An integrated artificial intelligence and E3SM hierarchical modeling framework for elucidating environmental responses of soil carbon and nutrients dynamics and its implications for land carbon-climate feedback

Yang Song, Department of Hydrology and Atmospheric Sciences, University of Arizona (Principal Investigator)
Forrest M. Hoffman, Oak Ridge National Lab (Co-Investigator)
Nathaniel Collier, Oak Ridge National Lab (Co-Investigator)
Umakant Mishra, Sandia National Lab (Co-Investigator)

We propose developing a comprehensive hierarchical framework that merges artificial intelligence (AI) with DOE’s Energy Exascale Earth System Model (E3SM). This framework will explicitly represent the environmental impact on soil biogeochemistry, optimizing both model complexity and computational efficiency. Our central goals are to identify crucial microbial processes that reduce E3SM simulation uncertainties and to quantify land carbon feedback to atmospheric CO₂ and physical climate over the long term, considering the intricacies of soil biogeochemical modeling and uncertainties in input data.

To achieve this research objective, we will develop an integrated AI-E3SM hierarchical framework that can employ different model complexities to simulate microbially mediated soil carbon and nutrient dynamics in response to changing environments. We will apply the developed AI-E3SM framework and International Land Model Benchmarking (ILAMB) metrics to identify key environmental response mechanisms of soil biogeochemical processes and soil environment initialization datasets to reduce E3SM biases in simulating carbon, water, and energy cycles across time and space. We will conduct the 1pctCO₂ simulation experiment to assess E3SM land carbon feedback to atmospheric CO₂ concentration and physical climate and quantify their uncertainties from different complexities of soil biogeochemical modeling.

Our project will address Biogeochemical processes, feedback, and interactions within the Earth system by delivering three key outcomes, including (1) A grid-scale global distribution of soil enzyme functional composition contributing novel benchmark data on microbial functional diversity to the ILAMB benchmarking system; (2) An integrated AI-E3SM hierarchical modeling framework that can leverage AI to enhance E3SM’s predictive capabilities, computational efficiency and uncertainty quantification in modeling soil carbon-climate feedback; and (3) An insight into key soil biogeochemical processes and input data that can mitigate uncertainty in E3SM simulation and quantifying land carbon feedback to atmospheric CO₂ concentration and physical climate at different temporal scales. These outcomes will facilitate the development of E3SM and advance Earth system models’ capacity to integrate the power of AI to improve the prediction of soil carbon-climate feedback across time and space. Moreover, the designed AI-E3SM will enable the assimilation of gene-to-ecosystem data products and leverage the application of multiple DOE facilities, such as JGI genomic databases, EMSL MoNet data network, NMDC, and ILAMB benchmark data.