Marine assemblages respond rapidly to winter climate variability

Even species within the same assemblage have varied responses to climate change, and there is a poor understanding for why some taxa are more sensitive to climate than others. In addition, multiple mechanisms can drive species' responses, and responses may be specific to certain life stages or times of year. To test how marine species respond to climate variability, we analyzed 73 diverse taxa off the southeast US coast in 26 years of scientific trawl survey data and determined how changes in distribution and biomass relate to temperature. We found that winter temperatures were particularly useful for explaining interannual variation in species' distribution and biomass, although the direction and magnitude of the response varied among species from strongly negative, to little response, to strongly positive. Across species, the response to winter temperature varied greatly, with much of this variation being explained by thermal preference. A separate analysis of annual commercial fishery landings revealed that winter temperatures may also impact several important fisheries in the southeast United States. Based on the life stages of the species surveyed, winter temperature appears to act through overwinter mortality of juveniles or as a cue for migration timing. We predict that this assemblage will be responsive to projected increases in temperature and that winter temperature may be broadly important for species relationships with climate on a global scale.


Marine assemblages respond rapidly to winter climate variability
Even species within the same assemblage have varied responses to climate change, and there is a poor understanding for why some taxa are more sensitive to climate than others. Multiple mechanisms can drive species' responses, and responses may be specific to certain life stages or times of year. To test how marine species respond to climate variability, we analyzed 73 diverse taxa off the southeast U.S. coast in 26 years of scientific trawl survey data and determined how changes in distribution and biomass relate to temperature. We found that interannual variation in distribution and abundance for many taxa was explained by temperature during the previous winter. Further, the direction and magnitude of response of individual species to winter climate was predictable from their thermal preference. Both warm-and cold-water species were sensitive to winter conditions, though often in opposing directions. A separate analysis of annual commercial fisheries landings revealed that winter temperatures may also impact several important fisheries in the southeast U.S. Based on the life stages of the species surveyed, winter temperature often appears to act through overwinter mortality of juveniles or as a cue for migration timing. We predict that this assemblage will be responsive to projected increases in temperature, and that winter temperature may be broadly important for species relationships with climate on a global scale. determined how changes in distribution and biomass relate to temperature. We found that 24 interannual variation in distribution and abundance for many taxa was explained by temperature 25 during the previous winter. Further, the direction and magnitude of response of individual 26 species to winter climate was predictable from their thermal preference. Both warm-and cold-27 water species were sensitive to winter conditions, though often in opposing directions. A 28 separate analysis of annual commercial fisheries landings revealed that winter temperatures may 29 also impact several important fisheries in the southeast U.S. Based on the life stages of the 30 species surveyed, winter temperature often appears to act through overwinter mortality of 31 juveniles or as a cue for migration timing. We predict that this assemblage will be responsive to 32

Introduction 36
The global redistribution of species is a prominent impact of climate change (Parmesan  Therefore, we would expect that the thermal preference of a species might be a useful predictor 61 for how it will respond to climate variability in a region (Simpson et al., 2011

Temperature data 99
We calculated annual winter and spring temperature anomalies using both atmospheric 100 and water temperatures at five locations in the southeast U.S. (Fig. 1). However, we found that 101 temperature anomalies from the North Carolina atmospheric data effectively represented 102 interannual temperature variation throughout the southeast U.S., and so we only used the North 103 Carolina atmospheric data for analysis with biological data ( Fig. 2; Appendix S1; Table S2). The 104 North Carolina atmospheric data also had better data coverage than the other sources.  Table S1). Satellite derived sea surface temperature data were examined as 109 an alternative data source (Reynolds et al., 2007). However, these data had an unrealistic decline 110 in winter temperatures over the past 25 years that was not present with buoy data, so we did not 111 use satellite data. 112 Annual anomalies were calculated using mean daily temperature values, and days with 113 which begins two weeks before the average initiation of the spring trawl survey and ends when 125 90% of the survey is typically complete (Fig. S1). Two annual spring anomalies were missing for 126 the North Carolina atmospheric data and these values were estimated from regressions with 127 North Carolina water temperatures (Table S1). 128 North Carolina and Cape Canaveral, Florida (Fig. 1). Only a single depth stratum has been 133 sampled consistently, which occurs between the 4 and 10 m isobaths and is split into 24 strata 134 that follow the coastline. The deeper strata (10 -19 m depth) lacked spatial and temporal 135 coverage and so were not included in our analysis. Only taxa that were sorted to the species level 136 were included in analyses. Detailed information regarding the SEAMAP-SA survey is available 137 online (http://www.seamap.org/CoastalSurvey.html). 138 All analyses of survey data (distribution and biomass) were based on annual biomass 139 estimates within strata, and each season was analyzed independently. We calculated biomass for 140 each species in each stratum by dividing total captured biomass by area sampled per stratum, and  Only species that were collected at least an average of 10 times per year within seasons were 146 considered for analysis (Table S7). 147 Annual mean center of latitudinal biomass (centroid) was estimated for each species, 148 within each season, by calculating the stratum-biomass weighted average latitude. Our analysis 149 of distribution was restricted to latitude, because the strata for this survey are predominantly 150 arranged in a north-south direction. Annual biomass for each species was calculated within 151 were more likely to be associated with significant long-term trends in total biomass. 165

Response of assemblages to interannual temperature variation 166
We determined if winter and spring temperature anomalies explained annual variation in 167 centroid location and total biomass for both the spring and summer assemblages. The fall 168 assemblage was not included with these analyses because catches in this season are comprised 169 largely of young-of-the-year for many species (ASFMC, 2000). Also, we did not relate summer 170 temperatures to species data because regional temperature anomalies within the southeast U.S. 171 did not co-vary (Table S2) and because atmospheric and water temperatures were not as strongly 172 coupled (Table S1)  were used as the annual distribution or biomass anomalies. We then used linear regression to test 188 whether the annual anomalies of distribution and biomass were related to annual temperature 189 anomalies in winter ( Fig. 3b and e) and spring ( Fig. 3c and f). In addition, we modeled annual centroid and biomass anomalies with linear mixed effects 213 models where the species-specific relationship with winter or spring temperatures was included 214 as a random effect. We examined four response variablesdistribution and biomass responses in 215 both the spring and summer assemblagesand used AIC c to determine if winter or spring 216 temperatures better explained variation in species distribution and biomass. Further, we included 217 an interaction term (temperature × thermal preference) to determine if thermal preference 218 influenced species level responses to interannual temperature variability. Intercept terms did not 219 improve model fits and so were excluded; response variables were anomalies and so it was not 220 expected that intercept terms would explain any variation. Each group of models was weighed 221 against a null model that assumed data were best explained by a mean value. All modeling was 222 conducted with R version 3.3.1 (R Core Team, 2016) using the mgcv package for GAMs (Wood, 223 2011) and lme4 package for mixed effects models (Bates et al., 2014). 224

Commercial catch 225
To compare our results from the scientific surveys with an independent data set, we also 226 examined trends in commercial fisheries landings. We extracted commercial landings data for 23 227 species in the southeast U.S., ranging from North Carolina to Georgia, from the National Marine 228 was an obvious and rapid change in the fishery (Table S4) Table 1). 282 The influence of winter temperatures carried over into the summer assemblage, but this 283 effect was only observed with biomass and the effect was weaker. Across 64 taxa, cold-water 284 species generally had lower summer biomasses following warm winters, while warm-water taxa 285 showed no general response ( Fig. 5; slope(se) = 0.016(0.006), P = 0.01, r 2 = 0.08). The mixed-286 effects model supported the importance of thermal preference interacting with winter 287 temperatures (∆AIC = 2.6; Table 1). The distribution responses of species to winter temperatures 288 was not related to thermal preference within the summer assemblage (P = 0.55). 289 Generally, responses of species to spring temperature anomalies resulted in fewer 290 significant relationships compared to relationships with winter temperatures (Table S6). 291 Although the relationship of species biomass responses to spring temperatures was related to 292  (Table 1). Overall, winter temperatures appear to be more important than 314 spring temperatures for predicting the distribution patterns of this assemblage. 315

Effects of interannual temperature variation on commercial landings 316
Interannual variation in 7 of 23 (30%) southeast U.S. fisheries were significantly related 317 to winter temperatures. These fisheries included some economically valuable species such as 318 white and pink shrimp, which were positively related to winter temperatures, and Atlantic 319 croaker, summer flounder, and bluefish, which were negatively related to winter temperatures. temporal scales. We found that the effects of winter temperature were apparent in both the spring 336 and summer assemblages, which suggests that winter temperatures can broadly impact annual 337 recruitment and migration patterns. We note that summer temperatures were not related to winter 338 or spring temperatures (Appendix S1), so the response of the summer assemblage to 339 temperatures during the previous winter was not confounded by years that may have been 340 generally warm or cold in all seasons. 341 Despite finding substantial variation across taxa in how their distribution and biomass 342 responded to temperature, we found consistent similarities among species based on thermal 343 preference. Following mild winters, warm-water species were generally more abundant in the 344 northern region of the Southeast U.S. Shelf, which resulted in a north-shifted center of biomass 345 for these taxa. Conversely, cold-water species were less abundant following mild winters, 346 although they showed no general distribution response. The lack of distribution responses among 347 cold-water species may occur because many of these species are not as widespread throughout 348 the Southeast U.S. Shelf as warm-water taxa, which would have limited our ability to detect 349 distribution responses among them and might explain why biomass responses were more 350 strongly linked to temperature preference. 351 During the summer, thermal preference explained only a small amount of variation in 352 assemblage-scale responses to winter temperatures. This poor relationship was primarily due to 353 highly variable responses among warm-water species, which exhibited no consistent trend. The 354 biomass responses of cold-water species within the summer assemblage generally had negative 355 relationships with warmer winter temperatures, which were consistent to what was observed in 356 spring. There may be several explanations for why the summer assemblage responses were not 357 as strongly related to thermal preference as was the spring assemblage. First, the species included 358

Mechanisms behind species responses to winter temperatures 382
The mechanisms behind the observed responses to winter temperature variability depend 383 in part on the species and life stage captured by the SEAMAP trawl survey. Many of the taxa 384 included in our analysis are comprised mostly of age-1 individuals, particularly for the teleost 385 fishes and penaeid shrimp (ASMFC, 2000). Therefore, for many non-migratory species, 386 variability in distribution and biomass during spring and summer is likely to reflect spatial overwinter on the Southeast U.S. Shelf. Following winter, some of the age-1 fish remain in the 407 region, while others migrate north (Morley et al., 2013). More severe winters incur an energetic 408 deficit on age-1 bluefish (Morley et al., 2007). Therefore, the increased summer biomass on the 409 Southeast U.S. Shelf following colder winters may result from a greater portion of age-1 fish that 410 remain in the region due to reduced energy reserves. These age-1 fish typically comprise the 411 majority of the commercial catch (SAW assessment report, 2015), which may explain why 412 fishery landings were also significantly related to winter temperatures for this species. Our results provide insight into the mechanisms that may be driving changes in species 423 distribution and regional biomass, which is important because mechanisms behind changes in predicted by theory that associates geographic range of species with thermal tolerance (Sunday et 428 al., 2012). Given the sensitivity of this assemblage at an annual time scale, we expect that it will 429 be highly responsive to future long-term increases in temperature. 430 On the Southeast U.S. Shelf, winter temperatures may be of particular importance as a 431 biogeographic barrier for many tropical species. While this region appears to be suitable thermal 432 habitat for many tropical species during summer, sporadic cold winter temperatures may prevent Based on our findings, we expect that increased winter temperatures will affect the 441 Southeast U.S. Shelf in a number of important ways. First, we expect that species with more 442 southern affinities will experience increased juvenile survival during winter, which will lead to 443 greater abundance of species like white shrimp and star drum north of Florida. This would 444 impact overall community structure, as southern species will likely compete for the same niche 445 space as taxa that are currently common. For example, on the northern Gulf of Mexico coast, 446 many of the more northern taxa were lost entirely after ~3 °C of warming led to the introduction 447 of many tropical species (Fodrie et al., 2010). Second, the depth distribution of many warm-448 water species on the continental shelf may expand towards the coast, as winter temperature

Appendix S1 Spatial and temporal variation of SAB temperatures
Three sets of regression analyses were conducted to justify using North Carolina (NC) atmospheric temperatures for analysis of survey data, and to examine climate variability on the Southeast U.S. Shelf. First, the use of atmospheric data to represent coastal water temperatures was validated by regressions of annual water versus atmospheric temperature anomalies for each region and season, with the exception of South Carolina, which lacked water temperature data. All relationships between atmospheric and water temperature anomalies were highly significant (Table S1). Of the three seasons examined, correlations between atmospheric and water temperatures were highest in winter (Fig. 2), intermediate in spring (Fig. 2), and lowest in summer (Fig. S2). Further, Gray's Reef, the only offshore weather station used, exhibited the highest degree of air and sea temperature correlation, while Florida had the greatest amount of variation in regressions between air and sea temperatures, particularly during summer. Overall, atmospheric temperature data appear to be an effective method for estimating nearshore temperature conditions for marine taxa.
Second, we determined how NC atmospheric temperatures represent the entire nearshore Southeast U.S. Shelf, by regressing each regional temperature anomaly time series against the NC values, for all three seasons. The winter and spring temperature anomalies based on NC atmospheric data were highly correlated with both atmospheric and water temperatures in other regions (Table S2). Correlations were stronger with other atmospheric temperature anomalies compared to water temperatures. Further, correlations were generally stronger in winter. During summer, the only temperature time series that was correlated with NC was the atmospheric anomalies from South Carolina. These results suggest that the relative severity or mildness of the winter and early-spring season is consistent throughout the coastal SAB. Therefore, temperature data from Cape Lookout, NC accurately represent winter and spring conditions in the entire SAB. However, during summer, temperature conditions in the northern versus southern regions are decoupled.
The third analysis determined if the relative temperatures (e.g., a warm or cold year) in the winter season are indicative of conditions during the following spring and summer by regressing temperature anomaly values between seasons for each region. There was generally a weak relationship between winter and spring temperature anomalies (Table S3). However, this relationship was significant for three of the seven temperature time series, including NC atmospheric anomalies (Fig. S3). There was no significant correlations between winter and summer anomalies (all p > 0.27) or spring and summer anomalies (all p > 0.31) in any region.
In the northern region of the SAB, annual temperature anomalies were more variable in winter than in summer (F = 6.05, P < 0.001). Further, for the NC atmospheric data, the range of annual winter temperature anomalies was 4.5 °C, compared to 1.8 °C for annual summer temperature anomalies.

Long-term trends in species distribution
During spring, 16 of the 61 taxa examined for distribution shifts had significant trends in centroid location over the 26 year time period of the SEAMAP survey; 8 shifted southwards and 8 to the north. During summer, 14 of 64 taxa had significant trends; 6 shifted southwards and 8 northwards. During fall, 12 of 70 taxa had significant trends; 6 shifted southwards and 6 northwards. Species that had similar and significant trends in mean latitude for multiple seasons exhibited the greatest evidence for major distribution changes. Seven taxa had significant-unidirectional shifts in two or three seasons. Of these seven species, Selene setapinnis, Larimus fasciatus, Chilomycterus schoepfii, Gymnura micrura and Cynoscion nothus shifted northward while Limulus polyphemus and Ariopsis felis shifted to the south.          Positive values indicate species with higher spring (a) or summer (b) biomass following warm springs. Solid points represent taxa with significant (P < 0.05) relationships between biomass and spring temperatures at an annual scale. Dashed lines are 95% confidence intervals around the regression (a: P < 0.001, r 2 = 0.28; b: P = 0.003, r 2 = 0.12). Biomass responses of 52 species within the summer assemblage to winter temperature anomalies, related to temperature preference. Positive values indicate species with higher summer biomass following warm winters. Solid points represent species with significant (P < 0.05) relationships between biomass and winter temperatures at an annual scale. Dashed lines are 95% confidence intervals around the regression (P < 0.001, r 2 = 0.22). The species included in this regression represents a subset of the summer assemblage, and includes only species that are also present in the spring assemblage.