Clustering and classification of vertical movement profiles for ecological inference of behavior

Vertical movements can expose individuals to rapid changes in physical and trophic environments—for aquatic fauna, dive profiles from biotelemetry data be used quantify categorize vertical movements. Inferences on classes of movement typically rely subjective summaries parameters or statistical clustering techniques that utilize Euclidean matching with observation points. These approaches are prone subjectivity, error, bias. We machine learning a large dataset time series (N = 28,217 dives) for 31 post-nesting leatherback turtles (Dermochelys coriacea). applied dynamic warp (DTW) group (dive) by their metrics (depth duration) into an optimal number clusters. then identified environmental covariates associated each cluster using generalized additive mixed-effects model (GAMM). A convolutional neural network (CNN) model, trained standard shape types the literature, was classify dives within DTW shape. Two clusters were approach—these varied spatial temporal distributions, dependence covariates, sea surface temperature, bathymetry, height anomaly, time-lagged chlorophyll concentrations. CNN classification accuracy five 95%. Subsequent analyses revealed two differed composition shapes, dominated shapes indicative distinct behaviors (pelagic foraging exploration, respectively). The use these allowed discrete data, first second classifying (CNN). Statistical inference found relationships supporting hypotheses niche switching vertically structured behavior. This approach could similarly other animals utilizing dimension movements, including aerial, arboreal, species, efficiently identify different inform habitat models.