A hierarchical machine learning framework for the analysis of large scale animal movement data

Abstract

Background In recent years the field of movement ecology has been revolutionized by our ability to collect high-accuracy, fine scale telemetry data from individual animals and groups. This growth in our data collection capacity has led to the development of statistical techniques that integrate telemetry data with random walk models to infer key parameters of the movement dynamics. While much progress has been made in the use of these models, several challenges remain. Notably robust and scalable methods are required for quantifying parameter uncertainty, coping with intermittent location fixes, and analysing the very large volumes of data being generated. Methods In this work we implement a novel approach to movement modelling through the use of multilevel Gaussian processes. The hierarchical structure of the method enables the inference of continuous latent behavioural states underlying movement processes. For efficient inference on large data sets, we approximate the full likelihood using trajectory segmentation and sample from posterior distributions using gradient-based Markov chain Monte Carlo methods. Results While formally equivalent to many continuous-time movement models, our Gaussian process approach provides flexible, powerful models that can detect multiscale patterns and trends in movement trajectory data. We illustrate a further advantage to our approach in that inference can be performed using highly efficient, GPU-accelerated machine learning libraries. Conclusions Multilevel Gaussian process models offer efficient inference for large-volume movement data sets, along with the fitting of complex flexible models. Applications of this approach include inferring the mean location of a migration route and quantifying significant changes, detecting diurnal activity patterns, or identifying the onset of directed persistent movements.

Feedback loops between 3D vegetation structure and ecological functions of animals

“…Although GPS locations in themselves cannot capture many important aspects of animal behavior that might affect vegetation structure, machine learning can be used to infer behavioral states such as foraging or dispersing based on observed distributions of step lengths and turning angles, and where available, body orientation and acceleration. (Nathan et al2012;Torney et al 2021;Yu et al 2021). Hidden Markov Models, for example, estimate unobserved behavioral states using common metrics from GPS or accelerometer data (McClintock et al 2020;Klarevas-Irby et al 2021).…”

Inferring spatially varying animal movement characteristics using a hierarchical continuous‐time velocity model

“…Examining the landscape level drivers of movement has typically employed parametric functions of environmental covariates via hidden Markov models (Langrock et al, 2012 ) or step selection functions (Avgar et al, 2016 ; Thurfjell et al, 2014 ). Recently, non‐parametric approaches have been proposed that allow continuous and time‐varying movement parameters to be incorporated into models (Michelot et al, 2021 ; Torney et al, 2021 ). An important next step in the development of these models is to identify the underlying spatially varying factors that influence animal movement.…”

“…We employ a velocity‐based movement model which is linked to an underlying latent spatial field via the introduction of a novel non‐stationary covariance matrix. While previous works (Hooten & Johnson, 2017 ; Torney et al, 2021 ) have also applied hierarchical GPs to animal movement, the novelty of our approach is that by linking the velocity‐based movement model to latent spatial fields we are able to reveal the persistent, spatially varying movement characteristics of mobile animal populations based on positional data collected from multiple individuals. To enable efficient inference of data sets consisting of potentially millions of data points, we employ Bayesian variational learning (Blei et al, 2017 ) a novel approach in this context that replaces traditional computationally expensive sampling‐based inference with fast stochastic optimisation.…”

Classifying the activity states of small vertebrates using automated VHF telemetry

“…The behaviour of an animal can be fundamentally divided into active and passive (Halle & Stenseth, 2000), with the former requiring a much higher energy expenditure (Rowcliffe et al, 2014). Quantifying the distribution of activity periods throughout the day provides important insights into species’ responses to their environment, foraging strategies, bioenergetics and adaptations (Aschoff, 1966; Torney et al, 2021). Moreover, knowledge of the partitioning of sympatric species along the temporal niche axis can yield insights into the mechanisms that facilitate stable coexistence (Nakabayashi et al, 2021).…”