Interpretable machine learning for predicting climate change effect on agricultural land suitability in Eurasia

Abstract

Climate change is rapidly transforming temperature and precipitation patterns, posing serious challenges to agricultural sustainability across Eurasia. Accurately predicting changes in agricultural land suitability is essential for effective long-term planning. Conventional statistical models often fail to provide both high accuracy and interpretability. This project presents an interpretable machine learning approach to assess climate change impacts on agricultural land suitability. Multiple climate indicators, soil properties, and land-use parameters are used as input features. Machine learning models such as Random Forest and Gradient Boosting are developed for prediction. Model performance is evaluated using standard accuracy and error metrics. To enhance transparency, interpretability techniques such as SHAP are applied. These methods reveal the relative importance of climatic and environmental factors. Spatial analysis is conducted to identify regional variations in suitability changes. The model highlights vulnerable and resilient agricultural zones across Eurasia. The findings support explainable, data-driven agriculture.