Parent material modulates land use effects on soil organic carbon fractions but not inorganic carbon in subtropical red soils

Soil carbon pools, including soil inorganic (SIC) and organic carbon (SOC) along with its particulate (POC) and mineral-associated (MAOC) fractions, play a vital role in soil health and the global carbon cycle. Their spatial patterns and effect factors, particularly the interactions between parent material (PM) and land use (LU) in subtropical red soil regions, remain poorly understood. This study integrated field sampling (97 sites, 194 samples from 0-40 cm depth) across diverse PM and LU types in Ji’an City, China, with machine learning to map the spatial distribution of SIC, SOC, POC, and MAOC stocks. Machine learning models (RF, SVM, GBM, XGBoost) achieved satisfactory prediction accuracy for all carbon pools (R2 > 50%). Results revealed a fundamental decoupling of control factors, with SIC stocks (1.56–1.66 Mg C ha-1) were mainly influenced by climate (Aridity Index, AI; 10.4-15.4% individual effect), while SOC stocks (27.21–18.33 Mg C ha-1) and its fractions were primarily governed by LU through soil total nitrogen (TN; 45%–54% and 37%–54% individual effects for SOC and MAOC, respectively). A significant effect of LU, PM, and their interactions (P < 0.05) was observed. Paddy soils and Quaternary red clay-derived soil showed the highest SOC stocks. SIC did not correlate strongly with LU or PM but was higher in subsoil. LU exerted a direct negative effect on SOC fractions (path coefficients: -0.19 for POC, -0.49 for MAOC) but only affected SIC indirectly through soil pH and EC. PM directly influenced only MAOC. Environmental factors explained 65%–87% of SOC variance within specific LU or PM systems, with TN being key in croplands, while soil pH controlled SIC accumulation. This study underscores divergent biogeochemical mechanisms governing SIC and SOC, supporting tailored management strategies for improving carbon sequestration in subtropical ecosystems.