Big data-driven modelling reveals key mechanisms underlying soil organic carbon stabilization

Abstract:

Soil is the largest carbon pool of the terrestrial ecosystems, yet the mechanisms underlying soil organic carbon (SOC) stabilization are not well characterized nor incorporated in Earth system models. We developed the PROcess-guided DAta-driven deep learning modelling (PRODA) approach to explore mechanisms underlying global soil carbon dynamics. PRODA integrates data assimilation, deep learning, big data with more than 100,000 vertical soil organic carbon profiles, and the Community Land Model version 5 (CLM5) to optimize the model representation of SOC and obtain global retrievals of key biogeochemical processes. The PRODA-optimised CLM5 can represent 56(±2)% spatial variation of SOC across the world. Among all the retrieved biogeochemical processes, we found that microbial carbon use efficiency is among the most important processes in determining the spatial distribution of SOC over the globe. SOC decomposition rates and vertical transportation contribute most to SOC stabilization in subsurface soil and boreal regions. Soil textures were identified as the most significant environmental variables, followed by climatic variables, in regulating the soil carbon storage. Our results show that big data offers rich information to improve SOC representation in Earth system models and to reveal key mechanisms underlying global SOC stabilization.