Machine Learning Helps Improve How Earth System Models Represent Groundwater

The Science

Groundwater table depth (GWTD) influences how water moves through soils, rivers, and ecosystems. Hence, it also affects how water and energy move between the land and atmosphere. However, Earth system models often struggle to accurately represent groundwater at global scales. This study investigated whether using machine learning to calibrate models could improve the Energy Exascale Earth System Model (E3SM) Land Model’s simulation of GWTD. Researchers developed a framework supported by machine learning. They started with a stand-in computer model that they had trained on land model simulations. They then combined the surrogate model with global observations of GWTD to adjust key parameters related to water runoff in E3SM. This approach improved how well E3SM modeled GWTD compared to simulations that used default parameters. In particular, it improved E3SM’s predictions within the 1 to 5 meter depth range. At this range, groundwater can influence surface processes. However, this representation of GWTD did not consistently improve how E3SM represented other variables related to water. These variables included soil moisture, runoff, and evapotranspiration.

The Impact

This study provided a test of how using machine learning to calibrate an Earth system model can improve how it simulates groundwater. It demonstrated that machine learning can efficiently calibrate model parameters, which would otherwise require computationally expensive global simulations. However, it also revealed the limits of calibration strategies that rely on a single variable. The results showed that improving GWTD alone does not improve the model’s representation of related hydrologic variables. In addition, model calibration based on land simulations driven by atmospheric forcing produces different responses to changes in the model parameters in simulations where land and atmosphere are interactive. These findings highlight the importance of interactions among hydrologic processes in Earth system models. They suggest that to improve Earth system models’ performance, scientists may need to use calibration approaches that optimize multiple hydrologic variables. They may also need to run land and atmosphere models together, rather than land models alone. 

Summary

Accurately representing groundwater processes in Earth system models remains challenging. Groundwater interacts with soil moisture, runoff, evapotranspiration, and atmospheric processes across multiple spatial and temporal scales. In this study, researchers developed a machine learning framework to calibrate GWTD in the Energy Exascale Earth System Model (E3SM) Land Model (ELM). The approach used surrogate modeling to emulate the relationship between model parameters and simulated groundwater depth. This effort allowed researchers to efficiently explore parameter space that would otherwise require many expensive global simulations. The calibration targeted five parameters related to surface runoff and subsurface drainage processes. It also used global groundwater model data constrained by extensive observations as a benchmark. The researchers evaluated the model’s performance against multiple observation-based datasets for runoff, soil moisture, evapotranspiration, and baseflow. 

The machine learning calibration substantially improved simulated groundwater table depth, particularly within the 1 to 5 meter depth range, where groundwater influences land surface energy fluxes. However, the improved groundwater simulations did not consistently improve other hydrologic variables. In some regions, runoff and soil moisture simulations became less accurate. The model also tended to overestimate the baseflow index. The results showed that the improvements in modeling groundwater table depth were strongest in humid regions where groundwater had limited influence on surface processes. Comparisons between offline land simulations and coupled land–atmosphere simulations further showed that groundwater calibration had stronger impacts when atmospheric feedbacks were included. These findings indicated that calibration strategies targeting a single hydrologic variable may not improve overall model behavior. It highlights the importance of integrated calibration approaches that account for multiple hydrologic processes simultaneously in coupled land-atmosphere models.

Research Contact(s)

L. Ruby Leung
Pacific Northwest National Laboratory
Ruby.Leung@pnnl.gov

Yilin Fang
Pacific Northwest National Laboratory
Yilin.Fang@pnnl.gov

Funding

This research was supported as part of the Energy Exascale Earth System Model (E3SM) project, funded by the U.S. Department of Energy (DOE), Office of Science, Biological and Environmental Research, as part of the Earth System Model Development (ESMD) program area. 

Publication

Fang, Y. and Leung, L. R. Machine learning calibration of groundwater table depth in ELM: Impact on land surface hydrology and land–atmosphere fluxes. Journal of Advances in Modeling Earth Systems, 18, e2025MS005184 (2026). [DOI: 10.1029/2025MS005184]