Soil organic carbon mapping in India’s black soil region through machine learning

Abstract

Soil organic carbon (SOC) is a critical component of soil health, influencing productivity, ecosystem services, and carbon sequestration. This study applied Quantile Random Forest (QRF) modeling to predict SOC distribution across depth intervals (0-200 cm) using 88 soil profiles from India’s Black Soil Region (BSR). Observed, splined, and predicted SOC values were compared to evaluate prediction accuracy and spatial variability. Results indicate low SOC levels (mean = 0.5%) consistent with semi-arid agroecosystems, with highest concentrations in surface soils (0-15 cm) and a decline with depth. Prediction performance improved with depth (R² = 0.48 at 0-5 cm; 0.62 at 100-200 cm), reflecting greater stability of SOC in subsoils. Vegetation, slope, rainfall, and temperature emerged as key predictors of SOC distribution. Spatial mapping highlighted higher SOC in forested and hilly zones, and lower SOC in cultivated and degraded lands. Uncertainty analysis revealed greater variability in surface layers than subsoils. These findings provide insights into SOC dynamics in black soils and demonstrate the utility of machine learning for depth-resolved SOC prediction, supporting sustainable land management and carbon conservation.