
East River Catchment, Gunnison County, Colorado.
As world population grows, so do concerns that water availability and water quality will continue to diminish. Changes in land use, climate change, and extreme weather exacerbate these concerns, which threaten not only our fresh water supply but also systems that rely on watershed exports, such as hydropower and agriculture.
Although watersheds are recognized as Earth’s key functional unit for assessing water resources, a predictive understanding of how watersheds respond to environmental perturbations is challenging due to the complex nature of the system. What happens along the flowpath of water in a watershed drives complex, multi-scale interactions among the plants, microorganisms, organic matter, fluids, and minerals. These processes, which vary significantly across watershed compartments and gradients, are responsible for biogeochemical cycles that determine downstream exports of carbon, metals and nutrients, such as nitrogen.
While formidable challenges exist in gaining a predictive understanding of watershed hydro-biogeochemical processes across scales and ecosystem compartments, several recent developments now provide a springboard for advancing prediction over scales relevant for resource management, with sufficient resolution, and in a tractable manner. New networked and autonomous sensing systems are providing diverse, streaming datasets associated with a wide range of watershed processes. Toward-exascale computing capabilities are making it easier to simulate multi-scale hydrology-driven biogeochemical interactions–from bedrock to canopy, along significant lateral gradients, and across terrestrial and aquatic interfaces. And recent machine learning advances hold significant opportunity to transform our ability to rapidly use diverse datasets and physics-based models for rapidly predicting how watersheds respond to perturbations.
The new “ExaSheds” project led by Berkeley Lab PI Carl Steefel represents the first systematic effort to advance powerful machine learning and Exascale computing to transform our ability to predict watershed behavior and increase the use of ever-larger and more-complex data obtained from watershed field observations. In partnership with Oak Ridge National Laboratory, Pacific National Laboratory, and Los Alamos National Laboratory, the new ExaSheds project, which is funded by DOE Biological and Environmental Research, initially will take advantage of datasets being collected at the East River, Colorado watershed site that has been developed as part of Berkeley Lab’s DOE Watershed Function Science Focus Area.

The ExaSheds project is the first-ever systematic effort to advance powerful machine learning and Exascale computing to transform our ability to predict watershed behavior and increase the use of ever-larger and more-complex data obtained from watershed field observations.
Susan Hubbard is Berkeley Lab Associate Laboratory Director for EESA and lead of the Watershed Function SFA project. “The SFA’s vision is to transform watershed science through advanced computing and domain science,” she said. “Ultimately we hope to deliver predictions of how water availability and water quality change with early snowmelt and other perturbations – at unprecedented spatial resolution and mechanistic detail yet up to the river-basin scale.
“The ExaSheds team–which also includes co-PI Scott Painter (ORNL), Dipankar Dwivedi (LBNL), David Moulton (LLNL), Xingyuan Chen (PNNL), Ethan Coon (ORNL), and Ben Brown (LBNL)–will integrate and leverage spectacular watershed hydro-biogeochemistry and computational data science expertise in its effort to transform our ability to predict how watersheds are responding to a rapidly changing environment,” according to Hubbard.
“The ExaSheds project will develop groundbreaking capabilities critical for meeting this goal,” she said. “We expect the developed ExaSheds capabilities to be useful at the East River watershed site as well as to watersheds throughout the world.”