An information theory framework for movement path segmentation and analysis

Improved animal tracking technologies provide opportunities for novel segmentation of movement tracks/paths into behavioral activity modes (BAMs) critical to understanding the ecology of individuals and the functioning of ecosystems. Current BAM segmentation includes biological change point analyses and hidden Markov models. Here we use an elemental approach to segmenting tracks into µ‐step-long "base segments" and m-base-segment-long "words". These are respectively clustered into n statistical movement elements (StaMEs) and k "raw" canonical activity modes (CAMs). Once the words are coded using m extracted StaME symbols, those encoded by the same string of symbols, after a rectification processes has been implemented to minimize misassigned words, are identified with particular "rectified" CAM types. The percent of reassignment errors, along with information theory measures, are used to compare the efficiencies of coding both simulated and empirical barn owl data for a selection of parameter values and approaches to clustering.

Methods
The methods developed in this manuscript have been demonstrated on both empirical and simulated relocation data. The former corresponds to an adult female barn owl (Tyto alba) individual, which is part of a population tagged at our study site in the Harod Valley in northeast Israel. The simulated data has been generated using a two‐mode step‐selection kernel simulator called Numerus ANIMOVER_1 (Getz et al. (2023)).

Data processing has been carried out using a series of several machine learning and other algorithms presented in Varun Sethi's GitHub repository Hierarchical-path-segmentation-II.