Mapping soil organic carbon stocks in Swiss subalpine soils: External covariates versus field-mapped soil and vegetation types

Soil organic carbon (SOC) stocks in mountain terrain vary strongly over short distances, but dense sampling is rarely feasible, creating uncertainty in catchment-scale SOC assessments. Here, we test how much spatial prediction of SOC stocks in subalpine Swiss catchments improves when using freely available environmental covariates versus field-mapped categorical covariates. We sampled soils at 48 locations in each of two sites (Ar du Tsan and Blatt; ≈ 2100–2200 m a.s.l.) and quantified SOC stocks for the upper 50 cm. We mapped SOC using regression-kriging, with random forest regression on covariates followed by ordinary kriging of residuals, and evaluated alternative covariate sets using repeated training–validation resampling. Observed SOC stocks ranged from 16.7 to 622 Mg C ha -1 , with highest values concentrated in flat, wet areas associated with organic soils. External covariates (elevation, curvature, normalized difference vegetation index, geology) improved predictions relative to a mean baseline, reducing RMSE from 137 to 114 Mg C ha -1 in Ar du Tsan and from 165 to 122 Mg C ha -1 in Blatt; geology provided the largest gain, whereas curvature added little. Adding field-derived covariates had site-specific effects: in Ar du Tsan, including soil and vegetation types reduced RMSE to 85 Mg C ha -1 , while in Blatt the best model used soil type alone (RMSE 97.9 Mg C ha -1 ) and vegetation type did not improve performance. Overall, targeted soil-type mapping is a pragmatic approach to capturing high-carbon patches that dominate SOC prediction error in heterogeneous alpine landscapes, providing practical guidance for SOC monitoring in similar subalpine catchments.