A diverse group of toothed whale species inhabits the pelagic habitats of the western North Atlantic, competing for overlapping prey resources. Historical data deficits have limited fundamental research into many of these species, such as establishing baselines of distribution and abundance, so their occurrence and habitat use patterns are not well characterized. Periodic cycles in activity have been reported at a range of temporal scales for odontocetes in other regions, such as seasonal movements, foraging activity modulated by lunar cycles, and diel activity patterns. A variety of spatial, temporal, and behavioral separation strategies have also been observed among predator guilds in both marine and terrestrial systems, and these may also contribute to ob served spatiotemporal patterns in activity. Re cently, passive acoustic data has been applied to monitor odontocete species continuously, with im proved detection and species discrimination for some cryptic species. We used a long-term passive acoustic data set collected at sites spanning the western North Atlantic shelf-break region to quantify presence and characterize seasonal, lunar, and diel activity patterns for 10 species. Our re sults demonstrated strong regional preferences and clear patterns of spatiotemporal separation between species with similar foraging ecology. Latitudinal shifts in seasonal presence peaks may suggest meridional seasonal migrations for some dolphin species. We also observed strong diel activity patterns that were modulated by both seasonal and lunar cycles. This study reveals complex behavioral patterns arising in response to natural cycles playing out over multiple temporal scales and provides new in sights into habitat partitioning among toothed whale species.

Time series of labeled odontocete echolocation clicks were derived by (1) from a large passive acoustic data set collected through repeated mooring deployments. Clicks were detected and classified to species using a machine learning workflow, and then classification error was quantified by manual verification of a subset of the labeled data. For each species/group we binned the time series of labeled clicks into 5-minute time bins, then scaled the number of clicks per bin by recording effort as well as the classifier error rates which were calculated on a per-species per-deployment basis. For analysis of temporal patterns in species presence and activity, we considered binomial presence/absence in each 5-minute bin to be a more reliable metric of species presence than the actual number of clicks labeled to that species, since some clicks were isolated by the clustering algorithm and therefore were unavailable to be labeled by the classifier. To remove spurious presence bins based on very few detections, we set a minimum number of clicks per bin threshold to be considered “presence”: ≥50 clicks per 5-minute bin for delphinid species, and ≥20 clicks per 5-minute bin for beaked whales, sperm whales, and Kogia spp. These values were selected based on consideration of the click-production rates and group sizes of these species, but are essentially arbitrary. 

Odontocete species acoustic presence was detected and classified at the 11 western North Atlantic passive acoustic monitoring sites using a combination of automated methods and manual review, as detailed in (1) and (2). Each of these .csv files contains information on the acoustic presence of a single odontocete species/group, given in the file name, at each of the acoustic monitoring sites (listed above in this README file). The temporal granularity is 5-minute bins. The first column of each file gives the time stamps of all 5-minute bins spanning the study period (1 May, 2016 to 30 April, 2019). Subsequent columns correspond to species' acoustic presence at each of the acoustic monitoring sites, as indicated by the site name abbreviation in the column header, ordered from north to south. The values in each cell indicate the number of echolocation clicks labeled as the given species in a particular time bin at a particular site. These values have been scaled by both effort and classifier species- and deployment-specific false positive rates, often resulting in fractional click counts. Cells containing "NaN" correspond to periods of no effort, such as when there were gap between successive recording device deployments at a given site.