Improved estimation of soil organic carbon stock in subtropical cropland of Southern China based on digital soil mapping and multi-sources data

Accurate information on spatial patterns of Soil Organic Carbon (SOC) stock in cropland is crucial for soil quality management, agriculture production, and carbon cycling regulation. This research aims to produce the finest map (30 m) of SOC Density (SOCD) and estimate SOC stock in the plow layer of the cropland of Jiangxi Province, which is an important grain-production region in China but its knowledge of the quantity and spatial pattern of SOC stock majorly remain unclear. To fill this gap, we used the recursive feature elimination to choose the best predictors for mapping SCOD. Then we constructed a random forest model to map the SOCD in cropland across Jiangxi Province, quantified relative importance of predictors, and calculated the SOC stock using soil type method and the digital soil mapping method. Finally, structural equation model was adopted to analyze the impacts of various factors on SCOD. We hypothesized that using machine learning method and incorporating various covariates cover soil properties, terrain, climate, biota, lithology, and soil management policies can well capture the spatial variability of SOCD in the cropland. Our results indicating the averaged SOCD in cropland’s plow layer of Jiangxi Province is 2.95 kg m⁻². Our model explained 54% of the variance. The terrain factors had the largest contribution to mapping SOCD. The total SOC stock in the plow layer of the survey region is 9148.33 × 10⁴ tons. Soil properties, soil management practices, and lithological factors affect SOCD mainly through direct way while terrain, climate, and biota mainly via indirect way. Additionally, the mean value of SOCD significantly decreased from 3.23 in the 1980s to 2.95 kg m⁻² in the 2010s. Our results offer critical information on the spatial pattern in SOCD and its potential dominators, which enable us to make climate-smarter agricultural policies when making strategies for cropland management.