Novel statistical methods for inferring human impacts on animal movement and migration from large-scale datasets

Masolele, Majaliwa M. (2025) Novel statistical methods for inferring human impacts on animal movement and migration from large-scale datasets. PhD thesis, University of Glasgow.

Abstract

Multiple stressors contribute to the decline of numerous animal species within and outside protected areas worldwide. While our understanding of anthropogenic habitat loss and degradation, climate change, and anthropogenic pressures as potential drivers of these declines is improving, we still lack a mechanistic understanding of how their finescale effects translate into animal movement decisions and how these decisions ultimately influence survival and, in turn, population dynamics at a broad scale. Unravelling these patterns requires associations of spatial covariate fields and fine-scale movement data, typically collected using Global Positioning System (GPS) tags deployed on animals. These tags provide a bivariate time series of coordinates at defined intervals, facilitating insights into how animals move, where and when they forage, and the nature of both intra- and interspecific interactions.

Despite the availability of such movement data alongside the corresponding environmental data, significant analytical challenges persist. Habitat selection models, particularly resource selection functions (RSFs) and step selection functions (SSFs), represent a fundamental tool to identify the characteristics of suitable habitats for animals at both broad and fine scales. The core concepts underlying these methods are based on the ratio between habitat availability and habitat use by the animal. However, while these models enhance our understanding of habitat suitability, they often yield divergent conclusions even when applied to the same datasets, likely due to differences in their spatial and temporal scales of operation. A pressing question, therefore, is how parameters derived from fine-scale movement models can be reconciled to produce the patterns similar to those from broad-scale models and thereby improving our understanding of how animals’ use of space relates to the distribution of resources, risks, and environmental conditions. Addressing this challenge requires a modelling framework that enables parameter scalability, quantifies uncertainty, and remains computationally efficient while capturing the influence of spatial covariate fields, such as human-made infrastructure.

The objective of this thesis is to advance our understanding of how animal space use relates to the distribution of resources, risks, and environmental conditions by integrating and developing state-of-the-art multiscale statistical methods within a Bayesian framework while maintaining computational efficiency. This will enhance our ability to assess how animals respond to changing landscapes and climate conditions, predict future spatial distributions based on current patterns, identify the key drivers that displace or restrict animals from otherwise suitable habitats, and pinpoint critical habitats that should be preserved from human alteration. Throughout this thesis, I will focus on models of animal movement, particularly habitat selection models, and contribute to expanding the array of statistical methods available for analysing movement data. An overarching goal of the thesis is to develop methods that can be applied to the study of the Serengeti wildebeest migration, a vital ecological process in one of the most biodiverse ecosystems on earth. I will begin by reviewing existing and widely used methods in the literature. Subsequently, I introduce a multiscale step selection model that facilitates the estimation of long-term animal space use without requiring simulations from the fitted model, and I will leverage variational inference within a Bayesian framework to estimate selection and avoidance parameters from movement observations and environmental data while demonstrating the importance of formally quantifying uncertainty in these estimates.

The focus then shifts to examining the effects of anthropogenic structures, such as buildings, on the spatial distribution of migratory wildebeest using multiscale inference from the previous chapter. This analysis will provide insight into whether wildebeest select or avoid areas near buildings and how these selection patterns influence their space use at the population level within the ecosystem. These findings will be essential for a later chapter, where I simulate how wildebeest space use is expected to change in response to the introduction of new additional buildings in the ecosystem.

In Chapter 5, I use hierarchical sparse Gaussian processes to estimate the mean migration routes of the Serengeti wildebeest population. These modelled routes form the basis for improving spatial predictions of where wildebeest are likely to spend most of their time during critical life-history stages such as calving, weaning, rutting, or migration. This is achieved by integrating wildebeest space use patterns derived from local environmental features such as anthropogenic structures, as detailed in Chapter 4 with the population mean migration routes inferred here. The latter are used as a proxy for the influence of long-term spatial memory on movement decisions. This integrative modelling framework offers a more ecologically grounded understanding of wildebeest spatial distribution across specific days of the year and during key life-history events.

In Chapter 6, I will develop a novel simulation approach to model the placement of buildings in different scenarios and explore the impact of different allocation strategies on wildebeest space use. This will be achieved by simulating hypothetical building distributions using a nonlinear preferential attachment rule to place buildings at specific locations and incorporating an accept-reject mechanism to increase and decrease building clustering. Then I will estimate the new patterns of wildebeest space use using the methodology introduced in chapter 4 and quantify the shift from observed space use by employing the Kullback-Leibler divergence.

This thesis demonstrates that multiscale animal movement models provide valuable insights into how animal space use is shaped by the distribution of resources and risks in changing landscapes. A key finding is that considerable uncertainty can persist even in large telemetry datasets, underscoring the importance of quantifying uncertainty in resource selection analyses. The study on the spatial distribution of migratory wildebeest reveals that while these animals tend to avoid areas near anthropogenic structures, this behavior does not lead to complete exclusion. Instead, it results in a reduced duration of time spent in the vicinity of such structures. Furthermore, the study incorporating local environmental responses with long-term spatial memory effects reveals that spatial predictions of wildebeest distribution during key life-history stages, such as calving, are improved by reducing uncertainty about where populations are most likely to spend time on specific days or during particular events. Finally, a simulation study indicates that the impact on wildebeest space use is more pronounced when new developments occur in previously undeveloped regions or in isolation from existing infrastructure, highlighting the importance of strategic spatial planning in conservation efforts.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Additional Information: Supported by funding from the Engineering and Physical Sciences Research Council (EPSRC) scholarship.
• Subjects: Q Science > QA Mathematics
• Colleges/Schools: College of Science and Engineering > School of Mathematics and Statistics
• Funder's Name: Engineering and Physical Sciences Research Council (EPSRC)
• Supervisor's Name: Torney, Professor Colin and Hopcraft, Professor Grant
• Date of Award: 2025
• Depositing User: Theses Team
• Unique ID: glathesis:2025-85554
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 30 Oct 2025 12:14
• Last Modified: 04 Nov 2025 14:14
• Thesis DOI: 10.5525/gla.thesis.85554
• URI: https://theses.gla.ac.uk/id/eprint/85554

Related URLs

• Article DOI