Assessing biodiversity loss due to environmental changes using artificial intelligence techniques

Chia, Cheng Gun (2025) Assessing biodiversity loss due to environmental changes using artificial intelligence techniques. Final Year Project, UTAR.

Preview

PDF Download (4Mb) | Preview

Abstract

This study explores the potential of artificial intelligence (AI) techniques to enhance species distribution modelling (SDM) for assessing biodiversity loss due to environmental change, focusing on Strigiformes (owls) in Malaysia, which remain understudied and vulnerable to climate threats. Traditional SDM methods often struggle to capture complex ecological interactions because they rely on linear assumptions. There is a lack of comprehensive studies focused on predicting the future distribution of these species under varying environmental scenarios in Malaysia.To address these issues, this study proposes the use of machine learning and deep learning models, specifically Random Forests (RF) and Multi-Layer Perceptrons (MLP), complemented by Explainable AI (XAI) techniques, to improve predictive accuracy, robustness, and interpretability of SDMs. The models were developed with eight key environmental variables which are annual mean temperature, mean diurnal range, isothermality, annual precipitation, precipitation of wettest month, primary forest, secondary forest and urban area cover for the genus Ketupa (a genus of Strigiformes) in Malaysia. Data splitting techniques, including random and spatial block were evaluated to address spatial autocorrelation and improve model generalization. Spatial block sampling demonstrated superior performance, with smaller performance gaps in Area Under the Receiver Operating Characteristic curve (AUROC) and Area Under the Precision Recall curve (AUCPR) when tested on East Malaysia independent dataset, confirming its robustness for extrapolation. Environmental analysis identified urban area cover as the most influential predictor of habitat suitability, followed by annual precipitation. Response curve analysis revealed critical environmental thresholds that align with Ketupa’s ecological preferences for tropical lowland and wetland habitats. Habitat suitability mapping under future climate and land-use scenarios indicates a potential loss of high-quality habitat and a flattening of suitability gradients.

Actions (login required)