Mapping future soil carbon change and its uncertainty in croplands using simple surrogates of a complex farming system model

Soil organic carbon (SOC) in agricultural soils is vital for soil fertility for sustainable agricultural production and climate change resilience. Process-based farming system models are widely used to predict SOC dynamics in agricultural soils, but their application at regional scales is largely limited by computational requirements, data availability, and uncertainties in model predictions. Here we present an approach of combining a farming system model and a simplified surrogate model that emulates and mimics the behaviour of complex process-based models to predict SOC change (∆SOC) and its uncertainty in Australian dryland cropping regions under anticipated climate change. We first calibrated and validated the farming system model APSIM for simulating ∆SOC (0-30 cm soil) using data from 90 farming-system trials at 28 sites across the study regions. Next we conducted a comprehensive simulation across the region using the validated APSIM model to predict ∆SOC over the period 2009-2070. Then simple surrogate models were developed based on the APSIM outputs. The surrogate models were able to explain >96% of the variation in APSIM-predicted ∆SOC. Last the surrogate models were applied across the regions at the resolution of 1 km. In our simulations, Australian dryland cropping soils under farmers' common management practices and future climate conditions were a net carbon source (0.66 Mg C ha−1 with the 95% confidence interval ranging from −5.79 to 8.38 Mg C ha−1) during the 62-year period. Across the regions, simulated ∆SOC exhibited great spatial variability ranging from −108.8 to 9.89 Mg C ha−1 at the resolution of 1 km, showing significant (P < 0.05) negative correlation with baseline SOC level, temperature and rainfall, and positive correlation with pasture frequency (the duration of pasture in the rotation divided by the whole duration of the rotation) and nitrogen application rate. The uncertainty in ∆SOC and the underlying drivers were also assessed. This study presented a novel approach to efficiently predict future SOC dynamics and their uncertainty at fine resolutions, facilitating the development of site-specific management strategies for soil carbon sequestration.