Reconciliation of Process-Based Models and Observations: Collaborative Pathways to Improve Soil Carbon Predictions Across Sub-Saharan Africa

Process-based soil carbon (C) models are increasingly used to project regional and global C cycle responses to climate change. However, the development and evaluation of these models has largely focused on temperate regions of North America and Europe. This geographic bias raises a critical question: Do these models capture generalizable mechanisms that can be applied to underrepresented pedological regions, or do they encode processes specific to their developmental context? Through collaboration between modelers and experimentalists, we evaluated three process-based models—Century, Millennial, and MIMICS—across 777 topsoil samples spanning the climate and pedological diversity of sub-Saharan Africa. Despite their differences in mechanistic detail, all three models performed similarly (adjusted R² = 0.09–0.18) in predicting soil organic carbon (SOC) stocks. Using random forest algorithms trained on observed and modeled SOC data, we identified divergences between drivers of SOC. All three models overemphasized net primary productivity as a SOC driver and misrepresented the role of organo-mineral interactions. Bias analyses revealed that the three process-based models inadequately capture exchangeable calcium, which is increasingly recognized as an important control on SOC. Notably, increased mechanistic complexity did not improve transferability. Our results have significant implications for regional C budgets and global climate projections. They underscore the need for tighter feedback between modelers and experimentalists to incorporate region-specific biogeochemistry—particularly organo-mineral interactions and calcium dynamics—into future soil C models in (sub-)tropical regions. We propose that targeted experimental work on these mechanisms, coupled with model re-parameterization, offers a path toward more reliable climate projections.