Efficient abstracting of dive profiles using a broken‐stick model

Abstract

Summary For diving animals, animal‐borne sensors are used to collect time–depth information for studying behaviour, ranging patterns and foraging ecology. Often, this information needs to be compressed for storage or transmission. Widely used devices called conductivity‐temperature‐depth satellite relay data loggers (CTD‐SRDLs) sample time and depth at high resolution during a dive and then abstract the time–depth trajectory using a broken‐stick model (BSM). This approximation method can summarize efficiently the curvilinear shape of a dive, using a piecewise linear shape with a small, fixed number of vertices, or break points. We present the process of abstracting dives using the BSM and quantify its performance, by measuring the uncertainty associated with the profiles it produces. We develop a method for obtaining a confidence zone and an index for the goodness‐of‐fit (dive zone index, DZI) for abstracted dive profiles. We validate our results with a case study using dives from elephant seals (Mirounga spp.). We use generalized additive models (GAMs) to determine whether the DZI can be used as a proxy for an absolute measure of fit and investigate the relationship between the DZI and the dive shape. We found a strong correlation between the residual sum of squares (RSS) for the difference between the detailed and abstracted profiles, and the DZI and maximum residual (R4), for dives resulting from CTD‐SRDLs (69% deviance explained). On its own, the DZI explained a lower percentage of deviance which was variable for abstracted dives with different numbers of break points. We also found evidence for systematic differences in the DZI for different dive shapes (65% deviance explained). Although the proportional loss of information in the abstraction of time–depth dive profiles by BSM is high, what remains is sufficient to infer goodness‐of‐fit of the abstracted profile by reversing the abstraction process. Our results suggest that together the DZI and R4 can be used as a proxy for the RSS, and we present the method for obtaining these metrics for BSM‐abstracted profiles.