INTERPRETABLE MACHINE LEARNING FOR PREDICTING CLIMATE CHANGE EFFECTS ON AGRICULTURAL LAND SUITABILITY IN EURASIA

Abstract

Climate change poses significant challenges to global food security, particularly in regions with large agricultural landscapes such as Eurasia. This study presents an interpretable machine learning-based framework to evaluate and predict the impact of climatic variations on agricultural land suitability across Eurasia. A comprehensive dataset consisting of climate parameters, soil characteristics, topographic features, and historical crop productivity records was used for model training and evaluation. Machine learning algorithms were applied to assess current land suitability and project future scenarios influenced by rising temperatures, altered precipitation patterns, and soil degradation. To ensure transparency and trust in decision-making, explain ability techniques such as SHAP and feature importance analysis were incorporated, revealing the most influential environmental factors affecting suitability outcomes. The results demonstrate noticeable spatial shifts in suitable agricultural zones, with certain regions experiencing reduced productivity potential under future climate projections. This work emphasizes the importance of data-driven insights for policymakers, enabling climate-resilient agricultural planning, sustainable land management, and effective adaptation strategies in Eurasia.