Comparing geospatial techniques to predict SOC stocks

Soil organic carbon density (SOCD) has strong spatial variability and dependency, and its impact factors vary with changes in scale and geographic location. With 272 topsoil samples (0-30 cm) collected from Chahe Town in the Jianghan Plain, China, we (i) investigated the impacts of environmental variables and land cover types on the spatial distribution of SOCD; (ii) estimated the spatial distribution of SOCD by using global and local spatial interpolation models, including geographically weighted regression kriging (GWRK), regression kriging, geographically weighted regression (GWR), multiple linear regression (MLR), and ordinary kriging; and (iii) used mean absolute estimation error (MAEE), mean relative error (MRE), root mean square error (RMSE) and Pearson's correlation coefficients (r) to evaluate the performance of these models. SOCD was significantly correlated with elevation, normalized difference moisture index (NDMI), and the nearest distance to road (TRD) and residential area (p < 0.05). The SOCD ranged from 0.33 kg ha−1 to 10.14 kg ha−1 for the topsoil in the study area. Most of the study area, especially the middle region, exhibited SOCD ranging from 4.3 kg ha−1 to 7.2 kg ha−1. The highest SOCD value was in wetland (5.45 kg ha−1) and the lowest was in unused land (4.18 kg ha−1). The effects of different environmental variables on SOCD can be revealed by the coefficients of GWR and MLR. The spatial distribution map of dominant variables can help us distinguish the essential environmental influence variables for SOCD in different geographical locations. GWRK outperformed the other models in terms of the lowest MAEE (0.984 kg ha−1), RMSE (1.665 kg ha−1), MRE (0.190), and high r (0.559) values. Thus, GWRK is a promising approach for mapping SOCD at a local scale.