Spatial patterns of soil organic carbon stocks in German forests at 10 m resolution

This is the authors' abstract. We don't add key points for this paper.

Forest soils are highly heterogeneous and global maps often fail to capture the forest floor and subsoil, making depth-resolved mapping of forest soil organic carbon (SOC) quantities particularly challenging. SOC stock estimation involves two stages, calculating stock values from measured data and their subsequent spatial regionalization. Here, we tested different regionalization approaches using National Forest Soil Inventory (NFSI) data for German forests: A direct Calculate-Then-Model (CTM) approach, an indirect Model-Then-Calculate (MTC) approach, and an extended Model-Then-Model (MTM) approach based on random forest and residual kriging. The study is based on data from 1796 sampling locations from all over Germany, and the study aimed to (1) produce SOC stock maps for forest floor, topsoil, and subsoil at 10-m resolution, (2) compare direct and indirect approaches across depth intervals, and (3) evaluate the plausibility of predicted SOC distributions against measured inventory data. Total SOC stocks across forest floors, topsoil, and subsoil to 1 m ranged from 51 to 202 tC ha −1 , with pronounced regional contrasts across Germany. The Model-then-Model (MTM) approach performed best, achieving the highest predictive accuracy and the lowest errors in terms of calibration, goodness-of-fit, and systematic over- and underestimation along the SOC stock gradient. Cross-validated R 2 values reached 0.69 for forest floor, 0.66 for topsoil, and 0.42 for subsoil, outperforming the other approaches. The strongest predictors differed by depth, with humus-related and vegetation variables dominating forest floor SOC stocks, whereas topsoil and subsoil SOC stocks were mainly controlled by soil properties, especially predicted SOC concentration, coarse fragments, and bulk density. These results show that forest SOC stocks are depth-specific and require modeling approaches that reflect distinct ecological and pedological controls. The Model-Then-Model (MTM) framework improved predictive performance and map quality, making it a promising tool for forest-specific carbon accounting and climate-relevant SOC assessments for greenhouse-gas reporting.