A Dual-Driven Framework Integrating Remote Sensing and Spatial inference for Mapping Soil Organic Matter under Straw Cover

Crushing and spreading crop straw on the soil surface is a widely adopted agricultural management practice for preserving farmland and improving soil quality. However, satellite-based mapping of soil organic matter (SOM) typically requires exposed soil conditions. Persistent straw cover therefore poses a substantial challenge to conventional remote sensing-based SOM mapping. To address this issue, this study proposes a dual-driven SOM mapping framework (DDF) that integrates remote sensing inversion and spatial inference by distinguishing bare-soil and straw-covered areas. Within this framework, bare-soil areas are regarded as high-reliability observation areas, and their inversion results serve as conditioning information for spatial inference in observation-limited areas. In bare-soil areas, spectral variables were extracted from Sentinel-2 imagery, and sensitive features were identified using the Boruta–SHAP algorithm. A Random Forest model was then employed to map SOM in these high-reliability domains. In straw-covered areas, SOM was estimated through spatial inference constrained by the inversion results derived from bare-soil areas, enabling spatially continuous SOM mapping under heterogeneous surface conditions. The DDF achieved an R² of 0.82 in bare-soil areas and an R² of 0.76 in straw-covered areas. The results indicate that under spatially heterogeneous surface observation conditions, incorporating bare-soil inversion results as conditioning information into the spatial inference process effectively overcomes observation-induced mapping constraints and enables spatially continuous reconstruction of SOM at the regional scale.