci p/CWlin k/daily _ao_index /ao.shtml ) the melt season and hence to an accelerated ice melt (e.g., September sea–ice variability (Kapsch et al. 2016). During Kapsch et al. 2016). LIYs these patterns occur relatively early in spring of LIYs, To investigate the atmospheric circulation patterns associ- as compared to HIYs, and are rather persistent (Fig. 8), sup- ated with the springtime LWNT episodes, a SOM algorithm porting an early preconditioning and melt onset of sea ice. was applied to all spring days from 1979 to 2012. Two dis- ConLE anomalies are negative only for CP8, indicating that tinct atmospheric circulation patterns prevail: one similar other processes contribute to the positive LWNT anoma- to the Arctic Dipole anomaly, the other associated with the lies under those circulation conditions. One such process positive phase of the Arctic Oscillation. The AD-like pat- may be an eastward propagation of cloud anomalies from terns are characterized by low geopotential heights along the Nansen Basin and parts of the Barents and Kara Seas. the Russian coast and high geopotential heights along the Liu et al. (2007) argued that cloud anomalies propagate Canadian coast. The AO+ patterns are associated with low along the cyclone tracks that traverse from the North Atlan- geopotential heights over the majority of the Arctic, favor- tic along the Russian coast towards the East Siberian Sea. ing transport from Siberia or the Kara Sea into the study Specifically, CP9 and CP10 are characterized by cyclonic region. CP1, CP3 and CP4 are the most frequent of the AD flow, supporting the notion of enhanced transport across the patterns, occurring over a third of the time in spring (15, western boundary of the investigation area (Fig. 5). Note that 10 and 11% of the days, respectively; Fig. 5). The most fre- mechanical wind forcing associated with sea–ice transport quent AO+ patterns are CP9, CP10 and CP12, prevailing might also play a role; however, here we find no significant more than another third of the spring days (10, 11 and 16%, anomaly in the wind velocities for this specific AO+ pattern respectively). During LWNT episodes in LIY springs, the (not shown). AO+ patterns occur more frequently and AD patterns less The AD patterns favor the advection of warm and moist frequently, compared to HIY-spring LWNT episodes. air from the Pacific and adjacent seas into the investiga- During the LWNT episodes in spring of LIYs, the tion area during LIY LWNT episodes. Another mechanism AO+ patterns are associated with advection from Siberia associated with AD patterns that can contribute to nega- that that for most patterns favors anomalous transport of tive September SICs over the investigation area is asso- moisture, leading to ConLE being positive; ConDE anoma- ciated with sea–ice export and advection of heat within lies are negative, likely due to low temperatures over Siberia the ocean. Rigor et al. (2002) found that the AD patterns until June (not shown). The positive anomalies of ConLE are transport sea ice from the Siberian Arctic towards Canada, associated with more water vapor and clouds in the study which leads to ice divergence in the central Arctic and domain, leading to positive LWD anomalies. Increased LWD therefore over the investigation area. Wang et al. (2009) during spring is the primary cause for melt onset (Persson and Maslanik et al. (2007) showed that AD favors sea–ice 2012; Mortin et al. 2016) and is likely an important factor for export from the Atlantic side of the Arctic, leading to an 1 3 Summers with low Arctic sea ice linked to persistence of spring atmospheric circulation patterns Acknowledgements Open access funding provided by Max Planck increased ocean heat flux into the Arctic Ocean through Society. The authors are grateful to the anonymous reviewers for their the Bering Strait. The LWNT episodes from 23 April to constructive comments and valuable suggestions, which helped to sig- 19 May 1990 were associated exclusively with the patterns nificantly improve the manuscript. The ERA-Interim data is provided CP3, CP4, CP8 and CP10 and exhibit a SIC reduction by the ECMWF and was downloaded from the MARS data reposi- tory. The computations were performed on resources provided by the after AD patterns had prevailed (Fig. 3). As all of these Swedish National Infrastructure for Computing (SNIC) at Linköping patterns are associated with positive anomalies of surface University. MLK was funded by the Swedish research council Formas wind speeds over the investigation area (not shown), these through the ADSIMNOR project. JAF and NS are supported by the might be additional factors that contribute to the negative NASA Grant NNX14AH896. MLK is also grateful for support from the Swedish e-Science Research Center (SeRC). ice anomalies. Note that LWNT episodes occur not only in LIYs or Open Access This article is distributed under the terms of the Crea- HIYs. Throughout all the springs considered, 272 LWNT tive Commons Attribution 4.0 International License (http://creat iveco episodes occurred, spanning a total of 1156 days (Sect. 3.2; mmons.or g/licenses/b y/4.0/), which permits unrestricted use, distribu- Table 1). Significant lagged correlations between LWNT tion, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the episodes and anomalies of ConLE are evident in all years Creative Commons license, and indicate if changes were made. from 1979 to 2012, except 2010 (not shown). This indi- cates that the correlations between ConLE and episodes of enhanced LWNT are not unique to LIYs but, as would be expected, that enhanced moisture convergence always References triggers LWNT anomalies. During spring of LIYs, how- ever, LWNT episodes last considerably longer than aver- Burt MA, Randall DA, Branson MD (2016) Dark warming. J Clim age (Table 1), suggesting that atmospheric circulation 29:705–719. https ://doi.org/10.1175/JCLI-D-15-0147.1 patterns associated with enhanced moisture transport are Comiso JC, Parkinson CL, Gersten R, Stock L (2008) Accelerated decline in the Arctic sea ice cover. Geophys Res Lett 35:L01703. more persistent in LIY springs. https ://doi.org/10.1029/2007G L0319 72 Two questions were posed at the beginning of this Cox CJ, Uttal T, Long CN, Shupe MD, Stone RS, Starkweather S paper. Regarding the first question—do similar atmos- (2016) The role of springtime arctic clouds in determining autumn pheric circulation patterns prevail in spring of all LIYs— sea ice extent. J Clim 29(18):6581–6596. https://doi.or g/10.1175/ JCLI-D-16-0136.1 the answer is inconclusive. We do find similarities in the Dee DP, Uppala S (2009) Variational bias correction of satellite radi- atmospheric circulation in spring of LIYs, but several ance data in the ERA-Interim reanalysis. QJR Meteorol Soc flow patterns emerge, and some important flow patterns 135:1830–1841. https ://doi.org/10.1002/qj.493 emerge in both LIYs and HIYs. However, we find a pos- Dee DP et al (2011) The ERA-Interim reanalysis: configuration and performance of the data assimilation system. QJR Meteorol Soc sible relationship between circulation patterns in LIYs 137:553–597. https ://doi.org/10.1002/qj.828 and the positive phase of the Arctic Oscillation, which Devasthale A, Sedlar J, Koenigk T, Fetzer EJ (2013) The thermody- might favor episodes of enhanced LWNT due to increased namic state of the Arctic atmosphere observed by AIRS: com- energy convergence over the investigation area. Note, parisons during the record minimum sea ice extents of 2007 and 2012. Atmos Chem Phys 13:7441–7450. https://doi.or g/10.5194/ however, that some patterns, such as CP1, CP6 or CP10, acp-13-7441-2013 are associated with energy divergence throughout LWNT Ding Q et al (2017) Influence of high-latitude atmospheric circula- episodes in LIYs and HIYs (not shown), which leaves a tion changes on summertime Arctic sea ice. Nat Clim Change substantial gray zone in which other processes, such as 7:289–295. https ://doi.org/10.1038/nclim ate32 41 Döscher R, Vihma T, Maksimovich E (2014) Recent advances in sea–ice dynamics and mechanical wind forcing or oce- understanding the arctic climate system state and change from anic processes, may or may not play important roles for a a sea ice perspective: a review. Atmos Chem Phys Discuss given year. As to the second question—whether there are 14:10929–10999. https ://doi.org/10.5194/acpd-14-10929 -2014 systematic differences in the characteristics of the atmos- Eastman R, Warren SG (2010) Interannual variations of Arctic cloud types in relation to sea ice. J Clim 23:4216–4232. https ://doi. pheric transport events during spring of LIYs and HIYs— org/10.1175/2010j cli34 92.1 the answer is definitely yes, particularly with respect to Graversen RG, Källén E, Tjernström M, Körnich H (2007) Atmos- strength and duration of the LWNT events. Both intensity pheric mass-transport inconsistencies in the ERA-40 reanalysis. and persistence of thermodynamic forcing from the atmos- Q J R Meteorol Soc 133:673–680. https ://doi.org/10.1002/qj.35 Graversen RG, Mauritsen T, Drijfhout S, Tjernström M, Mårtensson phere in spring are larger for LIYs than for HIYs (Table 1). S (2011) Warm winds from the Pacific caused extensive Arctic The significant difference in the persistence and timing of sea–ice melt in summer 2007. Clim Dyn 36:2103–2112. https :// the individual flow patterns in spring prior to September doi.org/10.1007/s0038 2-010-0809-z LIYs and HIYs suggests that the distinct circulation pat- Hewitson BC, Crane RG (2002) Self-organizing maps: applications to synoptic climatology. Clim Res 22:13–26. https://doi.or g/10.3354/ terns (CP9, CP10 and CP12) tend to occur earlier in spring cr022 013 and thus have longer time to act on the surface during LIYs (Figs. 8, 9; Table 1). 1 3 M.-L. Kapsch et al. IPCC (2013), Climate change 2013: the physical science basis. Contri- Overland JE, Wang M (2005) The third Arctic climate pattern: 1930s bution of Working Group I to the Fifth Assessment Report of the and early 2000s. Geophys Res Lett 32:L23808. https ://doi. Intergovernmental Panel on Climate Change. In: Stocker TF, Qin org/10.1029/2005G L0242 54 D, Plattner G-K, Tignor M, Allen SK, Boschung J, Nauels A, Xia Overland JE, Wang M, Salo S (2008) The recent Arctic warm Y, Bex V, Midgley PM (eds) Cambridge University Press, Cam- period. Tellus Ser A 60:589–597. https ://doi.or g/10.111 bridge, pp 1535. https ://doi.org/10.1017/CBO97 81107 41532 4 1/j.1600-0870-2008 Jakobson E, Vihma T, Palo T, Jakobson L, Keernik H, Jaagus J Persson POG (2012) Onset and end of the summer melt season over (2012) Validation of atmospheric reanalyses over the cen- sea ice: thermal structure and surface energy perspective from tral Arctic Ocean. Geophys Res Lett 39:L10802. https ://doi. SHEBA. Clim Dyn 39(6):1349–1371. https ://doi.org/10.1007/ org/10.1029/2012G L0515 91s0038 2-011-1196-9 Kapsch M-L, Graversen RG, Tjernström M (2013) Springtime atmos- Reusch DB, Alley RB, Hewitson BC (2005) relative performance of pheric energy transport and the control of Arctic summer sea–ice self-organizing maps and principal component analysis in pat- extent. Nat Clim Change 3:744–748. https ://doi.or g/10.1038/ tern extraction from synthetic climatological data. Polar Geogr nclim ate18 84 29:188–212. https ://doi.org/10.1080/78961 0199 Kapsch M-L, Graversen RG, Tjernström M, Bintanja R (2016) The Rigor IG, Wallace JM, Colony RL (2002) Response of sea ice effect of downwelling longwave and shortwave radiation on Arctic to the Arctic oscillation. J Clim 15:2648–2663. https://doi. summer sea ice. J Clim 29:1143–1159. https ://doi.org/10.1175/ org/10.1175/1520-0442(2002)015<2648:ROSITT>2.0.CO;2 JCLI-D-15-0238.1 Serreze MC, Stroeve J (2015) Arctic sea ice trends, variability Kauker F, Kaminski T, Karcher M, Giering R, Gerdes R, Voßbeck and implications for seasonal forecasting. Philis Trans R Soc M (2009) Adjoint analysis of the 2007 all time Arctic sea– A373:20140159. https ://doi.org/10.1098/rsta.2014.0159 ice minimum. Geophys Res Lett 36:L03707, https ://doi. Sheridan SC, Lee CC (2011) The self-organizing map in synoptic cli- org/10.1029/2008G L0363 23 matological research. Prog Phys Geogr 35(1):109–119. https :// Kay JE, Holland MM, Jahn A (2011) Inter-annual to multi-decadal doi.org/10.1177/03091 33310 39758 2 Arctic sea ice extent trends in a warming world. Geophys Res Lett Skific N, Francis JA (2012), Self-organizing maps: a powerful tool for 38:L15708. https ://doi.org/10.1029/2011G L0480 08 the atmospheric sciences, applications of self-organizing maps. Kohonen T (1982) Self-organized formation of topologically correct In: Johnsson M (ed) ISBN: 978-953-51-0862-7. InTech. https :// feature maps. Biol Cybern 43(1):59–69. https ://doi.org/10.1007/ doi.org/10.5772/54299 bf003 37288 Skific N, Francis JA, Cassano JJ (2009) Attribution of projected Kohonen T (2001) Self-Organizing Maps, Third Edition. Springer, changes in atmospheric moisture transport in the Arctic: a self- Berlin organizing map perspective. J Clim 22:4135–4153 Lee HJ, Kwon MvO, Yeh S, Kwon Y, Park W, Park J, Kim YH, Stroeve JC, Markus T, Boisvert L, Miller J, Barrett A (2014) Changes Alexander MA (2002) Impact of poleward moisture transport in Arctic melt season and implications for sea ice loss. Geophys from the North Pacific on the acceleration of sea ice loss in the Res Lett 41:1216–1225. https ://doi.org/10.1002/2013G L0589 51 Arctic since. J Clim 30:17:6757–6769. https ://doi.org/10.1175/ Thompson DWJ, Wallace JM (1998) The Arctic oscillation signature JCLI-D-16-0461.1 in the wintertime geopotential height and temperature fields. Geo- Lindsay R, Wensnahan M, Schweiger A, Zhang J (2014) Evaluation phys Res Lett 25(9):1297–1300. https://doi.or g/10.1029/98GL0 of seven different atmospheric reanalysis products in the Arctic. J 0950 Clim 27(7):2588–2606. https://doi.or g/10.1175/JCLI-D-13-00014 Trenberth, K. E. (1991), Climate diagnostics from global analysis: Liu Y, Key JR, Francis JA, Wang X (2007) Possible causes of decreas- Conservation of mass in ECMWF analysis. J Clim., 4, 707–721. ing cloud cover in the Arctic winter, 1982–2000. Geophys Res https://doi.org/10.1175/1520-0442(1991)004<0707:CDFGAC> Lett 34:L14705. https ://doi.org/10.1029/2007G L0300 42 2.0.CO;2 Maslanik JA, Fowler C, Stroeve J, Drobot S, Zwally J, Yi D, Emery W Wang J, Zhang J, Watanabe E, Ikeda M, Mizobata K, Walsh JE, Bai X, (2007) A younger, thinner Arctic ice cover: Increased potential for Wu B (2009) Is the Dipole Anomaly a major driver to record lows rapid, extensive sea–ice loss. Geophys Res Lett 34:L24501. https in Arctic summer sea ice extent? Geophys Res Lett 36:L05706. ://doi.org/10.1029/2007G L0320 43https ://doi.org/10.1029/2008G L0367 06 Mortin J, Howell SEL, Wang L, Derksen C, Svensson G, Graversen Wang C, Granskog MA, Hudson SR, Gerland S, Pavlov AK, Per- RG (2014) Extending the QuikSCAT record of seasonal melt– ovich DK, Nicolaus M (2016), Atmospheric conditions in the freeze transitions over Arctic sea ice using ASCAT. Remote Sens central Arctic Ocean through the melt seasons of 2012 and Environ 141:214–230. https ://doi.org/10.1016/j.rse.2013.11.004 2013: impact on surface conditions and solar energy deposition Mortin J, Svensson G, Graversen RG, Kapsch M-L, Stroeve J, Boisvert into the ice-ocean system, J Geophys Res Atmos. https ://doi. LN (2016) Melt onset over Arctic sea ice controlled by atmos-org/10.1002/2015J D0237 12 pheric moisture transport. Geophys Res Lett 43:6636–6642. https Zhang J, Lindsay R, Steele M, Schweiger A (2008) What drove the dra- ://doi.org/10.1002/2016G L0693 30 matic retreat of the arctic sea ice during summer 2007? Geophys Ogi M, Wallace JM (2012) The role of summer surface wind anomalies Res Lett 35:1944–8007. https ://doi.org/10.1029/2008G L0340 05 in the summer Arctic sea ice extent in 2010 and 2011. Geophys Res Lett 39:1944–8007. https ://doi.org/10.1029/2012G L0513 30 1 3
Climate Dynamics – Springer Journals
Published: May 30, 2018
It’s your single place to instantly
discover and read the research
that matters to you.
Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.
All for just $49/month
Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly
Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.
All the latest content is available, no embargo periods.
“Whoa! It’s like Spotify but for academic articles.”@Phil_Robichaud