MOBILE-PATH: Machine Learning Optimization for Lateral gene Exchange PATHways – Phase I
The Subsurface Critical Zone (SCZ), extending from the land surface through deep soils and into bedrock, plays a vital role in the movement and availability of water, nutrients, contaminants, organic matter, and critical minerals. The processes that shape these resources are closely connected: soil chemistry, microbial activity, plant roots, and changing water and oxygen conditions all affect how materials are stored, transformed, and transported underground.
Scientists use biogeochemical and reactive transport models to study these interactions, but current models can be computationally expensive and difficult to connect with the varied observations collected in the field and laboratory. Measurements may range from continuous records of soil moisture, temperature, and redox conditions to occasional samples that reveal soil structure, microbial communities, and organic matter. Using these different forms of evidence together is essential for representing subsurface processes accurately.
This project will develop an SI (Super Intelligence)-driven biogeochemical solver that learns from both process-based model simulations and real-world observations. By combining continuous measurements with discrete samples, laboratory analyses, and other data, the solver aims to improve predictions of subsurface geochemistry while reducing the computational effort required. The project will also connect the solver with established ecological and hydrological models through flexible interfaces, enabling it to work across different modeling systems and computing environments.
The team will evaluate the approach across a range of deep soil profiles and integrate it with models including the E3SM Land Model and Advanced Terrestrial Simulator. These evaluations will compare the accuracy and speed of SI-enabled simulations with current approaches, including PFLOTRAN. The project’s targets are to maintain or improve predictive accuracy while achieving at least twice the computational efficiency for biogeochemical simulations, and at least five times the efficiency for reactive transport simulations using the integrated models.
If successful, this work will make it possible to study complex subsurface systems more efficiently and with better use of available data. The resulting solver and integration framework could support research into nutrient cycling, organic matter degradation, contaminant movement, and critical mineral availability. It will also provide reusable capabilities for environmental modeling and help advance SI-enhanced scientific tools across the Genesis Mission.
Partners
- Princeton University
