Integrating omics, machine learning, and process-based land surface model to predict hydroclimate feedbacks of microbial functions and its implication for soil carbon emission

Climate change is altering the spatiotemporal pattern of precipitation toward more extreme conditions. However, it’s still unclear how a more variable hydroclimate influences soil biogeochemical cycling and resultant soil carbon emission. One key challenge is our limited understanding of how hydroclimate coupling with other environmental drivers regulates the composition and functions of soil microbial communities. Moreover, how this environmental feedback of soil microbial communities mediates soil biogeochemical processes. To overcome this challenge, we integrated published metagenomics datasets across the US to identify eight general soil enzyme functional classes (EFCs) involved in soil carbon (C), nitrogen (N), and Phosphorus (P) cycles. We then integrated this omics-informed microbial functional information with the corresponding hydroclimate and other environmental data to train and test a machine learning (ML) pipeline for predicting the spatial distribution of EFC composition across the US domain and its variability with changing hydroclimate. This ML-predicted microbial functional feedback to changing hydroclimate was finally coupled with the Community Land Model (CLM5.0) to assess its impact on microbially-mediated soil carbon emission. Our study showed that soil enzyme functional composition is sensitive to changing hydroclimate. Microbial communities decrease the investment in EFCs involved in SOM decomposition under drying conditions. Incorporating this microbial feedback to hydroclimate into the CLM5.0 captured soil carbon dynamics in the water-limited region. Output from this study, including the gridded EFC composition dataset and coupled model framework, can be applied to mitigate the uncertainty in projecting soil carbon-climate feedback under changing hydroclimate.