Modeling elephant migration and deforestation hotspots in Sri Lanka’s dry forests using hybrid CNN-LSTM architectures and cellular automata

Abstract

Sri Lanka’s dry forests, spanning approximately 15,500 km2, represent a biodiverse yet critically threatened ecosystem in South Asia, facing escalating pressures from climate change, agricultural expansion, and human-wildlife conflict. This study develops an integrated machine learning (ML) and remote sensing framework for spatiotemporal change detection (2015–2025) to address these challenges, leveraging Sentinel-2 multispectral imagery. This study proposes a hybrid 3DCNN-LSTM (Convolutional Neural Network–Long Short-Term Memory) model that integrates spectral-temporal feature fusion to achieve monsoon-resilient land cover classification, attaining 92.4% accuracy (K=0.89)—a 7.2% improvement over conventional SVM-PCA methods (85.2%, K=0.81). Key innovations include (1) a Genetic Algorithm-optimized Degradation Risk Index (GA-DRI) incorporating 12 ecological variables, and (2) Cellular Automata (CA) modeling of elephant migration under RCP4.5 climate scenarios. Results identify six deforestation hotspots (> 5km2 each) in Anuradhapura District, strongly correlated with agricultural encroachment (r = 0.78, p < 0.01) and declining groundwater tables (r = −0.65). This framework supports Sri Lanka’s National Adaptation Plan (2022–2030) and advances progress toward UN Sustainable Development Goal 15 (Life on Land) through actionable, high-resolution conservation metrics.