Mapping soil organic carbon sequestration potential in croplands using a combined proximal and remote sensing approach

Abstract

Spatially explicit mapping of soil organic carbon (SOC) sequestration potential helps identify areas with the largest deficit for SOC accrual, but its implementation in a digital soil mapping (DSM) framework requires high-quality observational datasets that are often too resource-intensive to establish. The objective of this study was to develop an integrated proximal and remote sensing approach to efficiently map SOC sequestration potential in an agricultural region (25, 596 km²) of northeast China. Proximally sensed soil mid-infrared (MIR) spectra were used to develop memory-based local learning models that allowed accurate predictions of the mineral-associate organic carbon (MAOC) and clay+silt fractions (R²=0.97 and 0.98), which were then used to obtain point-based (n = 1158) estimates of SOC sequestration potential based on the C saturation concept. Using the training dataset augmented by spectral inference, a DSM model was built via random forest (RF) regression, with a collection of remote sensing derived environmental covariates as predictors. The performance of the RF model (R²=0.65, RMSE=1.42 kg/m²) represented a 30 % increase from the alternative with no addition of spectrally inferred data. The resultant map depicted a west-to-east increase in SOC sequestration potential, largely driven by the increasing capacity for C saturation. Shapley Additive exPlanation (SHAP) analysis revealed that the functional relationships between covariates and the target property were site-specific, but the climatic variables (mean annual precipitation (MAP) and mean annual temperature (MAT)) were on average the most important factors, followed by soil spectral indices that are related to soil texture and SOC. From a spatial perspective, the positive effect of MAP on SOC sequestration potential turned negative for areas with the highest MAP, due to erosion-induced topsoil loss that led to coarsening soil texture and thus diminishing capacity for SOC storage. This developed approach provided insights into the patterns and magnitude of SOC sequestration potential, serving as a basis for targeted cropland management and erosion control practices.