Accurate estimation of soil organic carbon (SOC) stocks in mountain ecosystems is essential for understanding carbon–climate feedbacks. Mountain soils store large amounts of carbon due to low temperatures that constrain decomposition, making them sensitive to ongoing climate change. Here, we quantified SOC stocks in crystalline mountains in the southern Alps and compared two complementary modelling approaches: (i) a data-driven ensemble of machine-learning models grounded in the SCORPAN framework, and (ii) a simplified, process-inspired model representing carbon input–output balances. Samples ( N = 185) were collected across gradients of elevation, land-cover, and topography. Statistical models achieved higher accuracy (R 2 = 0.42) than the process-inspired model (R 2 = 0.28), identifying growing degree days, precipitations, and vegetation productivity as the main drivers of SOC variability. Partial response analyses indicated maximum SOC stocks under intermediate temperature, with lower stocks at warmer sites, highlighting strong temperature sensitivity. The process-inspired model provided complementary insights by estimating relative model-derived mean residence times (10–100 years), which increased with elevation and were highest in grasslands and heathlands. These patterns suggest that high-elevation SOC persistence is primarily driven by climatic constraints on decomposition rather than carbon inputs. Both modelling approaches consistently indicate a potential sensitivity of alpine SOC to warming. Divergences between model predictions on steep slopes further emphasize the role of geomorphological processes, such as erosion, in shaping SOC distribution. By integrating machine learning with a process-inspired framework, this study advances computational approaches for SOC mapping in complex mountain environments and highlights the complementarity of modelling approaches. • Semi-empirical and statistical models jointly improve C cycling understanding in mountains. • Statistical models best predicted SOC (R 2 ≈ 0.42), outperforming process-inspired (R 2 ≈ 0.28). • Semi-empirical model predicted SOC mean residence time from 10 to 100 years across the massif. • Climatic factors were the dominant SOC drivers in high elevations. • Both approaches suggest high warming vulnerability of SOC stocks in alpine belt.
Bonfanti et al. (Fri,) studied this question.