INTEGRATING REMOTE SENSING AND MACHINE LEARNING FOR PREDICTING CLIMATE-INDUCED WATER SCARCITY AND AGRICULTURAL VULNERABILITY IN PAKISTAN

Abstract

Climate-induced water scarcity and agricultural vulnerability pose critical challenges to Pakistan’s food security and rural livelihoods. This study developed an integrated framework combining remote sensing (RS) indicators and machine learning (ML) algorithms to predict water stress and crop vulnerability across Pakistan’s major agro-ecological zones. Historical RS datasets, including NDVI, NDWI, soil moisture, and land surface temperature, were processed and analyzed using Random Forest, Artificial Neural Networks, and Long Short-Term Memory models. Results indicated that soil moisture and NDVI were the most significant predictors of water scarcity and agricultural vulnerability. LSTM models provided superior temporal prediction accuracy (R² = 0.89), while Random Forest effectively identified spatial hotspots of risk. Spatial mapping highlighted southern Sindh and western Balochistan as regions with high vulnerability, while temporal trends revealed increasing water stress over the last decade. The integrated RS-ML framework offers actionable insights for policymakers, provincial authorities, and farmers, enabling proactive water management, climate adaptation planning, and sustainable agriculture in Pakistan.