TIES 2024

TIES 2024

Temporal Embeddings for Animal Movement Trajectory Interpolation

Conference

TIES 2024

Format: IPS Abstract

Keywords: ecology, machine learning, missing not at random, random forest

Session: Invited Session 11B - Predictive Modeling of Complex Environmental Data in Agriculture and Ecology

Thursday 5 December 3:30 p.m. - 5 p.m. (Australia/Adelaide)

Abstract

In ecology, a selection analysis compares the features at an observed location of the animal to features at unobserved locations to determine if there is non-random selection of values of those features. The desert bighorn sheep (DBS) bed down during the day to rest. When it is hot, they may seek refuge in higher elevations (more rugged terrain) including caves. The GPS devices cannot communicate with the satellites leading to a "missed fix". Ignoring the missed fixes may lead to bias and reduction in data size necessary for fitting the selection analysis. In order to interpolate missed fixes, we use temporal lag and lead embeddings in a random forest model to quickly identify diurnal patterns in DBS movement for downstream analysis.