Improved Broad-Scale Modelling of Soil Organic Carbon Dynamics Following Land-Use Changes

Land-use changes (LUCs) strongly impact soil organic carbon (SOC) stocks over decades. However, there are too few long-term field experiments where these SOC dynamics have been observed long enough to validate process-based models for large-scale use. We have developed a new data-driven space-for-time approach for model validation using empirical data from over 3000 sites in the German Agricultural Soil Inventory, including 212 sites with LUC between cropland and grassland. Machine-learning models trained on sites under permanent land use were used to predict equilibrium SOC stocks for similar sites with changed land use. We used this derived data set to assess how well the process-based model RothC describes SOC dynamics following LUC. The default version of RothC struggled to capture the fast changes in SOC following LUC since it was mainly driven by differences in carbon input quantity and quality. Losses in SOC after converting grassland into cropland occurred faster than modelled, and SOC accrual after converting cropland to grassland was faster than simulated. This suggested an additional carbon stabilisation mechanism connected to grassland land use. We extended the RothC model with an additional carbon pool that builds up rapidly after grassland establishment, similar to aggregate-protected SOC. This improved the model efficiency from 0.49 to 0.80 for transitional croplands and from −3.39 to 0.90 after establishing grassland. This improved model version, RothC-LUC, is suitable for simulating SOC dynamics following LUC between cropland and grassland on a broad scale, such as in national inventory reports on greenhouse gas emissions.