Interpretable Machine Learning for Sustainable Land use: Analyzing Climate Change Impacts on Agriculture in Eurasia

Climate change has been a significant threat to the sustainability and productiveness of agricultural systems in Eurasia necessitating appropriate analytical systems to establish the suitability of land. The study applies interpretable machine learning (IML) algorithms to evaluate how the change in climate variables has influenced the suitability of agricultural land, and in this case, this has been taken as temperature, precipitation, soil moisture and elevation. The inputs to be used in the ensemble model are a multi-source geospatial dataset, which involves WorldClim climate projections (2020-2100), FAO soil data, and a multi-source model MODIS vegetation index. The model is very predictive (R2 = 0.89) meaning that there was great uniformity between the recorded and the anticipated suitability indices in various agro ecological areas. The analysis performed by using SHapley Additive exPlanations (SHAP) model interpretation revealed that model most common factors used in depleting the suitability were temperature anomalies and the variability of soil moisture particularly in Central Asian and Eastern Europe. Conversely, latitudes that are northward are gradually becoming livable in the moderately warmer conditions and this is a sign of progressive poleward shift on habitable agricultural latitudes. The spatially explicit results will be relevant to the policy makers and land-use planners to know how it is necessary to adapt to the changes in the crop zoning, and climate-resilient agricultural policies. Overall, interpretable machine learning integration offers an evidence-based and transparent approach to understanding and minimizing the impact of climate change on the suitability of agricultural lands in Eurasia that can be utilized in the planning of evidence-based sustainability.