Machine Learning Assessment of Climate-Driven Shifts in Wildlife Population Dynamics Across Fragmented Landscapes

Abstract

The human factors that have a significant influence on wildlife population dynamics include climate
change and habitat fragmentation, which impact the ways species are distributed, abundant, and
resilient. The traditional ecological models do not usually exhibit good predictability, usually do not
account for the nonlinear interactions of climate variables and landscape structure. In this paper, a
machine learning-based model is used to evaluate climate-induced changes in wildlife populations in
fragmented natural environments, determine the most important environmental and spatial predictors,
and measure the predictive accuracy. Data on wildlife population, climate (change in temperature,
variability of precipitation), and fragmentation (patch size, edge density, habitat isolation, and
connectivity) were integrated and pre-processed. Models of ensemble machine learning, such as
Random Forest, Gradient Boosting, and Neural Networks, were trained and cross-validated with
cross-validation. The evaluation of model performance was done using MAE, RMSE, and R2, and
these features of importance and partial dependence analysis established prominent drivers of climate
and fragmentation. Higher vulnerability when compared to moderately fragmented (0.98 ± 0.14) and
low-fragmentation areas (1.12 ± 0.08) was observed with populations in highly fragmented habitats
having the lowest mean population index (0.74 ± 0.22). Gradient Boosting was the most predictive
(RMSE=0.168, MAE=0.119, R 2=0.81). The importance of features showed that temperature anomaly
(0.31) and patch size (0.27) are the most influential factors, and precipitation variability, habitat
isolation, and edge density were also significant. Partial dependence plots revealed that there were
nonlinear population changes in response to temperature changes, but these were sensitive to extreme
weather conditions. Machine learning has the potential to learn complex interactions between climate
and fragmentation, and is capable of identifying the major drivers of wildlife populations. The
framework gives practical information to conservation planning to assist in restoring habitat
connectivity and climate-adaptive management of fragmented landscapes.