BOZEMAN – A Montana State University snow and atmospheric scientist is working with a team of researchers to harness artificial intelligence for improved precipitation forecasting in the mountain West.
The project, funded by the U.S. Department of Energy through its Genesis Mission, aims to train AI to inform more accurate precipitation forecasting that could help ensure natural hazards preparedness, consistent hydropower generation and protection of the nation’s grid infrastructure. The project is being led by DOE’s Lawrence Berkeley National Laboratory in partnership with Lawrence Livermore National Laboratory, Boise State University and MSU.
Scientists currently rely on models that aren’t yet able to make highly accurate precipitation forecasts in complex, mountainous terrain, explained Marianne Cowherd, the leader of MSU’s portion of the project and an assistant professor in the Department of Earth Sciences in the College of Letters and Science.
People are also reading…
“One of the problems is that we don’t have a great understanding of how precipitation forms in clouds and how that results in where, when, how much and what phase of precipitation ends up falling,” Cowherd said. “That can have a big impact – we might see rain on one side of a mountain, snow on the other side, and the rain runs off immediately while snow accumulates. We’re often very wrong.”
To solve that problem, the researchers will train an AI platform developed at the Berkeley lab to interpret errors and missed observations in real data. They will provide the platform with historic atmospheric data collected during a severe winter storm and ask it to compare those data against real-world observations and measurements made during that event – the so-called “Santa Slammer,” which dumped up to 100 inches of snow across parts of the Colorado Rockies in late December 2021.
Cowherd said humans understand that collected data is imperfect and rife with variables, but it would take people years to sort and compare the vast amount of data gathered during the Santa Slammer to understand why precipitation models failed to accurately forecast the event’s magnitude. AI could complete the job in a fraction of the time, though it first needs training to understand that errors are inherent in real-life data observation.
“AI assumes everything is a perfect observation, so we’re doing a lot of work to create this understanding of what real data can and can’t provide, including situations where we might have more than one instrument measuring the same thing and they don’t agree,” Cowherd said. “The process here is developing the software infrastructure and scientific guidance for how to connect this large, large dataset with lots of observations.”
To train the platform, the team is using Santa Slammer data collected by the DOE’s Atmospheric Radiation Measurement mobile facility. It includes full-scale radar scans from Earth’s surface to the upper atmosphere, as well as photos of precipitation particles and images taken by high-altitude weather balloons.
For the initial phase of the project, Cowherd and one of her graduate students have linked the AI platform to a forecasting model to create a feedback loop between the two as the system learns to properly interpret Santa Slammer data. At the end of the current phase, the team will demonstrate improvements in the AI-assisted observations by comparing forecasts from older and updated versions of the model.
In the spring, the team will apply for a second phase of funding to answer more fundamental questions about precipitation processes. Phase 2 could include teaching the AI platform to analyze other types of weather events in different locations and to assess the influence of distant atmospheric events on local conditions.
Cowherd said she is enjoying the collaboration with her national lab and Boise State colleagues, all of whom have different expertise related to their common interest. Ann Almgren, the Berkeley lab senior scientist who is heading the ARM-Driven Simulation for Physical Insight Research Engine, or ASPIRE, project, added, “Reliable power generation relies on understanding the big picture down to the smallest details, which requires true team science.”
The first phase of ASPIRE received $750,000 in Genesis Mission funding, $40,000 of which was allocated to MSU. Cowherd said the project aligns with MSU’s implementation of machine learning into its research and workflow.
“More accurate precipitation forecasting could strengthen our ability to prepare for natural hazards, manage critical water and energy resources, and build more resilient communities across Montana and surrounding states,” said Alison Harmon, vice president for Research and Economic Development. “This project reflects our ability to bring together deep scientific expertise, modern tools like artificial intelligence and strong partnerships to address significant problems.”
The Genesis Mission, announced in 2025, is a national initiative to unite DOE national labs, industry and academia to accelerate science through AI, according to the DOE website.
