Predicting and Mapping the Soil Organic Carbon Using Random Forest Model in Natural Forest Ecosystem of Central India

This is the authors' abstract. We don't add key points for this paper.

In the Pench Tiger Reserve (PTR) of central India, a study aimed to predict soil organic carbonI (SOC) levels to guide the eco-restoration of degraded sites impacted by human activities. Using a grid sampling method (100 mtr), 217 surface soil samples (0–30 cm depth) were collected across various terrains. The study employed a Random Forest Model (RFM) for predictive SOC mapping, incorporating environmental covariates derived from sampled points to capture non-linear relationships, which is an advanced rule-based method used at different scales. High-resolution digital elevation data from ALOS provided critical terrain attributes, including elevation, slope, aspect, multi-resolution index of valley bottom flatness (MRVBF), topographical wetness index (TWI), and channel network base level (CNBL). Vegetative factors such as the Normalized Difference Vegetation Index (NDVI) and the Soil-Adjusted Vegetation Index (SAVI) were obtained from Sentinel-2A satellite imagery. The model’s predictions for soil organic carbon (SOC) ranged from 0.65% to 4.4%. Its effectiveness was validated with a coefficient of determination (R2) of 73.5% and a concordance coefficient of 74%. The analysis highlighted Sentinel-2A bands—particularly bands 3 and 4—along with vegetation indices such as SAVI and NDVI as the most influential predictors for SOC mapping. These findings suggest that this model is well suited for assessing soil organic carbon (SOC) in the undulating terrains of degraded and natural forest areas.