Estimating Subsurface Properties from the Air: Linking Above and Below-Ground Observation

Machine learning finds relationships between above and belowground features to estimate bedrock properties across an entire watershed.

Digital elevation model of the East River Watershed, overlain by an aerial photograph and geophysical data used in this study.

[Reprinted under a Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0) from Uhlemann, S., et al. “Surface Parameters and Bedrock Properties Covary Across a Mountainous Watershed: Insights from Machine Learning and Geophysics.” Science Advances 8 (12), (2022). DOI:10.1126/sciadv.abj2479.]

The Science

Mountainous watersheds are often referred to as the world’s “water towers” because they provide more than half of earth’s freshwater. Climate change can influence watershed function and delivery to communities downstream. To predict the impact of this change, scientists must understand how water flows in the ground and how the earth’s properties affect this flow. However, measuring the earth’s properties is difficult—especially over a large area. Researchers have tested how to use observations from space or from the air to estimate the earth’s properties. The team demonstrated this method at a mountainous watershed close to Crested Butte, CO, one of the best characterized watersheds in the world. Results showed that, although the relationships are complex, the earth’s subsurface properties vary with properties on the earth’s surface, such as the angle of hillslopes, their gradient, elevation, and the vegetation that grows on them. Using these relationships, researchers can predict what the subsurface looks like and map features in the subsurface that are controlling groundwater flow.

The Impact

Protecting and monitoring groundwater is becoming increasingly critical in light of climate change and prolonged droughts. Understanding how the subsurface affects groundwater flow is crucial not only to predict how this resource may change over time but also to develop management approaches. This research shows that critical subsurface properties can be predicted from observations of the Earth’s surface, which are much easier to measure. Knowing the Earth’s properties will eventually lead to better management of groundwater resources and drought resilience.


Bedrock measurements are critical for predicting the hydrological response of watersheds to climate disturbances. However, estimating how water flows in bedrock over watershed scales is difficult, particularly in areas where bedrock may be cracked. By linking data from subsurface and surface measurements, researchers used machine learning to test the co-variability of above and belowground features throughout an entire watershed. The team studied the relationships between bedrock properties, surface formation features, and vegetation to show that relationships derived from machine learning can estimate most of their co-variability. Using these relationships, the team predicted bedrock properties across the watershed and showed that regions of lower variability provide better estimates. The results emphasize that this integrated approach can be used to derive bedrock characteristics on a smaller scale, allowing for a better understanding of subsurface variations across an entire watershed. Knowing how bedrock may vary with surface properties may be critical to assess the impact of disturbances on freshwater function in these ecosystems.

Principal Investigator

Sebastian Uhlemann
Lawrence Berkeley National Laboratory
[email protected]

Co-Principal Investigator

Eoin Brodie
Lawrence Berkeley National Laboratory
[email protected]


This material is based upon work supported as part of the Watershed Function Science Focus Area (SFA) funded by the Biological and Environmental Research (BER) Program within the U.S. Department of Energy’s (DOE) Office of Science (Award Number DE-AC02-05CH11231).


Uhlemann, S., et al. "Surface Parameters and Bedrock Properties Covary Across a Mountainous Watershed: Insights from Machine Learning and Geophysics." Science Advances 8 (12), (2022).