It is crucial to improve our understanding of the relative importance of multiple factors regulating soil organic carbon (SOC) persistence over continental scales. Previous studies found conflicting results on the importance of climate vs. geochemical factors (e.g., aluminum and iron mineral phases) for predicting SOC concentrations and stocks. Most of these studies utilized linear statistical models. To reconcile the controversy, we used data from > 2000 soil horizons from > 600 individual pits from National Ecological Observatory Network (NEON) sites spanning diverse ecosystems across North America, typically collected to one meter depth. We examined the relative importance of soil geochemical properties, climate, depth, and their interactions for predicting SOC concentration and stock using both linear mixed models and nonlinear models: generalized additive mixed models (GAMMs) and random forest models (RFMs).
Results/Conclusions
We found that climate was a fundamental driver of SOC concentration and played similarly important roles as some geochemical predictors; yet, this finding only emerged in the nonlinear models and was obscured in linear models. Most of the predictors had nonlinear threshold relationships with SOC. In all three models, depth, oxalate-extractable Al (Alox), pH, and exchangeable calcium plus exchangeable magnesium (Caex+Mgex) were important predictors for SOC concentration and stock while clay content and oxalate-extractable Fe (Feox) were weaker predictors. Even though climate, pH, and geochemical predictors were related, they were not redundant, and they frequently interacted: for example, Alox and Feox were most important in wet or cold climates. We observed a generally decreasing importance of geochemical predictors of SOC with increasing soil depth, challenging previous statements in the literature. Overall, our findings indicated that soil geochemistry, climate, and pH controlled SOC through direct effects as well as their interactions. Nonlinear models captured the dual importance of climate and geochemical predictors and showed threshold relationships between SOC and most of the predictors.