The CSU-MLP machine learning model has been used by researchers to test and refine the model, enhancing forecasters' confidence in their predictions and potentially saving lives.
Early warnings and precise predictions are critical as severe weather approaches, potentially saving lives.
Russ Schumacher, a professor at the Colorado State Climatologist's Department of Atmospheric Science, has led a team in developing an advanced machine learning model to improve hazardous weather prediction across the continental United States. The model, known as CSU-MLP (Colorado State University-Machine Learning Probabilities), has evolved to predict events such as tornadoes and hail four to eight days ahead of time, providing an essential opportunity for forecasters to provide information to the public.
Aaron Hill, a research scientist who has worked on improving the model for the past two or three years, led the team to publish their medium-range (four to eight days) forecasting capability in the American Meteorological Society journal Weather and Forecasting.
Aaron Hill, a research scientist, provides the CSU-MLP to forecasters at the Storm Prediction Center. Credit: Provided/Aaron Hill
The researchers have now teamed up with weather forecasters at the National Storm Prediction Center in Norman, Oklahoma, to test the model and refine it based on actual weather forecasters' practical considerations. The tool is not a stand-in for human forecasters' valuable knowledge, but rather a proactive, confidence-boosting tool that helps forecasters make informed decisions about future weather.
Hill said that our statistical models can be beneficial to operational forecasters as a guidance tool rather than as a replacement.
Israel Jirak, M.S. '02, Ph.D. '05, is a science and operations officer for the Storm Prediction Center and co-author of the paper. The collaboration with the CSU team was described as a "very successful research-to-operations endeavor."
Jirak said the CSU guidance tool is extremely reliable and powerful while also being practical for forecasters.
Allie Mazurek, a CSU Ph.D. student, talks about the CSU-MLP with forecaster Andrew Moore. Credit: Provided/Allie Mazurek
The model is trained on a large dataset of past weather data covering the continental United States. These data are combined with meteorological retrospective forecasts, which are model "re-forecasts" created from previous weather events. Storms and hail are examples of storms that have been studied in real-time.
Allie Mazurek, a Ph.D. student, is researching which atmospheric data inputs are the most important to the model's predictive capabilities. "If we can better understand how the model is calculating its conclusions, we may better diagnose why the model's predictions are correct or wrong in certain weather scenarios," she said.
Hill and Mazurek are working to make the forecaster's work more accurate and transparent.
It's heartening to know that years of effort in developing the machine learning tool are now making a difference in a public, operational setting.
Hill said he loves fundamental research and I enjoy learning new things about our environment. But having a system that is providing improved warnings and enhanced messaging around the possibility of severe weather is also rewarding.
Aaron J. Hill, Russ S. Schumacher, and Israel L. Jirak, 2 February 2023, Weather and Forecasting, DOI: 10.1175/WAF-D-22-0143.1.
