Background Soil organic carbon is an important component of carbon storage in terrestrial ecosystems, due to severe soil erosion in karst rocky desertification areas, soil fertility has decreased and soil organic carbon has been lost, causing significant impacts on the ecological environment of the watershed. In this study, the spatial heterogeneity of soil organic carbon content in karst rocky desertification areas was studied．The relationship between organic carbon content and other soil physicochemical properties, Environmental factor variables was discussed. This will provide a scientific basis for increasing soil carbon sink and improving soil quality.

Methods We collected soil samples from 0–20 cm and 20–40 cm, air dried them naturally and took an appropriate amount of sample to grind and sieve, and measured soil physical and chemical indicators. Firstly, based on soil carbon content data, we calculated the single point soil organic carbon density. Secondly, field soil sampling was carried out in the karst rocky desertification area of Lujiang River Basin. The spatial distribution of soil organic carbon content was predicted based on geographically weighted regression (GWR) model and the random forest (RF) algorithm in combination with topographic variables, climate variables and soil variables.

Results 1) GWR model predicted that the local determination coefficients for the average contents of soil organic carbon in topsoil (0–20 cm) and subsoil (20–40 cm) in the study area were between 0.42–0.65 and 0.48–0.62, and the explanatory factors had strong spatial heterogeneity. 2) The prediction accuracy of soil organic carbon content in topsoil and subsoil was 82.3% and 83.4%, respectively, based on the GWR model, the predicted values of soil organic carbon mass concentration in topsoil and subsoil were obtained 0.90–11.25 kg/m² and 0.85–5.49 kg/m² respectively, the maximum, minimum and average values of the topsoil were higher than those of the subsoil. 3) The spatial distribution of soil organic carbon contents was that the high values appeared in topsoil in the east and west wings of the Lujiang River Basin, while in subsoil in the west and northeast of the Lujiang River Basin. 4) Based on the analysis of influencing factors via RF algorithms, the most influential factor of the topsoil was alkali-hydrolyzed nitrogen (72.73%) and the most influential factor of subsoil was annual mean temperature (39.39%), both the upper and lower layers were significantly affected by ammonium nitrogen and total nitrogen. The effect of soil physicochemical properties on the spatial distribution of regional organic carbon content was remarkable than that of external natural environmental factors.

Conclusions This study reveals the spatial differentiation characteristics of soil organic carbon in the Lujiang River Basin and analyzes various indicator factors that affect the spatial distribution of soil organic carbon, which may provide a theoretical support for promoting soil organic carbon cycle, reducing soil erosion and improving soil fertility in Lujiang River Basin.