Improving soil organic carbon spatial distribution and interpretation of cross-scale drivers with probabilistic wetland representation

Despite inland freshwater wetlands containing approximately 30% of global SOC while only occupying 6-8% of terrestrial land area, they remain spatially under-represented, resulting in incomplete characterization of landscape-scale carbon distribution. Here, we assess if mapping new wetland areas, using the Wetland Intrinsic Potential (WIP) tool, improves modeling the landscape distribution of SOC stock and SOC concentration. We collected soil samples in four watershed study areas across the Pacific Northwest and evaluated both linear mixed models and quantile random forest models to predict SOC stock distribution and mineral soil SOC concentration. A random forest model with WIP as a predictor was the best performing and most accurate model for SOC stocks. However, WIP was not included in models for SOC concentration in the mineral soil. Instead, edaphic factors including depth, pH, and clay content were primary predictors of mineral SOC concentration, consistent with the well-known accumulation of coarse, particulate organic material in wetlands. Using the best SOC stock model, we mapped SOC stocks across the four watersheds with scaled up estimates of mean wetland SOC stocks of 114-243 Mg C ha-1 compared to mean upland SOC stocks of 99.5-162 Mg C ha-1. The differences between wetland and upland SOC stocks were contrasted in comparison of new model maps to current SOC stock maps from an ensemble of global and region models. This study demonstrates how incorporating probabilistic wetland maps can show wetland-driven patterns of large SOC stocks contained in organic soils.