Transmitted daily summary data

To characterize large-scale patterns of vertical habitat use across individuals and deployments, we analyzed satellite-transmitted daily summaries of time-at-depth (TAD) and time-at-temperature (TAT). For clustering, we used an arcsine transformation on the TAD and TAT proportions to improve normality of these skewed data. We created a distance matrix to assess the similarities of each day of TAD and TAT data based on the seven common depth bins and 11 common temperature bins. We computed pairwise distances among samples using the Manhattan (“city-block”) metric, which sums absolute differences across dimensions, rather than Euclidean distance as the latter is known to be sensitive to differences in the scale and units of variables; therefore, standardization is generally required when attribute scales differ substantially [39]. In contrast, Manhattan-type metrics are widely used for quantitative variables measured on discrete or binned scales (e.g., counts, depth bins). From the distance matrix, we created a cluster tree using the true Ward clustering method [40] because of its robustness when handling outliers. We assessed the optimal number of clusters based on our data using the NbClust package for R [41].

For each day represented by TAD and TAT data, a geographic position (latitude and longitude) was estimated using a state-space hidden Markov model from the HMMoce package [42] for R. Further description of this methodology can be found in [26]. These positions were grouped by cluster to explore the geographic distribution of the resulting vertical behavior patterns. Days assigned to a given cluster were also identified within the available transmitted (n = 14) and archived (n = 2) time series data from 16 tags to investigate the fine-scale vertical habitat use within each cluster. All time series data were artificially coarsened to a common resolution (600 seconds) to standardize for visualization. The depth-temperature time series data were used to identify the onset of the southerly migration for each individual by calculating the date when the shark crossed the Gulf Stream North Wall, traditionally defined as where the 15ºC isotherm occurs at 200 m depth [43].

Recovered high-resolution archival data

Archival data from the two recovered tags provided a unique opportunity to characterize vertical habitat use and behaviors at a higher resolution. These two tags recorded depth (pressure), temperature, and light-level data at either 3- or 30- second resolution (hereafter “Shark 1” and “Shark 2”, respectively). These data were used to 1) identify the periodicity and prevalence of diel patterns in depth (i.e., diel vertical migration [DVM] on ~24 hour period); 2) estimate depths of specific daytime isolumes; and 3) detect bioluminescence at depth as a proxy for potential prey distribution. These metrics were used by Klöcker et al. [44] to study basking shark behavior in the Northeast Atlantic, enabling comparison between regions.

We used a continuous wavelet analysis to evaluate the periodicity and presence of diel patterns in the high-resolution, depth-time series from the recovered tags (as described in [44]). Analyses were conducted with a Morlet wavelet (x₀ = 6) using the WaveletComp package for R [45]. In addition to the default workflow, which applies global standardization of the time series, we implemented a local variance normalization using a sliding-window (7-day) z-score standardization prior to the wavelet transformation. This procedure accounts for non-stationary variance observed in the depth-time series (e.g., associated with habitat transitions), thereby allowing consistent assessment of periodicity. We assessed the statistical significance of the wavelet spectrum by generating 1,000 simulated time series for each individual based on a first-order autoregressive process (AR [1]) with p = 0.7 and using the observed mean. Wavelet power values that exceeded the bootstrapped 95% confidence limits were considered statistically significant and interpreted as evidence of non-random vertical migratory behavior within the time series. DVM was inferred when the wavelet power values at the 24-h period were statistically significant (p < 0.05). Although an ideal sinusoidal depth cycle would produce a single 24-h peak, real depth trajectories often include sharp descents or ascents or seasonal shifts in daylight regime, which generate additional harmonics (e.g., 12 h, 6 h) with diminishing power at higher frequencies. Because these harmonics do not reflect interpretable ecological behavior, they were excluded from further consideration.

To characterize habitat-specific differences in light penetration and identify depths likely favored by light-sensitive prey, we estimated the daily depth of a representative isolume for each individual using tag-recorded light measurements during daytime conditions (sun angle α > 6°). Only days with at least 20 valid observations were retained, and a three-day moving average was applied to smooth the resulting isolume depths for visualization. We focused on light-level values near 30 (LL30), corresponding to an irradiance range (~10^-10-10^-11 W cm²) commonly associated with preferred light environments of mesopelagic crustaceans (including known copepod prey in the North Atlantic) and global deep scattering layer communities [46]. Although the occupied light level range of such species may be broader, LL30 served as an approximate indicator of the depth at which vertically migrating zooplankton may aggregate. In addition, Shark 2 did not consistently experience light levels around LL30, so we also calculated LL40 and LL50 for both tags. LL40 and LL50 are one half and one order of magnitude brighter than LL30, respectively. Previous studies have shown that prey scattering layers, including the primary DSL, can locally align with light levels several orders of magnitude above the reported photobehavioral sensitivity of mesopelagic crustaceans [47–49]. Because the Wildlife Computers light sensor records relative light in a narrow blue wavelength band (415–460 nm) on a logarithmic scale [50], light-level values cannot be converted directly into absolute irradiance, and readings are not comparable between tags.

In addition to isolume depth, we used high-resolution, light-level time series to detect potential bioluminescence events as an in situ proxy for the presence of mesopelagic organisms. Following a modified version of the approach of Braun et al. [31], we applied a series of filters to identify sharp, short-lived increases in light level consistent with bioluminescent flashes rather than ambient light fluctuations. Events were discarded if they occurred in waters at least two orders of magnitude brighter than the tag light sensor’s sensitivity floor, or if the tag had just moved into shallower, brighter water (i.e., > 1 m ascent between time steps). Candidate flashes required an increase of more than two light-level units relative to the preceding two measurements and a peak value that exceeded expected noise based on the following two time steps and on the local standard deviation of the time series (calculated from 15–5 time steps prior). Because these filters eliminate periods where detection is unreliable, the method yields presence-only rather than quantitative estimates. All putative bioluminescence detections were visually checked against previously described flash characteristics for this light sensor to confirm or reject their classification. Because of these conservative quality-control measures, only the light-level data from the 3-second archival dataset was used to detect bioluminescent events.

Vertical habitat use in contrasting oceanographic regimes

Behaviors within the ‘Epipelagic’ cluster occurred primarily on or near the Northeast US continental shelf during summer. Shelf waters in this region support predictable seasonal peaks in copepod abundance (e.g., Calanus spp.) and host a suite of planktivores that track these resources [5,52–54]. Despite being well-documented surface feeders in shelf systems [21], basking sharks in this study also used near-bottom waters, consistent with vertically compressed prey fields generated by shallow bathymetry [55]. Previous autonomous vehicle observations in Scottish shelf systems similarly documented basking sharks spending considerable time within a few meters of the seabed, although no feeding was observed during those short tracking periods [56], which mirrors results from recent satellite tag-based results from this region [44]. When zooplankton vertical migration is constrained by bottom depth, dense aggregations can form near the seafloor, a mechanism also exploited by North Atlantic right whales [57]. Together, these observations indicate that shelf-associated vertical behavior is consistent with exploitation of seasonally concentrated zooplankton resources. However, behavior associated with the epipelagic cluster was present year-round (Fig 1C). Just a few (2−4) individuals remained at high latitudes rather than migrating south during winter/early spring. This behavior suggests that, while most sharks track seasonal productivity and seemingly more hospitable habitats, they can overwinter in cold shelf environments. Basking sharks exhibit regional endothermy [58], which may facilitate tolerance of low temperatures (as low as −0.6ºC; [44]) and allow extended use of high-latitude habitats. Similar behavior has been documented in the Northeast Atlantic, where one of two individuals tagged off the coast of Norway overwintered in the Barents Sea [44] and at least six of 28 individuals tagged off the United Kingdom overwintered within proximal waters [25]. Such variability is also observed in other large pelagic fishes, in which some individuals migrate to distinct overwintering grounds (e.g., common thresher shark [59]) or forego a seasonal migration altogether (e.g., albacore tuna [60]), likely mediated by elevated thermal capacity. In this context, the year-round presence of the epipelagic cluster likely reflects individual variation in movement strategy rather than a breakdown of the broader pattern of seasonal prey tracking.

Once sharks crossed the Gulf Stream, vertical behavior shifted markedly. The ‘Mesopelagic’ cluster was defined by persistent use of 400–1000 m depths and strong DVM (Fig 2). Offshore depth distributions overlapped with depth strata associated with potential mesopelagic prey layers in the Sargasso Sea and adjacent regions [15,61,62]. This shift reflects a transition from spatially constrained shelf foraging to overlap with vertically structured prey fields in oligotrophic offshore waters.

Depth use also varied seasonally within the offshore cluster. Early in migration, Shark 1 occupied depths associated with the deeper, weakly migrating secondary DSL before shifting toward shallower depths later in winter and spring. Previous research indicates peak calanoid copepod abundance in the upper ~500 m of the Sargasso Sea during spring, consistent with the shallower dive behavior observed during that period [61]. Seasonal analyses of backscatter in the North Atlantic similarly show peak optical scattering from mesopelagic organisms is deepest in winter and shoals in spring [63]. However, spatial heterogeneity in mesopelagic layer depth across the basin is also pronounced, suggesting that the observed depth transition could reflect both seasonal vertical migration and spatial movement across distinct hydrographic regimes [63,64]. The seasonal transition in basking shark vertical habitat use in this study suggests flexible use of mesopelagic prey layers across changing oceanographic conditions and dynamic prey resources, and potential tracking of different layers over time.

High-resolution light data from Shark 1 provide independent, presence-only evidence of organismal encounter at mesopelagic depths and into the upper bathypelagic. Numerous short-duration bioluminescent flashes were recorded, including events at depths exceeding 1000 m. Mesopelagic fishes (e.g., myctophids and bristlemouths), cephalopods, and many crustaceans (including copepods) exhibit bioluminescence [65,66], and such signals are common within scattering layers. Although detection probability varies with ambient light conditions and these data do not quantify prey density, they confirm that sharks occupied depth ranges inhabited by bioluminescent mesopelagic organisms and were within ~ 3m of these taxa (i.e., the maximum detection distance of the tag’s light sensor [50]). Importantly, mesopelagic occupancy was observed across individuals through clustering analysis, independent of bioluminescence detections from the two recovered tags.

A proposed ecological link to the secondary deep scattering layer

A consistent feature of the offshore records was repeated use of depths associated with the secondary DSL in the southern Sargasso Sea (~800–900 m), as evidenced in the time series data in Cluster 2 (Fig 2B) and the high-resolution archival data for Sharks 1 and 2 (Fig 4, 5). Scattering layers in this region are vertically structured and taxonomically diverse [15,18,62]. The primary DSL typically occupies ~400–700 m during the day and contains small mesopelagic fishes such as myctophids, along with crustaceans and zooplankton [15,61]. In contrast, deeper and more weakly migrating or non-migratory layers near ~800–900 m are commonly associated with bristlemouths (e.g., Cyclothone) and other small mesopelagic and bathypelagic taxa, such as hatchetfishes and dragonfishes [15,62]. Acoustic classification and net sampling indicate that these deeper layers comprise mixed assemblages of small fishes, crustaceans, gelatinous organisms, and other “fluid-like” scatterers, with relative contributions varying regionally and seasonally [14,15,17,18,62].

Small-bodied fishes such as bristlemouths are abundant within deep, weakly migrating scattering layers. The family’s most prolific genus, Cyclothone, is believed to be the most abundant vertebrate on Earth [67,68]. Although Cyclothone are not confined exclusively to the secondary DSL, they are commonly associated with this layer in the Sargasso Sea [15,69]. These fishes are small (typically ~20–70 mm), weak swimmers, and frequently captured in large numbers in midwater trawls [70–72], suggesting limited evasion capacity relative to larger, more agile mesopelagic fishes. Their body size overlaps with the size range of prey items documented in basking shark stomach contents [21], indicating that small mesopelagic fishes fall within the range that basking sharks are capable of retaining during filtration.

Repeated occupancy of depths corresponding to the secondary DSL is notable because this layer has not traditionally been considered a major foraging target for large vertebrate predators [31], most of which are pursuit feeders that target individual prey items. For most predator species, foraging on small-bodied organisms at 800–900 m imposes substantial energetic and physiological constraints, including thermal limitations in ectothermic sharks [73] and breath-hold constraints in air-breathing divers [74,75]. Thus, the energetic return on individually small prey may not offset the costs of repeated deep diving for most predators.

In contrast, basking sharks combine two traits that may relax these constraints: bulk filter feeding and physiological capacity for sustained deep occupancy. As filter feeders, energetic return depends primarily on prey density and volumetric encounter rate rather than individual prey size [76,77]. Energy intake scales with the product of prey concentration and filtration rate, meaning that even extremely small organisms can contribute substantially to net gain when encountered in sufficiently dense aggregations [78]. At the same time, basking sharks exhibit regional endothermy and substantial thermal inertia [44,58], permitting prolonged residence in deep offshore waters. Few large pelagic vertebrates combine efficient bulk filtration of small prey with the physiological capacity to remain at depth for extended periods. This combination in basking sharks may work synergistically to reduce both encounter and physiological constraints that otherwise limit access to deep, weakly migrating scattering layers for many other predator species.

Dense, multi-taxa assemblages of small-bodied mesopelagic organisms may represent energetically viable resources for basking sharks during offshore residency. The abundance, depth distribution, and limited escape capacity of Cyclothone illustrate how biomass concentrated in deeper scattering layers could be accessible to and profitable for a predator that combines bulk filtration with sustained deep habitat use. The repeated occupancy of these depths that we found, together with the reported composition of scattering layer communities and basking shark physiology and feeding morphology, supports a plausible trophic link between this species and biomass in the deep pelagic ocean. Direct dietary or biochemical evidence from offshore individuals will be required to resolve the extent to which mesopelagic vertebrates contribute to basking shark diet and to close this important gap in our understanding of their trophic ecology.

Dietary Records and Offshore Foraging Potential

Nearly all diet data published on basking sharks derive from surface observations or stranded individuals collected in shelf habitats [21,79,80]. While basking sharks are generally considered zooplanktivores, stomach contents have included fish eggs, pelagic shrimp, and other small crustaceans [21,79]. No confirmed records document consumption of mesopelagic fishes, but this absence likely reflects strong geographic sampling bias toward coastal systems. The only documented deep-water prey ingestion involved pelagic shrimp (Sergestes similis) 40–54 mm in length consumed at>100m depth [21,79], demonstrating that basking sharks are capable of locating, capturing, and ingesting relatively large, mobile prey items at depth. These shrimp are generally considered macroplanktonic and are known to form dense aggregations, suggesting that such feeding may occur within concentrated prey fields. Regardless, because dietary data are almost exclusively derived from shelf-associated individuals, the contribution of offshore mesopelagic prey to overwintering foraging ecology remains unresolved.

Limitations and future work

Although this study provides evidence of overlap with mesopelagic prey resources, several limitations preclude a more definitive interpretation of basking shark foraging ecology. Concurrent hydroacoustic data on scattering layer structure and composition were not available during tag deployments and were therefore inferred from published observations. Previous work has attempted basin-scale alignment of predator dive data with modeled DSL depth [31], but model-derived estimates of the primary DSL vary substantially across regions and seasons and do not provide any inference on secondary DSLs. Accordingly, our inferences rely on clustering of vertical habitat use, regional descriptions of scattering layer structure, and in situ proxies such as isolume depth and bioluminescence detection. Future research integrating multi-sensor biologgers, animal-borne cameras, and surface or in situ acoustic surveys (e.g., [81,82]) would substantially improve understanding of prey community composition and distribution and illuminate predator–prey interactions at depth.