Projected shifts in loggerhead sea turtle thermal habitat in the Northwest Atlantic Ocean due to climate change

It is well established that sea turtles are vulnerable to atmospheric and oceanographic shifts associated with climate change. However, few studies have formally projected how their seasonal marine habitat may shift in response to warming ocean temperatures. Here we used a high-resolution global climate model and a large satellite tagging dataset to project changes in the future distribution of suitable thermal habitat for loggerheads along the northeastern continental shelf of the United States. Between 2009 and 2018, we deployed 196 satellite tags on loggerheads within the Middle Atlantic Bight (MAB) of the Northwest Atlantic continental shelf region, a seasonal foraging area. Tag location data combined with depth and remotely sensed sea surface temperature (SST) were used to characterize the species’ current thermal range in the MAB. The best-fitting model indicated that the habitat envelope for tagged loggerheads consisted of SST ranging from 11.0° to 29.7 °C and depths between 0 and 105.0 m. The calculated core bathythermal range consisted of SSTs between 15.0° and 28.0 °C and depths between 8.0 and 92.0 m, with the highest probability of presence occurred in regions with SST between 17.7° and 25.3 °C and at depths between 26.1 and 74.2 m. This model was then forced by a high-resolution global climate model under a doubling of atmospheric CO2 to project loggerhead probability of presence over the next 80 years. Our results suggest that loggerhead thermal habitat and seasonal duration will likely increase in northern regions of the NW Atlantic shelf. This change in spatiotemporal range for sea turtles in a region of high anthropogenic use may prompt adjustments to the localized protected species conservation measures.

Warming ocean temperatures due to climate change are already having a measurable impact on ecological processes 1 . An emerging body of research has documented distribution shifts 2 , phenological changes to seasonal migration and reproduction 3 , and trophic mismatch 4 in a wide variety of marine taxa. All of these changes increase the difficulty of managing commercially valuable marine species 5 and protecting endangered and threatened animals 6 .
Understanding species distribution and habitat preferences are becoming fundamental components to developing effective resource management and conservation strategies 7,8 . Fisheries bycatch is one of the most serious threats to sea turtles around the world 9,10 . Attempts to mitigate bycatch levels are often based on an understanding of when and where a species occurs over time and how interactions occur with the fishing gear 11,12 . With the advent of time/area closures in fisheries management, more research is being conducted to understand the spatio-temporal nature of by-catch species 11 . In the Pacific, fisheries interactions with loggerhead sea turtles (Caretta caretta) have resulted in temporary area closures, and vessels must comply with stringent regulations to prevent the incidental capture of this species 13 . While these types of regulations have resulted in reduced Methods Loggerhead satellite tagging. All loggerheads were tagged between 2009 and 2018. The majority of turtles (n = 190) were tagged within the MAB between May and September. Two turtles were tagged in eastern GB and one was tagged off-shelf near southern MAB at the edge of the Gulf Stream, a fast moving current that flows along the US eastern seaboard pulling warm water from the Gulf of Mexico northward and eastward 34 . Three additional turtles were tagged off of the North Carolina coast in February. Curved carapace length, from anterior notch to posterior tip, was measured for each captured turtle. Most (n = 186) turtles were equipped with a satellite relay data logger manufactured by the Sea Mammal Research Unit at the University of St Andrews; and ten turtles were equipped with a Wildlife Computers SPOT tag. See Patel et al. 35,36 , Winton et al. 29 and Crowe et al. 37 for full details on capture and handling protocols, satellite tag parameterizations, and more details from the satellite tag data.
All fieldwork was approved by the US National Marine Fisheries Service (NMFS) Atlantic Institutional Animal Care and Use Committee and the US Endangered Species Act (ESA) Section 10(a) ( www.nature.com/scientificreports/ movement model fitted to location data using the software 'Template Model Builder' 38 in R 39 . Daily position estimates were interpolated from each movement model's output 29,40,41 to correct for the different transmission rates of each tag and to reduce autocorrelation in position estimates. Prior to fitting the movement model, all location coordinates were re-projected into the oblique Mercator center projection centered on 35.0°N, 75.0°W using R package 'rgdal' 42 and a speed filter with a max speed of 5 km/hr was applied to remove errant telemetered locations 43 . For analysis, loggerhead locations were aggregated over the NMFS Atlantic Marine Assessment Program for Protected Species (AMAPPS) spatial grid that has a 10-km resolution 29 . For model fitting, we used loggerhead location estimates from continental shelf waters (depths < 200 m) between 33.5°N and 41.6°N, which encompasses the MAB as delineated by NMFS statistical areas 44 and corresponds to the highest density area traversed by satellite-tagged loggerheads 29 . Although on occasion loggerheads were tracked in waters deeper than 200 m and north of 41.6°N latitude, those locations were removed from this study due to the low sample size and the higher incidences of fisheries interactions occurring on shelf waters 45 Locations were binned by month to match the resolution of climate projections 46 and aggregated over the 10-km resolution AMAPPS spatial grid. The AMAPPS grid was bounded by the coastline to constrain loggerhead space use to the ocean.
Characterizing the bathythermal range of loggerheads. Although we understand loggerheads are likely influenced by a large range of environmental variables, our goal was to investigate how the distribution of loggerheads may change in response to warming water temperatures associated with climate change. To model spatial variation in the occurrence of tagged loggerheads and identify the proportion of the observed variation related to water temperature, generalized linear models (GLMs) were applied to estimate the relationship www.nature.com/scientificreports/ between the probability of loggerhead presence, SST, and depth. We modeled the occurrence, y it, (0 = absent, 1 = present), of tagged turtles in grid cell i during time step t as the outcome of a Bernoulli random variable: where p it is the probability that a tagged turtle was present. We modeled p it as a function of SST and depth as: where the logit link function constrains p it from 0 to 1, β 0 is an intercept term; β 1 and β 2 represent a quadratic effect of SST (which allows for a non-linear relationship); β 3 and β 4 a quadratic effect of bottom depth. Depth data for daily loggerhead locations were obtained from the gridded ETOPO1 bathymetry data set 47 . For observed ocean temperature data (2009-2018), we used the Optimum Interpolation Sea Surface Temperature (OISST) product for the same time period of the turtle tracking. OISST is a combination of observations from different platforms (satellites, ships, buoys) and is produced at a 1/4° spatial resolution 48 . Daily OISST satellite composites were obtained from the NOAA CoastWatch Program (http:// coast watch. pfeg. noaa. gov/ erddap/ gridd ap/) using functions available in the R package 'rerrdap' 49 and averaged together to create monthly climatologies to match the output of the climate model projections. These composites were then up-sampled to align with the AMAPPS grid by simple averaging. All model variants were fit via maximum likelihood methods using the package 'Template Model Builder' 38 . All parameters were treated as fixed effects; the final gradient value for parameters and the hessian matrix were inspected for each model fit to confirm convergence. We used the Akaike information criterion (AIC) 50 and the percent deviance explained 51 for model selection. Given the small number of explanatory variables considered, we used a forward, step-wise selection approach to identify the most parsimonious combination of regression terms 52 . Individual terms were retained in the model if their inclusion resulted in a lower AIC and increased the proportion of the deviance explained relative to the best-fitting model from the previous step. To assess the fit of the selected model and identify potential model misspecification, we examined standard residual diagnostic plots using normalized, randomized residuals 53 . Visualizations of model results were produced using functions available in the R package 'tidyverse' 54 .

Forecasting the distribution of loggerheads under climate change.
To investigate how the distribution of loggerheads may shift under climate change, the selected model was fitted to long-term (80 year) projections of SST in the MAB, SNE, GB, and GOM, the most northern portion of loggerhead range within the western Atlantic Ocean 29 . Projections were based on a climate change scenario from the National Oceanic and Atmospheric Association's (NOAA's) high-resolution global climate model (CM2.6) as described and validated by Saba et al. 46 for the Northwest Atlantic. Unlike most global climate models that have a warm bias due to misrepresentation of the position of the Gulf Stream, CM2.6 resolves the Gulf Stream, regional ocean circulation, and bathymetry of the Northwest Atlantic shelf 46 much more realistically. Overall, CM2.6 outperformed all coarser models that were assessed 46 . Many studies that have projected marine species habitat shifts in the Northwest Atlantic have relied on this climate model 20,55-57 .
The SST output from CM2.6 represents a monthly deviation from a historical average derived from control simulations (deltas). The CM2.6 output was rasterized onto a 0.1° × 0.1° mesh and then synced to the AMAPPS grid. The SST deltas were then added to the mean monthly SST values for the observed time period. Along with depth, which we assumed remained constant, the projected monthly SST was used to project the probability of loggerhead presence from the MAB north to GOM within the continental shelf for 80 years conditioning on the fitted model. For visualization purposes, observed and projected data were grouped into seasons based on both general climate trends for the region and turtle habitat usage patterns 29,36,58 . The projected probability of presence was then averaged across years (10 and 20 year bins). January through March were grouped into winter, April through June to spring, July through September to summer; and October through December to fall.
Quantifying climate-related shifts in distribution. To better understand and visualize the predicted changes in loggerhead occupancy (presence/absence) under climate change, we developed a binary classifier using the Index of Union (IU) approach to determine whether a cell would be occupied by a loggerhead turtle given the identified relationships 59 . This analysis was done using the R package "ROCR" 60 . In short, the IU approach attempts to find an optimal cut point (c) that correctly classifies the fitted continuous probabilities of loggerhead presence as a 1 (present) or 0 (absent). The optimal value of c is that which minimizes the IU criterion: where Se is the sensitivity (true positives / (true positives + false negatives)), Sp is the specificity (true negatives / (true negatives + false positives)), and AUC is the Area Under the Receiver Operating Curve. The optimal cut-point was found to be c = 0.08 (i.e., predicted probabilities ≥ 0.08 were classified as 1, otherwise 0), with Se (c = 0.08) = 0.84, Sp (c = 0.08) = 0.70, and the AUC = 0.85. Using the optimal cut-point, we classified seasonal averages of presence probabilities by projected decade to identify cells that could be occupied by loggerhead turtles based on the bathythermal habitat associated with observed loggerhead occupancy patterns. We labeled the IU region classified to have loggerhead presence as the 'core habitat' . The fraction of cells in the study area that could be occupied by loggerheads was then calculated to explore trends in projected occupancy over time. We also calculated the region with the highest probability of presence by taking the top 25% of the predicted habitat values 27 . www.nature.com/scientificreports/

Results
In total, 196 loggerheads from 2009 to 2018 were fitted with satellite tags (Table 1). Turtles were either late-stage juveniles or adults with a mean (± SD) curved carapace length of 80.0 ± 9.7 cm. We found no qualitative difference in the seasonal movement patterns between presumed late-stage juveniles and adults and as such pooled across life-stage for this analysis. Filtering the location estimates from these tags yielded 45,840 daily locations within the NW Atlantic (north of 33.5°N and west of 64°W, the approximate southern and western boundaries of the US northeast continental shelf Large Marine Ecosystem 61 ), of which 44,865 daily locations occurred on the continental shelf in the MAB and were used for model fitting.
Model selection for explanatory variables supported a relationship between loggerhead presence, SST, and depth. SST alone explained 15.4% of the variability in loggerhead presence, while including only bottom depth explained 4.1%. SST and depth combined explained 20.1% of the variability in loggerhead presence. Based on the fitted model, loggerheads occur at SST between 11.0° and 29.7 °C and at depths between 0 and 105.0 m (Fig. 2a,b). The overall predicted distribution for each month was consistent with the reconstructed tracks and indicated that the probability of loggerhead presence in the NW Atlantic shelf waters is highest from May through October. Portions of SNE and GB were estimated to have a higher probability of presence than the MAB during summer months.
Using the binary classifier resulted in a core habitat that consisted of temperatures between 15.0° and 28.0 °C and depths between 8.0 and 92.0 m (Fig. 2c). The highest probability of presence occurred in regions with SST between 17.7° and 25.3 °C, and depths between 26.1 and 74.2 m. More specifically, turtles tended to occupy regions of the MAB with SST closest to 21.5 °C at depths closest to 50 m.
The CM2.6 model projected that warmer SST would push farther inshore and north through all seasons (Fig. 3). Mean SST within the shelf region is expected to gradually increase in the first 40 years, then intensify over the following 40 years. The probability of presence for loggerheads in the MAB is projected to increase from the observed May-October season, to an April-December season within 20-60 years, encompassing the entire spring, summer and fall seasons (Fig. 4). In particular, fall is expected to have the largest increase in available thermal habitat for loggerheads, followed by spring (Fig. 5). Minimal changes in winter are expected in terms of available suitable habitat over the next 80 years, while available habitat in the summer is expected to slightly decline as the most southern portions of the MAB warm beyond the range of our established temperature env elop.

Discussion
Based on the current relationship between loggerhead occurrence and SST, projected increases in SST as a result of climate change will likely result in an expansion of potentially suitable loggerhead habitat in the NW Atlantic. While the CM2.6 model projects a ~ 3 °C increase in mean SST for all seasons within 60-80 years, this increase in SST is not expected to be uniform-the GOM is expected to warm faster than the more southern regions 46 . Shelf waters in the MAB are typically cooler than offshore waters with these two water masses bounded by the Slope Sea, a narrow band of ocean along the continental slope between the shelf and the Gulf Stream 62 . However, CM2.6 projects a change to these conditions with the southward flowing cool Arctic waters of the Labrador Current retreating along with a north-and shore-ward shift of the Gulf Stream, allowing warmer water to enter the NW Atlantic shelf region 46 . These oceanographic features are expected to provide continued suitable bathythermal habitat for loggerheads in the MAB, SNE and GB, and may also expand the range of suitable habitat beyond the observed season 29,58 and northward into GOM.
Our results matched well with previous research attempting to establish the bathythermal range for loggerheads in the region. Hawkes et al. 26 found that satellite tagged loggerheads occupied a similar range of temperatures (12.8°-30.4 °C) and depths (0.25-104.4 m), while Hawkes et al. 63 updated these results with additional telemetry data and narrowed the ranges to SST = 18.2°-29.2 °C and depth = 3-89 m. Additionally, Mansfield et al. 64 identified that tracked loggerheads that stayed within the neritic zone of the NW Atlantic experienced an SST range between 9.0° and 29.3 °C across all seasons. During the summer the MAB experiences a sharp  36 and those that remain in the shelf waters of the southern US regularly inhabit environments with a much higher SST than observed in the MAB 66 . In addition, with our tagged turtles inhabiting the northern edge of western Atlantic loggerhead distribution 29 , the range of SSTs and depths reported here may be slightly narrower than the temperature and depth ranges in which these animals are able to thrive. Loggerhead response may thus not match the rate of the projected northward shift in their available thermal envelope or their movements may be driven be other correlating factors 27 . Despite our use of only two explanatory variables (SST and depth), our model results showed similar patterns of loggerhead distribution to sightings, strandings, and bycatch data, with slight variations due to the www.nature.com/scientificreports/ unique practices of each fishery. Braun-McNeill et al. 67 found that 11 °C was a conservative minimum SST threshold that aligned well with sea turtle distribution in the NW Atlantic shelf region from ten years of sightings, strandings and bycatch data. Swimmer et al. 14 identified that loggerheads were most often caught in long lines when SST ranged between 18° and 24 °C and hook depth was 22 m or shallower; however, these results included a much larger portion of the greater Atlantic Ocean. Gillnet bycatch between Massachusetts and North Carolina, having occurred nearly year-round, was within a broader range of SST and depth (SST = 8.6°-27.8 °C; depth = 1.8-76.8 m 68 ). Observed bycatch in scallop dredges was limited to SST between 18° and 25 °C and depths of 36-68 m 69 . These values from the scallop fishery aligned closest to our ranges for highest probability of turtle presence (SST = 17.7°-25.3 °C and depths = 26.1-74.2 m) because by-caught turtles in the scallop fishery had a high spatiotemporal overlap with when and where we caught the majority of our tracked loggerheads. Simultaneous integration of multiple data streams during statistical model estimation could help with more robust characterization of habitat for marine species in addition to this corroborative evidence, particularly for cases with incomplete and imperfect data resolution and could be a target for future research. We built upon projections calculated by Witt et al. 70 of suitable loggerhead habitat by developing probability models with monthly projections. Witt et al. 70 created annual projections using a 15 °C threshold and suggested that for 90% of the year the MAB and areas north are unsuitable habitats for loggerheads, even as ocean temperatures warm. However, Witt et al. 70 added that during summer months, loggerheads would regularly forage farther north than their annualized habitat suitability contours. Results of our winter projections matched well with annualized contours from Witt et al. 70 , indicating that loggerheads would have a very low probability of entering the MAB during this season, remaining farther south, or potentially in warmer offshore waters. However, throughout the remainder of the year, we projected that the loggerhead thermal habitat envelope would expand into MAB shelf waters earlier in the spring, continue moving north beyond the observed range in SNE and GB, and retreat south later in the fall. This corresponds closely with the trend of the spring and fall 15 °C SST threshold, with this isotherm continuing north in both shelf and offshore waters throughout the next 80 years.
Although observed data from aerial surveys indicates that SNE and GB are already suitable for loggerheads during portions of the year 18 , turtles are likely not as abundant farther north due to the shorter thermal window and the existing availability of prey resources in the MAB. With the projected increased thermal window, loggerheads may have more time to explore and actively forage within the northern shelf regions, while reducing www.nature.com/scientificreports/  www.nature.com/scientificreports/ competition for prey resources in the more heavily populated MAB, creating higher value to the longer distance migration 71 .
In general, there are likely other shifts in loggerhead distribution that could occur as environmental conditions in the NW Atlantic change. For example, rising SST is expected to contribute to an increase in hurricane activity and intensity within this region 72 . Crowe et al. 37 identified that some loggerheads in the MAB initiated their southern migrations well in advance of the established seasonal trends due to the passing of Hurricane Irene. As a result, the seasonal movement patterns we have projected may be disrupted by an increase in hurricane activity. In addition, for this study we focused on foraging within the continental shelf; however, if conditions become unsuitable for loggerheads in the MAB, turtles may forage off-shelf for extended periods of time instead of migrating north 73,74 . Our tagging data do indicate that loggerheads venture off-shelf on rare occasions, and adult and sub-adult loggerheads have been tracked foraging in pelagic environments within the NW Atlantic and throughout the world 64,[74][75][76][77][78] . In the MAB, loggerheads have been observed foraging pelagically on jellyfish 79,80 , and the off-shelf regions adjacent to GOM, GB, MAB and SNE in the Northwest Atlantic are known migratory corridors and feeding grounds for leatherback turtles, obligate jellyfish foragers 81,82 .
As the thermal habitat in the MAB through GB shelf region changes, this will also likely cause shifts in prey densities and species compositions. Using the same CM2.6 global climate model, Kleisner et al. 20 described the shifts in available thermal habitats for over 30 commercially valuable marine species within the NW Atlantic continental shelf. In general, Kleisner et al. 20 projected the expansion of available thermal habitats for southern species, and the reduction in available thermal habitats for northern species during the spring and fall seasons. Amongst these species, the vulnerability of the Atlantic sea scallop (Placopecten magellanicus) to climate change may provide an indication of how turtles may shift habitat usage. Atlantic sea scallops are a known prey item for loggerheads 83 and there is generally a high overlap between loggerhead and sea scallop distribution in the MAB based on preferred depth range 84 . Recent research by Rheuban et al. 85 has found that each scallop stock (MAB and GB) may react differently to climate change and that the more northern GB population may be slightly buffered from negative impacts due to the different mechanisms for larval recruitment between the stocks. However, using changes in ocean temperature and salinity from CM2.6, Tanaka et al. 55 projected substantial scallop habitat declines throughout the MAB and GB but increased habitat in coastal GOM. As a result, the benthic community could substantially change in the MAB, potentially causing loggerheads to seek other, perhaps more northerly, shelf habitats for prey resources.
Climate based shifts in the distributions of sea turtles and commercially harvested species may change future patterns of bycatch. Changes to loggerhead range and seasonality may create spatial overlap with fisheries that have not previously needed sea turtle conservation measures. In the NW Atlantic, Kleisner et al. 20 identified that most commercial fisheries would likely have to change their practices due to climate change, as distances and directions from ports to fishing grounds are expected to substantially change due to projected shifts in commercial species' distributions. For example, the Atlantic sea scallop fishery developed gear modifications, (Turtle Deflector Dredge and Turtle Chains 83 ) to reduce the bycatch of sea turtles and mandated these measures for boats fishing in the MAB, specifically west of − 71°W, from May through November 86 . Our model projects that within the next 20-40 years, loggerheads could forage within the NW Atlantic shelf region outside the spatial and temporal range these scallop gear modifications are required. Because the scallop industry has already developed a dredge effective at reducing turtle bycatch, adjusting the gear to remain efficient in the more northern scalloping grounds and expanding its usage could be an effective solution with minimal economic impacts to the fishery 83 . However, northern fisheries that use pelagic long lines, trawls and gillnets have the potential to see increases in turtle bycatch if fisheries management does not adapt to projected environmental changes. For example, the bottom trawl fishery operating in the MAB, SNE and GB, from 2014 to 2018, had the highest number of estimated sea turtle interactions occur north of 39°N, which is farther north than in previous years 87 .
Overall, sea turtle seasonal habitat usage and distribution is certainly linked to a broader range of environmental variables beyond SST and depth 27 , as well as biological factors like the availability of prey resources 88 , reproductive cycles and life stage 89 . However, given the availability of data and what is known about loggerhead ecology in general, the type of SDM we present provides a reasonable assessment of the potential drivers for the distribution of this cohort of loggerheads 30 . To truly determine how climate change will impact these turtles will require continued monitoring, particularly in the MAB, SNE and GB. Our results can guide expectations for likely future turtle distributions and inform discussions to plan for climate change-resilient conservation measures. www.nature.com/scientificreports/