Soil organic carbon modeling in cropland under several climatic scenarios using machine learning in western India

This study investigated how shared socioeconomic pathways (SSPs) scenarios affect farmland soil organic carbon (SOC). The effects of historical precipitation, temperature, soil type, soil features, land shape, farming methods, and land use changes on SOC trends from 1982 to 2024 were examined in Karvir area Maharashtra, India. Predictions used machine learning modelers Random Forest (RF), Extreme Gradient Boosting (XGB) and Support Vector Regression (SVR). Coefficient of determination (R²), mean error (ME), and root mean square error (RMSE) were optimal for XGB at 0.998, 0.000, and 0.225 g/kg, respectively. Using remote sensing data processed in Google Earth Engine (GEE) and Geographic Information System (GIS), the model was validated by a certified local soil testing laboratory, achieving an R² of 0.88, an RMSE of 1.9 g/kg, and an ME of 1.5 g/kg. Temperature and rainfall under high SSPs drastically reduced projected SOC. In the worst case (SSP5-8.5), the mean SOC in 2040 declined from 46.4 to 24.2 g/kg in 2100. Since 2018, conservation agricultural technologies like no-till farming, mulching, and composting have boosted SOC while farmland has decreased over 42 years. These findings give a paradigm for climate-adaptive management alternatives that promote SOC sequestration, soil health, and sustainable agriculture.