Olivier Devineau, Alina L. Evans, Andrea Friebe, Martin Mayer, Boris Fuchs, Alexandra Thiel

Brown bear (Ursus arctos) females are known to reduce their movement when accompanied by dependent offspring. In this study, we used GPS data from 76 sexually mature (age > 4 years) Scandinavian brown bear females (N=155 years of data) between 2006 and 2023 to identify movement patterns associated with reproductive status: solitary vs. with cubs of the year (COY). To classify reproductive status, we trained a Random Forest model (RF) on data obtained from solitary females (N=93) and females with COY (N=62) whose status had been determined based on aerial surveys. We used two movement metrics derived from GPS data (mean daily step length and home range) and two activity metrics derived from bi-axial accelerometer data (mean daily activity and the proportion of passive state). Based on the knowledge that a female returns to solitary behavior within a few days after losing her entire litter, we estimated the timing of 17 litter loss events that occurred during the active season (from April 10th to July 31st) with two machine learning algorithms (the previous Random Forest model (RF) and a Hidden Markov Model (HMM)) detecting shifts in reproductive status based on the four previous metrics.
The HMM performed better than the RF, with an overall accuracy in predicting the reproductive status of a female on a given day of 92% and 86% respectively. The accuracy of both methods was highest at the beginning of the active season and decreased from May to July as the movement of the two reproductive statuses became less distinguishable, probably due to the combination of an increased offspring mobility and a decrease in roaming behavior at the end of the mating season.
These algorithms allow for remotely detectable and accurate estimates of reproductive status, and a higher temporal resolution of litter loss events compared to aerial surveys. They can be particularly useful for individuals that have evaded annual monitoring and when cub losses occur in early spring.