The Three Gorges Reservoir Area (TGRA) in China has extensive non-timber product plantations (NTPP), in which integrated management based on intensive fertilization and weeding were required to maintain and improve yields for a long time. Uncertainties still existed regarding the compound effects of environment and the long-term integrated management on soil organic carbon content (SOC) in NTPP. Data from 341 sampling plots covering six primary NTPP types were collected to investigate the influence of environment and management on topsoil (0–10 cm) SOC of NTPP using a coupled algorithm based on machine learning and structural equation modeling. Results showed significant differences and spatial variabilities in SOC content among different types of NTPP. Integrated management accounted for approximately 53% of the accumulation of topsoil organic carbon, surpassing the total contribution of topography, climate, vegetation, and soil properties in NTPP of TGRA. SOC content increased with available nitrogen for NTPP at all altitudes in TGRA. The study highlighted the potential of enhancing SOC storage through adaptive integrated management in NTPP of vast areas. Improving soil organic carbon stock in large area of non-timber production plantations would benefit the realization of carbon neutralization in next decades.

Accurate monitoring of forest soil organic carbon (FSOC) is essential for climate change mitigation and biodiversity conservation. Although digital soil mapping has proven effective in capturing the spatial variability of FSOC, its application remains challenging in ecologically heterogeneous regions such as China. This study developed a national 90 m gridded FSOC content product (four depth intervals) with quantified uncertainty at 90 m resolution, covering four depth intervals from 0 to 100 cm (0–20, 20–40, 40–60, and 60–100 cm). By integrating 8,709 soil profiles with 41 environmental covariates within a quantile regression forest (QRF) framework enhanced by forward recursive feature selection (FRFS), we simultaneously predicted FSOC content and quantified associated uncertainties. The model demonstrated robust and unbiased performance, with 10-fold cross-validation mean explained variance (MEC) ranging from 0.69 to 0.80 across soil depths. This high-resolution FSOC content dataset provides a critical spatial baseline for forest carbon management under China’s “Dual Carbon” strategy, with direct applications in carbon sink verification and soil carbon sequestration potential assessment.