To improve the accuracy of remote sensing inversion modeling of soil organic matter (SOM) at the field scale, this study selected a 41.3 ha (hm²) field in the black soil region of northeastern China as the study area. The spatial differences of soil physical and chemical properties, reflection spectra and topographic factors, as well as the relationship between spectral index, topographic factor and SOM were analyzed. Multiscale segmentation and hierarchical clustering were also conducted based on measured SOM samples, bare-soil Sentinel-2A images and 4-m-resolution digital elevation model (DEM) data. Then, the stepwise regression SOM prediction models were constructed for the entire study area and its subregions. The results led to the following conclusions: (1) SOM presents significant spatial heterogeneity in the study area; (2) the SOM prediction accuracy of single model for the entire study area was low (R² = 0.16, RMSE.cal = 1.61, and RMSE.val = 1.45); and (3) entire study area can be divided into four subregions, i.e., the “sedimentary area,” “deposition-buffer area,” “erosion-buffer area” and “erosion area,” the SOM prediction model was established for each subregion according to its SOM content and spectral as well as topographic characteristics, and SOM prediction accuracy for the entire study area clearly improved (R² = 0.58, RMSE.cal = 1.17, and RMSE.val = 1.30). In this study, prediction models employing the subregion method were established and represent a new approach for SOM inversion research in the future. The results can be used as a reference for the study of SOM inversions of farmland blocks in the black soil region of northeastern China and provide theoretical and technical support for soil carbon pool estimations and precision fertilization.