Simulation of decadal variations in soil organic carbon stocks in French forests

Forest soils play a critical role in the global carbon cycle, yet predicting their response to environmental change remains limited by the scarcity of models validated against long-term observations. Here, we tested whether AMG, a simple soil organic carbon model originally developed for croplands, could simulate the dynamics of SOC stocks in temperate forests. We adapted the original AMG model to forest conditions, creating the AMG-Forest model. To better represent forest soil functioning, we added to the original AMG model a new compartment representing the surface litter layer dynamics. We further initialized soil carbon pools using Rock-Eval® thermal analysis, which helps distinguish more stable from more active forms of soil carbon. We calibrated and evaluated the AMG-Forest model over SOC data obtained from two campaigns performed on average 15.5 years apart over 97 plots in the French long-term forest monitoring network (RENECOFOR). The AMG-Forest model reproduced the average trend toward SOC accumulation observed in the RENECOFOR forest soils, but its site-level predictive skill was poor and no better than a mean-model baseline; it also failed to capture inter-site variability and systematically overestimated sequestration rates, likely because calibration conducted separately for broadleaf and coniferous plots could not compensate for missing site-specific controls such as belowground carbon inputs, litter quality, and microclimatic regulation of decomposition. A complementary Random Forest analysis, using standard pedoclimatic descriptors, likewise explained little of the observed variance in the SOC trends. Although part of this unexplained variance likely reflects measurement uncertainty and missing drivers, the observed decoupling between litter and mineral soil carbon stocks was robust, indicating that litter accumulation did not translate consistently into mineral SOC gain over the resampling interval. These results show that transferring cropland soil carbon models to forests requires more than simple structural change and reparameterization. Reducing uncertainties in forest soil carbon accounting will therefore require better constraints on both candidate forest-specific controls, such as microclimate and belowground carbon inputs, and the observations used for model calibration and evaluation.