Bayesian Spatio-Temporal AI for Modelling and Forecasting Animal Movement Behaviour and Identifying Its Drivers

This project seeks research partners to develop Bayesian spatio-temporal artificial intelligence models for modelling and forecasting animal movement behaviour and identifying its ecological and environmental drivers. It will integrate animal tracking records with Earth observation, climate, habitat and human activity data to understand movement patterns, predict behavioural responses to changing conditions and support wildlife conservation.

Animal movement reflects interactions between individual characteristics, resource availability, landscape structure, weather, seasonal cycles and human disturbance. Tracking datasets provide detailed observations, but irregular sampling, measurement errors, missing records and differences between individuals complicate analysis. The project will combine Bayesian statistical modelling with AI to capture these relationships while explicitly representing uncertainty.

The central research question is how Bayesian spatio-temporal AI can improve forecasts of animal movement while providing interpretable evidence about the factors associated with behavioural changes.

The objectives are to:

1. Establish a reproducible framework for integrating animal tracking and environmental datasets.

2. Develop Bayesian and hybrid AI models for movement trajectories, speed, direction and behavioural states.

3. Identify and quantify associations between movement behaviour and ecological, climatic and human-related factors.

4. Assess forecast uncertainty, model robustness and transferability across individuals, seasons and locations.

5. Translate validated findings into tools for habitat management, connectivity assessment and conservation planning.

The research will begin with data quality assessment, including tracking accuracy, sampling intervals, missing observations and spatial and temporal alignment. Depending on availability, GPS and biologging records will be linked to vegetation indices, land cover, rainfall, temperature, water availability, terrain and indicators of human disturbance. Data access agreements and safeguards will protect sensitive species locations.

Model development will compare established movement models with Bayesian hierarchical, state-space and spatio-temporal approaches, alongside suitable machine learning and deep learning methods. Hybrid architectures will investigate how AI can capture complex nonlinear patterns within a probabilistic framework. Hierarchical structures will account for variation among individuals, populations and locations, while latent-state models will examine behaviours such as resting, foraging and travelling. Behavioural labels will be validated against independent observations where available.

The analysis will distinguish forecasts from explanations of movement drivers. Posterior estimates, interpretable model components and sensitivity analyses will assess relationships with environmental conditions. Associations will be interpreted cautiously, recognising confounding and the limits of observational data. Where data and study design permit, mechanistic hypotheses about resource selection, seasonal movement and responses to disturbance will be tested.

Validation will use chronological testing, rolling-origin forecasting and holdout evaluations across individuals and locations. Performance will be assessed through trajectory and displacement errors, behavioural classification where relevant, probabilistic forecast scores and prediction interval calibration. Posterior predictive checks will assess model adequacy. Robustness tests will examine irregular sampling, missing inputs and environmental conditions outside the training range. Comparisons with simpler benchmarks will establish whether additional model complexity improves forecasting.

The cooperation will bring together expertise in Bayesian statistics, mathematical modelling, AI, movement ecology, remote sensing and conservation. South Eastern Kenya University could contribute Bayesian and hybrid forecasting methods, uncertainty quantification and postgraduate training. Partners could contribute animal tracking datasets, ecological expertise, Earth observation processing, field validation and conservation applications. Target species, study areas, responsibilities and data access will be agreed jointly.

Expected outputs include reproducible analytical workflows, validated movement forecasting models, estimates of movement-related environmental associations, joint publications and a prototype tool for exploring forecasts and uncertainty. Collaboration will also support joint postgraduate supervision, researcher exchanges and capacity building.

The project aims to strengthen understanding of animal responses to environmental change and provide credible evidence for protecting habitats, maintaining movement corridors and managing human–wildlife interactions. Contact me if interested: rmutwiri@seku.ac.ke