[Influencing Factors Analysis and Spatial Estimation of Soil Organic Carbon Content in the Hilly Areas of the Loess Plateau].

Soil organic carbon （SOC） is a crucial indicator for assessing soil fertility. Understanding its spatial distribution patterns and influencing factors is essential for enhancing agricultural sustainability and securing national food security. This study focused on the Fuxian County, Yan'an City, and Shaanxi Province, selecting 22 environmental variables related to SOC formation from four types of environmental factors： topography, climate, vegetation, and soil. Three digital soil mapping methods, random forest （RF）, support vector machine （SVM）, and geographically weighted regression （GWR）, were employed to establish SOC content estimation models. The influencing factors and spatial distribution characteristics of SOC content at 0-20 cm soil depth for the entire study area, garden land, cultivated land, and forest land were analyzed. The results showed that： ① The average （SOC） across the entire region of the Fuxian County was 8.54 g·kg, with garden land at 6.44 g·kg, cultivated land at 7.49 g·kg, and forest land at 10.22 g·kg. The coefficients of variation were 36.90%, 19.24%, 29.88%, and 32.56%, respectively, all of which fall into the moderate degree of variation. ② In the entire region of the Fuxian County, topography, climate, vegetation, and soil factors all significantly affected the distribution of SOC, with notable differences in their effects on SOC. In forest land, slope （SLP）, mean annual temperature （MAT）, and bulk density （BD） had significant negative effects on SOC, while mean annual precipitation （MAP） and total nitrogen （TN） had significant positive effects on SOC. In garden land, pH, TN, and total potassium （TK） all had significant positive effects on SOC. In cultivated land, MAP had a significant negative effect on SOC, while TN had a significant positive effect. ③ Comparing the performance of different estimation models, the RF estimation model used in this study had the highest , the lowest root mean square error （RMSE） and mean absolute error （MAE） values, and the smallest model prediction error. In the linear fitting between measured and estimated values, the RF model's fitting accuracy reached above 0.85, demonstrating the best estimation performance among the models. ④ Utilizing the RF model for spatial estimation of SOC content in the Fuxian County revealed a distribution pattern of lower concentrations in the east and higher in the west. The results can provide decision-making reference for the optimization and adjustment of land-use structure in hilly areas of the Loess Plateau and offer technical support for the accurate estimation of SOC.