Biome- to global-scale controls over soil carbon storage: divergence in observations and process-based models

Abstract:

The storage and persistence of soil organic carbon (SOC) exhibit considerable heterogeneity and are governed by several key factors, including climate, plant productivity, edaphic properties, and disturbance history. However, it remains unclear which of these predictors dominate in distinct biomes and across spatiotemporal scales. Historically, soil and ecosystem model development has focused on the role of climate and plant productivity in regulating SOC storage. Emerging theories, however, place a growing emphasis on litter quality and mineralogical properties in controlling the long-term persistence of SOC. Here we use global soil observations and an ensemble of soil biogeochemical models to quantify the relative importance of key state factors – namely, mean annual temperature, mean annual precipitation, net primary productivity, and clay and silt content – in explaining biome- to global-scale variation in SOC stocks. We use a machine-learning approach to disentangle the role of covariates and elucidate individual relationships with SOC, without imposing any relationships apriori. We find that SOC predictability is significantly lower for soil profile measurements than for gridded data products and biogeochemical models – with 24%, 60%, and up to 87% of variance explained, respectively. Furthermore, in comparing measurements and models, we observe a mismatch in the importance of key predictors. Models overrepresent the importance of temperature and primary productivity (especially in forests and grasslands, respectively), while measurements suggest a greater relative importance of mineralogy. We observe a non-linear decrease in SOC with increasing temperature and a linear increase in SOC with increasing clay and silt content across data sources, though the magnitude varies substantially between models and measurements. Interestingly, SOC-predictor relationships also vary between the soil profile measurements and gridded data products, suggesting that underlying environmental sensitivities may not be adequately captured in gridded products. Elucidating the role of key SOC controls, and the discrepancies between models and measurements, is essential for improving and validating process representations in soil and ecosystem models for predictions under novel future conditions.