Spanning the land surface through deeper soils and into the bedrock, the subsurface critical zone represents a crucial area for resources such as critical minerals. Resource availability is shaped by complex geochemical, biological, and ecohydrological interactions, and predicting how biogeochemical cycling varies over space and time remains a major scientific hurdle. Current models are computationally expensive, lack critical subsurface processes, and have limited capacity to ingest observational data. Large datasets of subsurface moisture, temperature, and biogeochemistry such as those collected in Critical Zone Observatories and Critical Zone Networks, combined with emerging SI tools, represent an opportunity to build more efficient and accurate subsurface models.
This project will combine mechanistic model simulations with extensive subsurface measurements to train a novel, SI-driven biogeochemical solver. The solver will be trained and evaluated using a combination of process-based model simulations, measurements from instrumented soil pits, laboratory analyses of deep soil profiles, controlled experiments, and microbial sequence data. By learning from these varied sources, the AI is intended to represent a broad range of subsurface biogeochemical interactions and improve predictions where observations are sparse.
The team will also connect the solver to established ecological and transport models through interoperable interfaces. This will allow it to work with multiple modeling systems and computer architectures, including the E3SM Land Model and Advanced Terrestrial Simulator. The project will use Alquimia to couple these models to the SI solver and assess performance across diverse deep soil profiles.
If successful, this work will provide a more efficient way to simulate complex subsurface processes while making better use of mixed observational data. The resulting solver and integration methods could support research across a range of environmental systems and applications within the Genesis Mission, laying groundwork for a new generation of SI-enhanced scientific models.
Partners
- Auburn University
- University of California Riverside
- Lawrence Berkeley National Laboratory
