Behaviour identification and time-activity budget estimation of the threatened little bustard using accelerometry

Abstract

Understanding animal ecology requires knowing what animals do. GPS devices provide detailed insights into space use, but they lack behavioural information until recent advancements in accelerometry have bridged this gap. In this study, machine learning methods were applied to accelerometry data to identify and classify behaviours in little bustard, Tetrax tetrax, a species whose population is declining because of habitat loss and degradation. Using recordings of four captive individuals, models were fitted to classify key behaviours, standing, lying, vigilance, locomotion, foraging and male courtship, with separate models for each sex because of behavioural differences. In addition, different sampling frequencies, balancing methods and data-splitting approaches were tested to examine interindividual variation and the effect of sample size. Results revealed that models built with data sampled at 10 Hz performed similarly to those sampled at 20 Hz. Male models slightly outperformed female models, achieving precision and sensitivity exceeding 0.87. Male-specific behaviours, such as courtship, attained F1-scores above 0.8. The application of the models to 10 free-ranging individuals showed marked seasonal and sexual differences in time—activity budgets. Males changed their behaviour seasonally, devoting more time to vigilance, locomotion and courtship during the breeding season and to foraging in winter. On the contrary, females showed a more consistent behaviour pattern year-round, predominantly resting, although lying increased during the breeding season, likely reflecting incubation. These findings indicate the potential application of machine learning and accelerometry to monitor behaviours in freeranging little bustards, offering a valuable tool to understand activity patterns and develop conservation strategies for this threatened species