Early warning and accurate forecasts are critical as we approach potentially life-threatening hazards such as severe weather, heavy rain, hail or hurricanes. Colorado State University weather researchers have given storm forecasters a powerful new tool to improve the reliability of their forecasts, potentially saving lives.

In recent years, Russ Schumacher, a professor in the Department of Atmospheric Sciences and a climatologist at Colorado State, has led a team to develop an advanced machine learning model to improve severe weather forecasting across the continental United States. Originally trained on historical extreme precipitation data, the model known as CSU-MLP (Colorado State University Machine Learning Probability) was developed to accurately predict events such as hurricanes and hail four to eight days in advance; this is a critical window for the spread of forecasters. information among the public for the preparation.

The team, led by research scientist Aaron Hill, who has worked to develop the model for the past two and a half years, recently published its mid-term (four to eight days) forecasting capabilities in the American Meteorological Society’s journal Weather. And guess.

The researchers have now teamed up with forecasters at the National Storm Prediction Center in Norman, Oklahoma to test the model and develop it based on practical evaluations from real forecasters. The tool does not replace the invaluable skills of human forecasters, but instead provides an agnostic confidence-building measure to help forecasters decide whether to issue public alerts about potential weather.

“Our statistical models can benefit operational forecasters as a guide product, not as a backup product,” Hill said. Said.

Israel Jirak, MS ’02, Ph.D. ’05 is a science and operations officer at the Storm Prediction Center and co-author of the paper. He described the collaboration with the CSU team as “a very successful project from research to operation”.

“They have developed a probabilistic machine that is statistically robust and capable and learns the severe weather pattern that is practically useful to forecasters,” said Jirac. Forecasters in Oklahoma use CSU guidance every day, especially when they need to publish a medium-term forecast.

Nine years of weather data

The model was developed on a very large dataset containing nearly nine years of detailed historical weather observations across the continental United States. These data are combined with meteorological posterior forecasts, which are model “re-predictions” based on the results of past weather events. CSU researchers took environmental factors from these model predictions and correlated them with past severe weather events such as hurricanes and hail. The result is a model that can work with current weather events in real time and generate probabilities of such hazards from four to eight days ahead, based on current environmental factors such as temperature and wind.

Alli Mazurek, Ph.D., is working on the project and trying to figure out which atmospheric inputs are most important to the model’s predictive capabilities. “If we can better parse how the model makes its predictions, we hope we can better diagnose why the model’s predictions are good or bad in certain weather conditions,” he said.

Hill and Mazurek are working to make the model not only more accurate, but also more understandable and transparent to the estimators who use it. For Hill, it’s great to know that years of work perfecting the machine learning tool are now making a difference in public workplaces.

“I love basic research. I love learning about our atmosphere. However, having a system that provides advanced alerts and enhanced messaging about the severe weather threat is extremely helpful,” Hill said.