DISTRIBUTION OF SOIL INORGANIC CARBON IN THE CONTIGUOUS UNITED STATES

Soil inorganic carbon (SIC) influences soil infiltration, sequesters carbon, and affects global climate models. Despite its significance, research on SIC formation, mapping, and modeling has been limited. Unlike soil organic carbon, SIC typically accumulates in the subsoil. This and the considerable spatial variability pose challenges to accurately mapping and modeling data through interpolation, as uncertainty amplifies with distance from sampled areas. This research leverages alternative land surface parameters, paired with satellite imagery and machine learning methods (Random Forest Regressor (RFR) and Random Forest Classifier (RFC) models), to reduce uncertainty in unsampled regions compared to traditional interpolation methods. Using this method, we generated a SIC distribution map across the Contiguous United States (CONUS). The results indicate the moderate effectiveness of the RFR model with an RMSE of 3.97, and MAE of 3.1. The RFC model classified SIC into three classes with an overall accuracy of 0.56 with the lowest underestimation and overprediction belonging to SIC≥126 g C/kg soil. An evaluation of feature importance confirmed that soil pH is a primary determinant influencing SIC accumulation. Moreover, we identify a novel precipitation threshold of 1700 mm for SIC accumulation across the CONUS.