The role of oceanographic scales in shaping highly mobile marine predator distributions

Abstract

Marine animals live in a dynamic environment, where a wide range of drivers and processes impact their movements and distributions. These processes occur over multiple spatio-temporal scales, from fine scale phytoplankton blooms and zooplankton patches to larger scale climatic events such as El Niño or climate change. In a dynamic ocean, the predictability of ocean features and processes vary across multiple scales. Marine animals interact with all these processes, and they all have the potential to impact animal distribution. However, which processes and scales predominantly predict the distributions of highly mobile predators is currently unknown. Here, we use electronic tagging data (265 sharks tagged in the Pacific) to investigate the scales of environmental selection of three pelagic shark species (the salmon shark Lamna ditropis, the blue shark Prionace glauca, and the shortfin mako Isurus oxyrinchus) across an array of spatio-temporal resolutions (from 9 km – 1 day to 500 km – climatology) for both Eulerian and Lagrangian variables. While Eulerian and Lagrangian variables at all scales tested have predictive power, we find that the 100 km – 1 year scale best predicted predator locations, indicating that larger scale, annually averaged signals outperform the other scales in predicting predator location.

Introduction

Animal movement is a fundamental characteristic of life, playing a key role in the fate of individuals as they move across different spatio-temporal scales in order to fulfill their life history requirements, such as acquiring food, finding mates, escaping predation, or seeking favorable environmental conditions^{1, 2–3}. These movements are driven by a range of factors, including both internal (navigation capacity, mobility, physiology, life history) and external environmental conditions⁴. However, it remains unclear how changes in external environmental conditions across different time and space scales affect animal movements and distributions.

In the marine environment, a wide range of external processes may influence the movements, and resulting distribution, of animals. Marine ecosystems are inherently dynamic, with processes occurring over multiple nested spatial and temporal scales⁵. The scales of these processes can be visualized using a Stommel diagram⁶, a 2D representation of small scale processes that merge to form new phenomena at larger spatio-temporal scales (Fig. 1) from fine (localized processes such as small phytoplankton blooms) to large (climatic events such as El Niño) spatio-temporal scales. Marine organisms move in a fluid environment, where characteristics of the habitat are modified by currents, eddies, and fronts. As such, Eulerian variables (measured at fixed points in time and space) are not the only variables that matter. Lagrangian structures, which take into account the history of the water parcel, can carry strong environmental cues and are also drivers of marine organisms’ distribution^{7, 8–9}.

[See PDF for image]

Fig. 1

Stommel diagram showing the main oceanic biophysical processes, along with the scales investigated in this project. Lettered panels show the FTLE (Finite Time Lyapunov Exponent) field at different scales, on December 31^st, 2000. A: 9 km – 1 day, B: 9 km – 8 days, C: 50 km – 30 days, D: 100 km – 365 days, E: 500 km – climatology. ENSO: El Niño Southern Oscillation, NAO: North Atlantic Oscillation.

These cascading processes, both Eulerian and Lagrangian, interact across spatial and temporal scales, creating emergent bio-physical phenomena known to influence the distribution of marine predators. Small scale features (on the scale of hours to days, and 10’s of meters to kilometers) have been observed to attract sharks^10,11 and to increase foraging activity in seabirds¹². Mesoscale flow features (on the scale of weeks to months, 10’s of kms) have been shown to aggregate a variety of predators⁹, including sharks⁸, elephant seals¹³, penguins^14,15, tunas¹⁶, and seabirds⁷. Seasonal processes, including changes in abiotic and biotic conditions, drive migrations of countless marine species, including fishes, whales, sharks, and sea turtles^{17, 18, 19, 20–21}. Global scale processes, such as the El Niño Southern Oscillation, have also been shown to impact predator movements and population dynamics, from the regional distribution of sharks²², to movement patterns of leatherback turtles to their nesting grounds²³. Finally, climate change has already modified the distribution of marine species across taxa^{24, 25–26} and is projected to continue impacting the distribution of animals well into the future²⁷.

Given the variability and complexity of life history requirements, animals likely select habitats across multiple time and space scales to acquire resources and enhance their fitness. This is why strong arguments can be made about the importance of different scales in understanding animal distributions, which often leads to the truism that all scales are ecologically important. Here, we compare these distinct scales against each other to understand which are the most explanatory in predicting the distributions of these marine organisms. The benefit in analyzing how features of different scales influence animal distributions provides basic insight into the dominant oceanographic scales driving highly mobile predator distributions in the marine environment, and consequently help researchers predict species’ responses to global change^28,29 and design effective management of marine resources. For example, a species whose dominant scale is 9 km – 1 week may be more reactive, modifying its distribution rapidly in response to smaller scale dynamics, such as blooms, eddies, or even the passing of hurricanes. In contrast, a species whose dominant scale is 500 km – climatology will exhibit movements driven by much larger scale features, such as tracking large-scale oceanographic features or processes, climatological dynamics, or climate change. In between these scales, the distribution of animals will be primarily driven by intermediate scale phenomena, such as seasonal processes or interannual variability of oceanographic conditions. For example, a species primarily following seasonal cycles may exhibit marked seasonal migrations, and have a dominant selection scale of around 50 km – 30 days. The goal of this study is to understand the relative importance of Eulerian and Lagrangian oceanographic predictors along the spatial and temporal cascade of oceanic processes.

Here, we investigate the scales of environmental selection of three pelagic shark species, the salmon shark Lamna ditropis, the blue shark Prionace glauca, and the shortfin mako Isurus oxyrinchus. We selected these three species because they have been relatively well studied in the eastern North Pacific³⁰. They are highly migratory, pelagic species that have extensive tracking data sets and large home ranges, making them suitable for testing patterns of habitat selection across spatial and temporal scales. Using existing tagging data^{30, 31, 32–33}, we developed species distribution models (SDM) at an array of spatio-temporal resolutions (from 9 km – 1 day to 500 km – climatology, as well as a model with variable-dependent spatio-temporal resolution for comparison). We investigate which scale is the most predictive of these marine predators’ distributions and explore the relative importance of the different Eulerian and Lagrangian variables at each scale.

The scale at which organisms associate the most with environmental parameters was found to vary as a function both of the environmental variable and of the species considered. Yet, the 100 km – 365 days scale (annual) resolution performs better than the other fixed resolutions for the three species we considered, revealing the potential importance of larger spatial and temporal scale processes (e.g. El Niño Southern Oscillation) as opposed to fine-scale processes when considering habitat selection of highly migratory pelagic sharks. In addition, we show that Lagrangian variables significantly contribute to our understanding of and ability to predict pelagic animal distributions. As expected, carefully selecting variable-dependent resolutions within a single model outperforms all fixed resolution models, and provides an important benchmark for comparing the relative differences between the fixed resolution models. We provide a method based on a metric called the Information Value (a metric used to distinguish binary outcomes such as presence/absence based on a continuous variable such as environmental parameters) to guide future studies when investigating environmental selection and when selecting the most relevant resolutions to build species distribution models.

Results

To examine the movements of sharks and define their distributions in relationship to ocean characteristics, we used tracks from ARGOS satellite tags (Wildlife Computer SPOT tags) from a large tracking data set (TOPP)³⁰, processed with a state-space model³⁴. For three shark species, we built species distribution models (SDM) using eight environmental variables at 6 different spatio-temporal scales (Fig. 1 and Supplementary Table 1). Five models were built at scales that match those of satellite oceanographic data sets at 9 km – 1 day, 9 km – 8 days, 50 km – 30 days, 100 km – 365 days, and 500 km – climatology, which follows the cascade of emergent processes in space and time. In addition, a model (referred to as “Flexible scales”) was built using different scales for each variable, to account for the fact that organisms may select for different environmental variables at different scales. These flexible scales were chosen by maximizing the Information Value (IV).

The IV of oceanographic variables varied with spatio-temporal scales (Fig. 2). For Lagrangian variables, the scales with the highest IV were very coarse, clustered around 500 km and climatological time scales. The relative importance of Eulerian variables had greater variability across different scales. For example, the scale of 500 km – climatology was most important for photosynthetically active radiation (PAR), whereas intermediate temporal scales (8 to 30 days) were most important for euphotic depth (Zeu) and chlorophyll concentration (chl), and fine spatial scales (9 km) were most important for sea surface temperature (SST).

[See PDF for image]

Fig. 2

Information value for all environmental predictors (columns) and species (rows) considered. Black square outlines refer to the scales with the highest Information Value (IV), and red squares to the actual scale used in the “Flexible scales” model (in case the highest IV was at the data-poor 9 km – 1 day scale, which had significant gaps in satellite coverage). Higher (lighter) values reflect greater predictive power. Acronyms used for environmental variables are as follows: chl is chlorophyll concentration, SST is sea surface temperature, PAR is photosynthetically active radiation, Zeu is euphotic depth, FTLE is finite-time Lyapunov exponent, Delta is dilation rate, and Z is bathymetry.

The performances of the SDMs built at fixed scales are qualitatively similar for all three species (Fig. 3 and Supplementary Fig. 4). SDM performances were evaluated using three metrics: deviance explained, area under the curve (AUC), and true skill statistic (TSS). Except for the deviance explained for mako sharks, the fixed scale model with the best performances is 100 km – 365 days (performances were compared using ANOVA and Tukey HSD tests, p < 0.05) across all metrics. Nevertheless, the flexible scales model performed better than all fixed scale models when evaluated with test data, except for mako sharks, where performances of the flexible scales model and the 100 km −365 days model were comparable.

[See PDF for image]

Fig. 3

Mean (+/- SD) true skill statistic (TSS) for blue sharks, shortfin mako sharks, and salmon sharks at different scales. Model simulations were run 50 times to generate the means and standard deviations. Significant statistical differences between scales are denoted by different letters (ANOVA and Tukey HSD, p < 0.05).

Using boosted regression trees to develop species distribution models allows us to evaluate the relative contributions of the different oceanographic predictors to predicted distributions (Fig. 4 and Supplementary Fig. 6). Relative contributions vary across scales and species. For blue sharks, all predictors have a relatively similar contribution at the 100 km – 365 days scale, but Lagrangian predictors are much less important in the flexible scales model. For mako sharks, the dilation rate (a Lagrangian predictor describing particle aggregation) is the most important variable in both the 100 km – 365 days and flexible scales models, closely followed by SST. For salmon sharks, bathymetry is the most important predictor at the 100 km – 365 days scale, followed by dilation rate and SST. In the flexible scales case, SST is the most important predictor, closely followed by euphotic depth and chlorophyll concentration. For these two best performing models, the relative contributions of all variables are fairly similar. This is not the case for the finer scales or the 500 km – climatology (Supplementary Fig. 6). For models at these scales, for all three species, bathymetry is generally the predictor with the most important contribution, followed by SST and chlorophyll (fine scales) or the day of year (500 km – climatology scale).

[See PDF for image]

Fig. 4

Relative importance (%, mean +/- SD) of each environmental variable for models built using the 100 km – 365 days (a-c) and flexible scales (d-f), for blue sharks (a,d) shortfin makos (b, e), and salmon sharks (c, f). Acronyms used for environmental variables are defined in the legend of Fig. 2 (except for DOY, which stands for day of year).

These results will likely be influenced by the attributes of the data that were used to build the models, so we tested the sensitivity of the outputs to the characteristics of the two streams of data we used: animal locations and environmental variables. For animal locations, the salmon shark locations (and their associated uncertainty) used were processed in a state-space model³⁰. We compared the original state-space model performances to tracks built using a more recent state-space model called AniMotum³⁵ (Supplementary Fig. 8 and 9). The model performances are relatively similar for all scales, with the flexible scales model still performing the best on test data. Note that the relative performances of fixed scales are more consistent for the models built using AniMotum-processed data. For example, the TSS of the 50 km – 30 days, 100 km – 365 days, and 500 km – climatology AniMotum models are statistically similar. For environmental variables, we tested the sensitivity of our results to different Lagrangian data sources. Our Lagrangian predictors are based on HYCOM, a data-assimilative model, and we ran our fixed scales models using OSCAR, a dataset of sea surface currents directly estimated from sea surface height, surface winds, and SST³⁶. Because the native spatial resolution of OSCAR is 1/4°, we only built models at the scales 50 km – 30 days, 100 km – 365 days, and 500 km – climatology (Supplementary Fig. 12 and 13). The performances of models built with both types of tracking data outputs are similar, and the relative contributions of Lagrangian variables are also similar, in particular at the 100 km – 365 days scale. The main difference is for mako sharks at the 500 km – climatology scale, where the dilation rate has the strongest contribution of all variables (in front of bathymetry and day of year, which were the most important in the model based on HYCOM).

Finally, we visually compared the outputs produced by our species distribution models (Supplementary Fig. 10 and 11). Predicted distributions of sharks over the model spatial extent are qualitatively similar across scales, and consistent with the known distributions of these organisms in the East Pacific^31,32,37. Because of data gaps in environmental observations, there are more gaps in predictions at the finest (9 km) spatial scales.

Discussion

We used satellite tracking data of three shark species and compared species distribution models built at different spatio-temporal scales to determine which scale best predicts the distributions of marine predators. We found that all scales have predictive power, including very coarse spatial grids at a climatological time scale. The most predictive fixed spatio-temporal scales were found to be 100 km – 1 year. Surprisingly, using the finer spatio-temporal scales did not lead to better predictions of animal locations, with model performances steadily increasing as we use coarser scales, until we reach the 100 km – 1 year scale.

This 100 km – 1 year scale consists of an average of processes such as changes in mixed layer depth and seasonal biomass cycles (Fig. 1 and Supplementary Fig. 14). While all these processes are usually tightly controlled by seasonality, one of the primary driving forcings of ocean ecosystems³⁸, our results indicate that sharks in this analysis appear to not respond to seasonality as reliably as to annual averages of oceanographic factors, influenced by multiannual processes such as El Niño Southern Oscillation. Broadly, our results indicate that understanding the history of a water mass, both in a Eulerian (broad scale yearly averages) and a Lagrangian (along-trajectory averages) framework, improves our understanding of the distribution of highly mobile predators.

Our results indicating annually averaged conditions are more predictive of animal locations does not mean finer scales are not important for animal movement and behavior. While the dominant scale driving shark movements herein is 100 km – 1 year, other scales are still important and hold significant predictive power as to where marine animals are distributed. In addition, we could not test the importance of scales finer than 9 km – 1 day as there is no finer environmental data available for all our environmental products for such a large region – and the spatio-temporal resolution of the animal movement data used here will influence these results (for example, tags recording very precise locations at a high frequency would be more appropriate to investigate overlap with very fine scale features than ARGOS tags which have a location accuracy on the scale of 100’s of meters to several kms). Several animals, such as penguins¹⁵, basking sharks¹⁰, elephant seals³⁹, and whales⁴⁰ have been found to associate with very specific features (patches) when foraging. It is therefore likely that animals use environmental factors at different cascading scales when moving. For example, they could use broad scale annual features to decide on their migratory pathways and destinations – within 100 kms – and then use smaller scale, immediate cues to forage in very small-scale features when encountering a high-density food patch and engaging in area-restricted search^41,42. Future work could include environmental products (such as prey or predator distribution) that may be important at very fine scales. Including these environmental variables would probably modify our understanding of the predominant scales driving movement, as these scales are specific to each environmental product used – as exemplified by the performances of our flexible-scale models. For example, fine-scale spatio-temporal selection for food resources was shown for blue whales in the California Current Ecosystem, an area encompassed by our study⁴³.

Animals can readily detect these fine-scale features, such as high-density food patches, when in their vicinity (e.g. by vision, olfaction, mechanoreception, or other sensory cues), but they can likely not sense environmental cues at a 100 km resolution. Therefore, they may rely on their prior experiences and memory when deciding where to move at larger scales. External factors interact with the navigation capacity of the individual, its internal state, and its memory to create the movement path⁴. There are several examples of predators using memory to navigate in the ocean, such as blue whales^44,45 and sharks^46,47. They could rely on geomagnetic maps to navigate the pelagic⁴⁸, bathymetric features such as the slope or specific structures (e.g. seamounts), or on chemosensory cues when homing in on a coastal location or prey^49,50. In addition, the importance of various scales could vary with the time of year or the life history of the animal: fine-scale spatial or temporal features could be important when searching for foraging grounds or when deciding on when to migrate (such as e.g. for blue whales⁵¹), but less important outside of these or other similarly sensitive periods. When roaming the pelagic, animals most likely do not target very precise, distant locations (such as a specific eddy), but they may direct their movements towards areas where they have historically encountered habitat that fulfills some basic life history requirement, for example where eddies consistently occur (e.g., for foraging, mating, resting, gestating, etc.). Given that pelagic sharks are highly mobile and are long lived, to at least 15–30+ years^{52, 53–54}, their behaviors, movements, and patterns of habitat selection are likely influenced by their lifetime of experiences. This suggests that they may have a greater capacity to learn what features or regions are on average more suitable, and use this knowledge to cue in on historically productive regions more than do less mobile or shorter-lived pelagic species. For example, salmon sharks demonstrate high season-dependent fidelity to well-identified high-use areas and bathymetric features in the North East Pacific⁵⁵.

This study also sheds light on what environmental variables are the most important for predicting shark distribution at each scale. Bathymetry and SST appear to be the most important predictors of animal distribution at finer scales, while the two best scales (100 km – 1 year and flexible scales) have a more balanced contribution from all variables (Supplementary Fig. 6). Lagrangian features (and in particular Delta, the dilation rate) appear particularly important at these scales, especially for mako and salmon sharks (Supplementary Fig. 6). This emphasizes the importance of considering Lagrangian features in addition to Eulerian variables when building species distribution models. It is noteworthy that Lagrangian features are typically signatures of mesoscale activity (scale of 10 s of kilometers), but that their contribution appears the most important when averaged over larger scales, at a yearly or even at a climatological scale. This emphasizes that animals cannot predict Lagrangian features, but rather that they position themselves in areas of high Lagrangian feature activity, to enhance their potential feature encounter rate – which can have foraging and physiological benefits^8,56. Given that mako sharks and salmon feed more on epipelagic prey than blue sharks, which are feeding primarily on mesopelagic prey such as squids^57,58, it makes sense that they would be more associated with mesoscale features that aggregate surface prey.

One caveat of this analysis is that we used Lagrangian features computed with outputs from the HYCOM model, which can sometimes misplace eddies by tens of kilometers^59,60. While this would explain why Lagrangian features are most important at coarser spatial scales, we also ran our analysis using OSCAR-derived Lagrangian features, which are not subject to the same bias as they are directly derived from satellite observations. (Supplementary Fig. 12 and 13). Results are qualitatively similar when using either Lagrangian sources, further strengthening the confidence in our results.

In addition to providing insight into the ecology of highly mobile pelagic predators, our results have practical applications. Depending upon the question and species, scientific efforts may not have to focus on acquiring ecological data at very fine scales when building species distribution models. In fact, the flexible scale models are only slightly better than fixed scales at 1 year, 100 km. Our results are consistent with previously published studies using SDMs on pelagic species, where SDMs built using pop-up archival tags (position is computed using a geolocation algorithm incorporating day length, timing of local noon, and sea surface temperature – leading to much larger position uncertainties, generally greater than 50 or 100 km) led to better predictions than SDM built with SPOT tags, which are more accurate with positions ranging from 100’s of meters to kilometers⁶¹. Conversely, the importance of Lagrangian features in our results emphasize that adding more environmental products (as opposed to more precise products), and products that incorporate the history of a body of water, can lead to better predictions. Another series of metrics that can be used are sub-surface metrics (e.g. isothermal layer depth and other indices of water column structure⁶²). Considering only surface metrics biases our vision of the 3D environment in which marine animals live and may create mismatches between our understanding of movement and the actual biological and ecological drivers of animal behavior^63,64. This is particularly relevant for species, like the blue shark, that primarily forage or move at depth. Movements to specific locations are much harder to interpret or predict if they involve subsurface processes, as subsurface observations are much rarer. One such example is the migration of great white sharks to the white shark café in the Northeast Pacific: uncertainties remain as to whether white sharks’ migration is motivated by reproductive needs or by the possibility to forage on mesopelagic resources⁶⁵.

Nevertheless, we want to stress that we found that bathymetry, SST, and chlorophyll are the most important predictors at smaller scales, supporting the historical use of these variables in models. Rather than just using either very fine or large scales, animals are likely making decisions based on multiple temporal and spatial scales when moving through their environment⁶⁶. Hence, considering multiple scales simultaneously may also increase the predictive power of SDMs. In addition, the scale at which different variables influence a species distribution can provide information on the nature and scale of the ecological dynamics that drive animal distribution.

Despite all these considerations, knowing what scale to use when building SDMs may still be challenging for other species than those investigated here, as most do not have a large amount of high-resolution SPOT tracks available. Building models with flexible scales for each species based on the relative selection of each variable (using e.g. max IV as presented here) can help resolve some of these issues. These results could then be integrated into management activities, such as dynamic ocean management measures, or the designation of marine protected areas – if it turns out that large spatial temporal scales are enough to predict with high accuracy the distribution of marine predators.

Materials and methods

Animal tracks, animal pseudo-absences, and multi-scale oceanographic products

Species presence data were obtained satellite tags deployed during the Tagging of Pacific Predators (TOPP) project³⁰ (Supplementary Table 2), and processed using a state-space model to obtain daily position estimates and associated uncertainties³⁴. A total of 69 tracks were used for blue sharks (Supplementary Fig. 1), 76 tracks for shortfin makos (Supplementary Fig. 2), and 120 tracks for salmon sharks (Supplementary Fig. 3). Animal pseudo-absences were generated using a convex hull approach (i.e. drawing random pseudo-absences in the convex hull defined by animal presence location), maximizing the environmental separation with animal presence compared to Brownian motion or Correlated Random Walks^67,68.

Both animal presence and pseudo-absences were matched to environmental variables (see Supplementary Table 1 for the list of environmental variables used) at multiple scales. Spatio-temporal scales included all combinations of the following spatial and temporal scales: 9 km, 50 km, 100 km, and 500 km; and 1 day, 8 days, 30 days, 365 days, and a climatology. The climatology is also referred to as the 10 years scale in this article, as it was produced by averaging all environmental data available over a 10-year period. All spatio-temporal scales coarser than 9 km – 1 day were obtained by averaging the available data⁶⁹. Environmental variables’ values at animal locations (and their corresponding pseudo-absences) with an uncertainty larger than the spatial scale of interest were discarded, to ensure that we would only retain environmental variables corresponding to the actual location of animals. For example, for an animal location with a standard error of 60 km (in either latitude or longitude), we would discard all environmental variables at a spatial scale of 9 and 50 km, but retain spatial scales of 100 and 500 km. See section “Sensitivity analysis to data sources” below for more details on the impact of this assumption.

Maximum information value

For all species and environmental variables, we computed the Information Value (IV) at each scale (Fig. 2). The IV is a data analysis tool that helps to determine which factor has the strongest predictive power or influence on the outcome of a binary dependent variable (here, presence or pseudo-absence of organisms)^70,71. The IV is computed by first dividing the values of the continuous environmental variables into n (here, n = 5) bins and calculating, for each bin i, the fraction of presence and absence records in each bin:

Then, for each bin, we compute the weight of evidence (WOE) defined as:and the information value is computed as:

The “flexible scales” model was run using, for each environmental variable, the scale with the highest information value (Fig. 2) – unless this scale was 9 km – 1 day, as this fine scale has a significant number of data gaps due to incomplete satellite coverage and cloud cover. In this case, the scale with the second highest IV was selected.

Species distribution modelling

For each species (blue shark, shortfin mako, and salmon shark), six species distribution models (SDM) were built. The probability of species presence was modeled as a function of environmental variables using boosted regression trees (BRTs)⁷². Five SDMs were built at fixed scales (9 km – 1 day, 9 km – 8 days, 50 km – 30 days, 100 km – 365 days, 500 km—climatology), following the spatio-temporal scales at which most oceanic processes occur (Fig. 1). The sixth SDM (referred to as “flexible scales”) was built using a different spatio-temporal scale for each variable, corresponding to the scale with the maximum information value (or the second-highest information value, in case the highest information value was at the data-poor 9 km – 1 day scale). Co-linearity between the different environmental variables was not an issue as the BRT framework automatically handles co-linearity effects^62,72. All models were built using a Bernoulli response, appropriate to the response variable (presence and absence). All models were built using a tree complexity of 3, a learning rate of 0.01, and a bag fraction of 0.6^62,72.

Resulting models were evaluated using explained deviance, Area Under the receiver operating Curve (AUC), and true skill statistics (TSS). Explained deviance gives an estimate of how well the model explains the data, while AUC and TSS assess the predictive power of models on test data not used to build the model. Each model was built using 80% of the available data, and AUC and TSS were evaluated with the remaining 20% of the data. Reported model statistics are the means and standard deviations of the outcomes of 50 model iterations, each built with a different fraction of the available data. Differences in model statistics between spatio-temporal scales were assessed for significance using ANOVA and a Tukey Honest Significance Differences (HSD) post-hoc test^62,73.

Variable importance was computed for each model, and habitat suitability maps were computed using final models for winter (January-March, averaged over the period 2004–2008) and summer (July–September, averaged over the period 2004–2008) to better visualize model outputs.

Sensitivity analysis to data sources

Because of (i) the differences in animal location data accuracy, and (ii) the availability of environmental products at different spatio-temporal scales, the amount of data available for the different scales can vary tremendously. In addition, each animal has unique surface intervals and thus provides distinctly different amounts of raw data. Therefore, two different sets of models were generated. The first set of models only uses animal locations with less than 9 km uncertainty at all scales, to ensure that the same input data was used for all models (Fig. 3, results presented in the main manuscript). The second set of models uses all matched environmental data, so that the impact of increased sample size at coarser spatial scales can be assessed (Fig. 3 and Supplementary Fig. 5–7).

How ARGOS animal location data are originally processed may also influence output results. Therefore, we applied our framework to salmon shark data processed by two different state-space models. We compared data processed by the first state-space model^30,34 applied to the TOPP dataset (hereafter called “Block et al. 2011” and used by default in this study) to data processed by a newer state-space model (called “AniMotum”³⁵. The similarity of the results (Supplementary Fig. 8 and 9) show that this framework is not overly sensitive to the type of post-processing used for animal location data. This may in part be due to the fact that we only selected the most precise data points, thereby eliminating model interpolations between raw ARGOS hits that tend to artificially straighten tracks.

Finally, the Lagrangian variables used in this analysis were generated using HYCOM, a data-assimilative model. While this entails perfect coverage over our study period at a fine resolution, Lagrangian Coherent Structures computed using HYCOM outputs can be misplaced by 50 to 100 km^59,60,74. Therefore, we also generated SDM using OSCAR, a global surface current database^36,75. Because of the native resolution of OSCAR data (1/4^°), we only built SDM with OSCAR at the 50 km – 30 days, 100 km – 365 days and 500 km – climatology resolutions (Supplementary Fig. 12 and 13). The performances of models run with both types of Lagrangian data are similar (Fig. 3 versus Supplementary Fig. 12). In addition, while HYCOM- and OSCAR-derived Lagrangian can appear different at these scales, the relative importance of Lagrangian products remains similar, especially at the 50 km – 30 days and 100 km – 365 days scales (Supplementary Fig. 6 versus 13).

Acknowledgements

This work was supported by NASA Earth Science division, ROSES 2020 Program (award number 80NSSC21K1145). James Ganong and Mike Castleton who assisted on supplying data were funded by the Block laboratory of Stanford University. We thank the two anonymous reviewers whose comments and suggestions helped us improve this work.

Author contributions

Study design: JP, ABC, and MJO. Environmental data processing: HSH and MS. Animal tracks acquisition: ABC and BAB. Analysis: JP, with contributions from ABC and MJO. Writing: JP, with contributions from all authors.

Funding

Earth Sciences Division, award #80NSSC21K1145.

Data availability

All code developed for this project is available at https://gitlab.com/ud3/scales-of-top-predator-habitat-selection and is available from the corresponding author on reasonable request. All oceanographic data are available at http://ocean-data.ceoe.udel.edu/ and described in Pinti et al. 2024⁷⁵, electronic tag data are available at https://mola.stanford.edu/DataLinks/; and are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1038/s41598-025-06486-9.

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

1. Dingle, H. Migration: The Biology of Life on the Move. 2nd Edition. (Oxford University Press, 2014).

2. Shaw, AK. Drivers of animal migration and implications in changing environments. Evol. Ecol.; 2016; 30, pp. 991-1007. [DOI: https://dx.doi.org/10.1007/s10682-016-9860-5]

3. Lubitz, N et al. The role of context in elucidating drivers of animal movement. Ecol. Evol.; 2022; 12, pp. 1-19. [DOI: https://dx.doi.org/10.1002/ece3.9128]

4. Nathan, R et al. A movement ecology paradigm for unifying organismal movement research. Proc. Natl. Acad. Sci. U.S.A.; 2008; 105, pp. 19052-19059.2008PNAS.10519052N1:CAS:528:DC%2BD1cXhsFamurfP [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/19060196][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2614714][DOI: https://dx.doi.org/10.1073/pnas.0800375105]

5. Haury, L. R., McGowan, J. A. & Wiebe, P. H. Patterns and Processes in the Time-Space Scales of Plankton Distributions. in Spatial Pattern in Plankton Communities (ed. Steele, J. H.) 277–327 (Springer US, Boston, MA). https://doi.org/10.1007/978-1-4899-2195-6_12. (1978).

6. Stommel, H. Varieties of oceanographic experience. Science; 1963; 139, pp. 572-576.1963Sci..139.572S1:STN:280:DC%2BC3cvgvFantg%3D%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/17788284][DOI: https://dx.doi.org/10.1126/science.139.3555.572]

7. Tew Kai, E et al. Top marine predators track Lagrangian coherent structures. Proc. Nat. Acad. Sci. USA.; 2009; 106, pp. 8245-8250.2009PNAS.106.8245T [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/19416811][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2677090][DOI: https://dx.doi.org/10.1073/pnas.0811034106]

8. Arostegui, MC; Gaube, P; Woodworth-Jefcoats, PA; Kobayashi, DR. Anticyclonic eddies aggregate pelagic predators in a subtropical gyre. Nature; 2022; 609, pp. 535-540.2022Natur.609.535A1:CAS:528:DC%2BB38XitlCiu7fO [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36071164][DOI: https://dx.doi.org/10.1038/s41586-022-05162-6]

9. Prants, SV. Marine life at Lagrangian fronts. Prog. Oceanogr.; 2022; 204, [DOI: https://dx.doi.org/10.1016/j.pocean.2022.102790] 102790.

10. Sims, DW; Quayle, VA. Selective foraging behaviour of basking sharks on zooplankton in a small-scale front. Nature; 1998; 393, pp. 460-465.1998Natur.393.460S1:CAS:528:DyaK1cXjs1Kgurc%3D [DOI: https://dx.doi.org/10.1038/30959]

11. Hoffmayer, ER; Franks, JS; Driggers, WB; Oswald, KJ; Quattro, JM. Observations of a feeding aggregation of whale sharks, rhincodon typus, in the north central gulf of mexico. Gulf Carib. Res.; 2007; 19, pp. 69-73.

12. Lieber, L et al. Selective foraging behavior of seabirds in small-scale slicks. Limnol. Oceanogr. Lett.; 2023; 8, pp. 286-294. [DOI: https://dx.doi.org/10.1002/lol2.10289]

13. Keates, TR et al. Foraging behavior of a mesopelagic predator, the northern elephant seal, in northeastern Pacific eddies. Deep Sea Res. Part 1 Oceanogr. Res. Pap.; 2022; 189, [DOI: https://dx.doi.org/10.1016/j.dsr.2022.103866] 103866.

14. Oliver, MJ et al. Central place foragers select ocean surface convergent features despite differing foraging strategies. Sci. Rep.; 2019; 9, 157.2019NatSR..9.157O [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/30655549][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6336854][DOI: https://dx.doi.org/10.1038/s41598-018-35901-7]

15. Veatch, JM et al. Lagrangian coherent structures influence the spatial structure of marine food webs. Commun. Earth Environ.; 2025; 6, pp. 1-11. [DOI: https://dx.doi.org/10.1038/s43247-025-02101-x]

16. Pagniello, CMLS et al. Tagging of Atlantic bluefin tuna off Ireland reveals use of distinct oceanographic hotspots. Prog. Oceanogr.; 2023; 219, [DOI: https://dx.doi.org/10.1016/j.pocean.2023.103135] 103135.

17. Luschi, P; Hays, GC; Del Seppia, C; Marsh, R; Papi, F. The navigational feats of green sea turtles migrating from Ascension Island investigated by satellite telemetry. Proc. Royal Soc. B: Biol. Sci.; 1998; 265, pp. 2279-2284.1:STN:280:DyaK1M%2FpvVWnug%3D%3D [DOI: https://dx.doi.org/10.1098/rspb.1998.0571]

18. Block, BA et al. Migratory movements, depth preferences, and thermal biology of Atlantic bluefin tuna. Science; 2001; 293, pp. 1310-1314.2001Sci..293.1310B1:CAS:528:DC%2BD3MXmtFCqtb0%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/11509729][DOI: https://dx.doi.org/10.1126/science.1061197]

19. Weng, KC et al. Migration and habitat of white sharks (Carcharodon carcharias) in the eastern Pacific Ocean. Mar. Biol.; 2007; 152, pp. 877-894. [DOI: https://dx.doi.org/10.1007/s00227-007-0739-4]

20. Horton, TW et al. Straight as an arrow: Humpback whales swim constant course tracks during long-distance migration. Biol. Let.; 2011; 7, pp. 674-679. [DOI: https://dx.doi.org/10.1098/rsbl.2011.0279]

21. Costa, DP; Breed, GA; Robinson, PW. New insights into pelagic migrations: Implications for ecology and conservation. Annu. Rev. Ecol. Evol. Syst.; 2012; 43, pp. 73-96. [DOI: https://dx.doi.org/10.1146/annurev-ecolsys-102710-145045]

22. Adams, GD; Flores, D; Flores, OG; Aarestrup, K; Svendsen, JC. Spatial ecology of blue shark and shortfin mako in southern Peru: Local abundance, habitat preferences and implications for conservation. Endangered Species Res.; 2016; 31, pp. 19-32. [DOI: https://dx.doi.org/10.3354/esr00744]

23. Saba, VS et al. An oceanographic context for the foraging ecology of eastern Pacific leatherback turtles: Consequences of ENSO. Deep Sea Res. Part 1 Oceanogr. Res. Pap.; 2008; 55, pp. 646-660. [DOI: https://dx.doi.org/10.1016/j.dsr.2008.02.006]

24. Pinsky, ML; Worm, B; Fogarty, MJ; Sarmiento, JL; Levin, SA. Marine taxa track local climate velocities. Science; 2013; 341, pp. 1239-1242.2013Sci..341.1239P1:CAS:528:DC%2BC3sXhsVWrs7nK [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/24031017][DOI: https://dx.doi.org/10.1126/science.1239352]

25. Braun, CD et al. Widespread habitat loss and redistribution of marine top predators in a changing ocean. Sci. Adv.; 2023; 9, pp. 1-10. [DOI: https://dx.doi.org/10.1126/sciadv.adi2718]

26. Mills, KE et al. Multispecies population-scale emergence of climate change signals in an ocean warming hotspot. ICES J. Mar. Sci.; 2024; 81, pp. 375-389. [DOI: https://dx.doi.org/10.1093/icesjms/fsad208]

27. Hazen, EL et al. Predicted habitat shifts of Pacific top predators in a changing climate. Nat. Clim. Change; 2013; 3, pp. 234-238.2013NatCC..3.234H [DOI: https://dx.doi.org/10.1038/nclimate1686]

28. Sequeira, AMM et al. Convergence of marine megafauna movement patterns in coastal and open oceans. Proc. Natl. Acad. Sci. U.S.A.; 2018; 115, pp. 3072-3077.2018PNAS.115.3072S1:CAS:528:DC%2BC1cXht1Chs7vO [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/29483242][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5866563][DOI: https://dx.doi.org/10.1073/pnas.1716137115]

29. Hazen, EL et al. Marine top predators as climate and ecosystem sentinels. Front. Ecol. Environ.; 2019; 17, pp. 565-574. [DOI: https://dx.doi.org/10.1002/fee.2125]

30. Block, BA et al. Tracking apex marine predator movements in a dynamic ocean. Nature; 2011; 475, pp. 86-90.1:CAS:528:DC%2BC3MXnvVekur4%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/21697831][DOI: https://dx.doi.org/10.1038/nature10082]

31. Weng, KC et al. Migration of an upper trophic level predator, the salmon shark Lamna ditropis, between distant ecoregions. Mar. Ecol. Prog. Ser.; 2008; 372, pp. 253-264.2008MEPS.372.253W [DOI: https://dx.doi.org/10.3354/meps07706]

32. Nasby-Lucas, N et al. Movements of electronically tagged shortfin mako sharks (Isurus oxyrinchus) in the eastern North Pacific Ocean. Anim. Biotelem.; 2019; 7, pp. 1-26. [DOI: https://dx.doi.org/10.1186/s40317-019-0174-6]

33. Weng, KC et al. Satellite tagging and cardiac physiology reveal niche expansion in salmon sharks. Science; 2005; 310, pp. 104-106.2005Sci..310.104W1:CAS:528:DC%2BD2MXhtVOjsr7O [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/16210538][DOI: https://dx.doi.org/10.1126/science.1114616]

34. Jonsen, I; Flemming, JM; Myers, R. Robust state-space modeling of animal movement data. Ecology; 2005; 86, pp. 2874-2880. [DOI: https://dx.doi.org/10.1890/04-1852]

35. Jonsen, ID et al. aniMotum, an R package for animal movement data: Rapid quality control, behavioural estimation and simulation. Methods Ecol. Evol.; 2023; 14, pp. 806-816. [DOI: https://dx.doi.org/10.1111/2041-210X.14060]

36. Dohan, K. Ocean Surface Current Analyses Real-time (OSCAR) Surface Currents - Final 0.25 Degree (Version 2.0). Preprint at https://doi.org/10.5067/OSCAR-25F20 (2022).

37. Maxwell, SM et al. Seasonal spatial segregation in blue sharks (Prionace glauca) by sex and size class in the Northeast Pacific Ocean. Divers. Distrib.; 2019; 25, pp. 1304-1317. [DOI: https://dx.doi.org/10.1111/ddi.12941]

38. Visser, AW et al. Seasonal strategies in the world’s oceans. Prog. Oceanogr.; 2020; 189, [DOI: https://dx.doi.org/10.1016/j.pocean.2020.102466] 102466.

39. Della Penna, A; De Monte, S; Kestenare, E; Guinet, C; D’Ovidio, F. Quasi-planktonic behavior of foraging top marine predators. Sci. Rep.; 2015; 5, 18063.2015NatSR..518063D1:CAS:528:DC%2BC2MXitVWqtr%2FP [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/26666350][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4678296][DOI: https://dx.doi.org/10.1038/srep18063]

40. Fahlbusch, JA et al. Blue whales increase feeding rates at fine-scale ocean features. Proc. Royal Soc. B: Biol. Sci.; 2022; 289, 20221180. [DOI: https://dx.doi.org/10.1098/rspb.2022.1180]

41. Dorfman, A; Hills, TT; Scharf, I. A guide to area-restricted search: A foundational foraging behaviour. Biol. Rev.; 2022; 97, pp. 2076-2089. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/35821610][DOI: https://dx.doi.org/10.1111/brv.12883]

42. Spencer, ET et al. Searching day and night: diel hierarchical search patterns in a large marine predator that never stops swimming. Anim. Behav.; 2025; 223, [DOI: https://dx.doi.org/10.1016/j.anbehav.2025.123152] 123152.

43. Ryan, JP et al. Oceanic giants dance to atmospheric rhythms: Ephemeral wind-driven resource tracking by blue whales. Ecol. Lett.; 2022; 25, pp. 2435-2447. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36197736][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9827854][DOI: https://dx.doi.org/10.1111/ele.14116]

44. Fagan, WF. Migrating whales depend on memory to exploit reliable resources. Proc. Natl. Acad. Sci. U.S.A.; 2019; 116, pp. 5217-5219.2019PNAS.116.5217F1:CAS:528:DC%2BC1MXlsFelt7k%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/30804183][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6431177][DOI: https://dx.doi.org/10.1073/pnas.1901803116]

45. Abrahms, B et al. Memory and resource tracking drive blue whale migrations. Proc. Natl. Acad. Sci. U.S.A.; 2019; 116, pp. 5582-5587.2019PNAS.116.5582A1:CAS:528:DC%2BC1MXlsFeksb0%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/30804188][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6431148][DOI: https://dx.doi.org/10.1073/pnas.1819031116]

46. Guttridge, TL; Myrberg, AA; Porcher, IF; Sims, DW; Krause, J. The role of learning in shark behaviour. Fish Fish.; 2009; 10, pp. 450-469. [DOI: https://dx.doi.org/10.1111/j.1467-2979.2009.00339.x]

47. Séguigne, C; Vignaud, T; Meyer, C; Bierwirth, J; Clua, É. Evidence of long-lasting memory of a free-ranging top marine predator, the bull shark Carcharhinus leucas. Behaviour; 2023; 160, pp. 1303-1318. [DOI: https://dx.doi.org/10.1163/1568539X-bja10240]

48. Lohmann, KJ; Goforth, KM; Mackiewicz, AG; Lim, DS; Lohmann, CMF. Magnetic maps in animal navigation. J. Comp. Physiol. A Neuroethol. Sens. Neural Behav. Physiol.; 2022; 208, pp. 41-67. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34999936][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8918461][DOI: https://dx.doi.org/10.1007/s00359-021-01529-8]

49. Grassman, M. Chemosensory orientation behavior in juvenile sea turtles. Brain Behav. Evol.; 1993; 41, pp. 224-228.1:STN:280:DyaK3s3jvVWjuw%3D%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/8477344][DOI: https://dx.doi.org/10.1159/000113842]

50. Nosal, AP; Chao, Y; Farrara, JD; Chai, F; Hastings, PA. Olfaction contributes to pelagic navigation in a coastal shark. PLoS ONE; 2016; 11, pp. 1-17. [DOI: https://dx.doi.org/10.1371/journal.pone.0143758]

51. Oestreich, WK et al. Acoustic signature reveals blue whales tune life-history transitions to oceanographic conditions. Funct. Ecol.; 2022; 36, pp. 882-895.1:CAS:528:DC%2BB38XhvV2lsrnI [DOI: https://dx.doi.org/10.1111/1365-2435.14013]

52. Goldman, KJ; Musick, JA. Growth and maturity of salmon sharks (Lamna ditropis) in the eastern and western North Pacific, and comments on back-calculation methods. Fish. Bull.; 2006; 104, pp. 278-292.

53. Natanson, LJ et al. Validated age and growth estimates for the shortfin mako, Isurus oxyrinchus, in the North Atlantic Ocean. Environ. Biol. Fishes; 2006; 77, pp. 367-383. [DOI: https://dx.doi.org/10.1007/s10641-006-9127-z]

54. Nakano, H; Stevens, JD. The Biology and Ecology of the Blue Shark, Prionace Glauca. Sharks Open Ocean: Biol., Fish. Conserv.; 2009; [DOI: https://dx.doi.org/10.1002/9781444302516.ch12]

55. Arnoldi, NS et al. Salmon shark seasonal site fidelity demonstrates the influence of scale on identifying potential high-use areas and vulnerabilities. Mar. Ecol. Prog. Ser.; 2024; 735, pp. 125-140.2024MEPS.735.125A [DOI: https://dx.doi.org/10.3354/meps14565]

56. Braun, CD; Gaube, P; Sinclair-Taylor, TH; Skomal, GB; Thorrold, SR. Mesoscale eddies release pelagic sharks from thermal constraints to foraging in the ocean twilight zone. Proc. Natl. Acad. Sci. U.S.A.; 2019; 116, pp. 17187-17192.2019PNAS.11617187B1:CAS:528:DC%2BC1MXhs1yhurvK [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/31387979][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6717292][DOI: https://dx.doi.org/10.1073/pnas.1903067116]

57. Kubodera, T; Watanabe, H; Ichii, T. Feeding habits of the blue shark, Prionace glauca, and salmon shark, Lamna ditropis, in the transition region of the Western North Pacific. Rev. Fish Biol. Fish.; 2007; 17, pp. 111-124. [DOI: https://dx.doi.org/10.1007/s11160-006-9020-z]

58. Preti, A et al. Comparative feeding ecology of shortfin mako, blue and thresher sharks in the California Current. Environ. Biol. Fishes; 2012; 95, pp. 127-146. [DOI: https://dx.doi.org/10.1007/s10641-012-9980-x]

59. Huntley, HS; Lipphardt, BL; Kirwan, AD. Lagrangian predictability assessed in the East China Sea. Ocean Model; 2011; 36, pp. 163-178.2011OcMod.36.163H [DOI: https://dx.doi.org/10.1016/j.ocemod.2010.11.001]

60. Muscarella, P et al. Do assimilated drifter velocities improve lagrangian predictability in an operational ocean model?. Mon. Weather Rev.; 2015; 143, pp. 1822-1832.2015MWRv.143.1822M [DOI: https://dx.doi.org/10.1175/MWR-D-14-00164.1]

61. Braun, CD et al. Building use-inspired species distribution models: using multiple data types to examine and improve model performance. Ecol. Appl.; 2023; 66, [DOI: https://dx.doi.org/10.1002/eap.2893] e2893.

62. Brodie, S et al. Integrating dynamic subsurface habitat metrics into species distribution models. Front. Mar. Sci.; 2018; 5, pp. 1-13. [DOI: https://dx.doi.org/10.3389/fmars.2018.00219]

63. Hebblewhite, M; Haydon, DT. Distinguishing technology from biology: a critical review of the use of GPS telemetry data in ecology. Philos. Trans. Royal Soc. B: Biol. Sci.; 2010; 365, pp. 2303-2312. [DOI: https://dx.doi.org/10.1098/rstb.2010.0087]

64. Irvine, LM et al. Ecological drivers of movement for two sympatric marine predators in the California current large marine ecosystem. Mov. Ecol.; 2025; 13, 19. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/40102967][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11917063][DOI: https://dx.doi.org/10.1186/s40462-025-00542-9]

65. Jorgensen, SJ et al. Eating or Meeting? Cluster Analysis Reveals Intricacies of White Shark (Carcharodon carcharias) Migration and Offshore Behavior. PLoS ONE; 2012; 7, pp. 1-10. [DOI: https://dx.doi.org/10.1371/journal.pone.0047819]

66. Mannocci, L et al. Temporal resolutions in species distribution models of highly mobile marine animals: Recommendations for ecologists and managers. Divers. Distrib.; 2017; 23, pp. 1098-1109. [DOI: https://dx.doi.org/10.1111/ddi.12609]

67. Hazen, EL et al. Where did they not go? Considerations for generating pseudo-absences for telemetry-based habitat models. Mov. Ecol.; 2021; 9, 5. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/33596991][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7888118][DOI: https://dx.doi.org/10.1186/s40462-021-00240-2]

68. Pinti, J; Shatley, M; Carlisle, A; Block, BA; Oliver, MJ. Using pseudo - absence models to test for environmental selection in marine movement ecology : The importance of sample size and selection strength. Mov. Ecol.; 2022; 10, 60. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36581885][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9798696][DOI: https://dx.doi.org/10.1186/s40462-022-00362-1]

69. Pinti, J; Shatley, M; Huntley, H; Carlisle, A; Oliver, M. The interplay between animal location accuracy and the decorrelation length scale of environmental variables when investigating environmental selection in marine organisms. Mar. Ecol. Prog. Ser.; 2024; 732, pp. 1-14.2024MEPS.732..1P1:CAS:528:DC%2BB2cXos12rtrc%3D [DOI: https://dx.doi.org/10.3354/meps14546]

70. Prabhakaran, S. InformationValue: Performance Analysis and Companion Functions for Binary Classification Models. R package version 1.2.3. Preprint at (2016).

71. Breece, MW; Fox, DA; Haulsee, DE; Wirgin, II; Oliver, MJ. Satellite driven distribution models of endangered Atlantic sturgeon occurrence in the mid-Atlantic Bight. ICES J. Mar. Sci.; 2018; 75, pp. 562-571. [DOI: https://dx.doi.org/10.1093/icesjms/fsx187]

72. Elith, J; Leathwick, JR; Hastie, T. A working guide to boosted regression trees. J. Anim. Ecol.; 2008; 77, pp. 802-813.1:STN:280:DC%2BD1cvgsFOqsQ%3D%3D [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/18397250][DOI: https://dx.doi.org/10.1111/j.1365-2656.2008.01390.x]

73. Fournier, A; Barbet-Massin, M; Rome, Q; Courchamp, F. Predicting species distribution combining multi-scale drivers. Global Ecol. Conserv.; 2017; 12, pp. 215-226. [DOI: https://dx.doi.org/10.1016/j.gecco.2017.11.002]

74. Thoppil, PG et al. Ensemble forecasting greatly expands the prediction horizon for ocean mesoscale variability. Commun. Earth Environ.; 2021; 2, 89.2021ComEE..2..89T [DOI: https://dx.doi.org/10.1038/s43247-021-00151-5]

75. Pinti, J; Shatley, M; Huntley, HS; Oliver, MJ. Global multi-scale Eulerian and Lagrangian oceanic environmental variables for the period 2000-2009. ESS Open Archive; 2024; [DOI: https://dx.doi.org/10.22541/essoar.171986449.92834256/v1]