INTERPRETABLE ML-BASED ANALYSIS OF CLIMATE CHANGE EFFECTS ON LAND SUITABILITY FOR AGRICULTURE IN EURASIA

This article introduces a machine learning (ML) approach to investigating the impact of climate change on the arable land potential of various regions in Eurasia. In order to guarantee that the results are easily comprehensible, researchers implemented potent machine learning methodologies, including Random Forests and Gradient Boosting, in conjunction with SHAP (SHapley Additive exPlanations) values. In order to accomplish this, we integrate data from a variety of geographic sources, which encompasses information on various types of soil, elevations, meteorological conditions (including temperature and precipitation), and historical land use changes. The findings indicate that the feasibility of cultivation in specific regions will be significantly impacted by the application of either the RCP8.5 or RCP4.5 alternative climatic scenario between 2050 and 2100. According to research, the frequency of farming in northern regions, including Russia and Kazakhstan, may increase as a result of climate patterns. However, the value of certain regions in central and southwest Eurasia may be significantly diminished due to a scarcity of water and rising temperatures. The identification of the primary change drivers in this study can be advantageous for sustainable agriculture, land use planning, and climate change adaptation. The interpretable machine learning approach will ensure that the model forecasts are both valuable and accurate for legislators and other stakeholders.