A Spatiotemporal Learning Framework for Anticipating LULC Shifts Under Climate and Anthropogenic Pressures

Anticipating land use and land cover (LULC) changes is essential for sustainable ecosystem management, water resource planning, and climate risk mitigation in rapidly evolving river basins. LULC trajectories emerge from complex interactions among vegetation phenology, climatic variability, topographic constraints, and anthropogenic pressures, rendering accurate forecasting both indispensable and methodologically challenging. This study presents an interpretable, uncertainty-aware predictive framework that integrates multi-source earth observation data with ancillary environmental and socioeconomic variables to capture these spatiotemporal dynamics. By explicitly modelling relationships between vegetation indices, climate drivers, topography, and human activities, the framework identifies key determinants of land conversion while quantifying prediction confidence. Unlike conventional Cellular Automata-Markov (CA-Markov) models constrained by uniform temporal intervals, the proposed approach accommodates irregular observation periods, thereby enabling effective learning from heterogeneous historical datasets. Additionally, the framework addresses a critical limitation of frequency-based models by enhancing prediction accuracy for underrepresented land cover classes. Application to the Godavari Basin, India, demonstrates strong overall predictive performance, with particularly robust results for grasslands, deciduous forests, urban areas, and water bodies. Persistent classification challenges between croplands and grasslands, as well as between evergreen and deciduous forests, underscore the subtle nature of these transitions and highlight the need for enhanced phenological characterization. Beyond predictive accuracy, the framework provides interpretability through its probabilistic architecture, offering valuable insights into how climatic and anthropogenic factors jointly influence land transformation pathways. The results deliver not only spatially explicit LULC projections for future scenarios but also a practical decision-support tool for policymakers and resource managers navigating the tension between development imperatives and ecosystem sustainability. Critically, such predictive capabilities are vital for anticipating compound hydroclimatic extremes where land-atmosphere feedbacks amplify drought, flood, and heatwave risks and for informing climate adaptation strategies in dynamic river basins and other climatically vulnerable regions.