How Far into the Medium Range Can Probabilistic Excessive Rainfall Forecasts be Extended?

The predictability limits for heavy precipitation can vary widely, from just an hour or two for localized convective systems to several days for atmospheric rivers and tropical cyclones. Until recently, NOAA’s Weather Prediction Center (WPC) issued Excessive Rainfall Outlooks (EROs) for forecast days 1 through 3; in early 2022, day 4 and 5 outlooks were added as experimental products. One tool available to WPC forecasters when generating the ERO is known as the Colorado State University-Machine Learning Probabilities (CSU-MLP) system, which uses global ensemble reforecasts, historical observations, and random forest machine learning models to produce “first guess” probabilities of excessive rainfall. CSU-MLP products were also recently extended from the day 1-3 guidance available in operations to forecast days 4-8.

Quantitative evaluation of CSU-MLP forecasts over a 2+-year period show that there is statistically significant skill out to forecast day 6, when using a smoothed daily climatology as the baseline. Although day 7-8 forecasts do not show positive skill in aggregate, they can provide useful guidance in some widespread heavy rain events. The skill of the medium-range forecasts is greatest along the west coast of the US, with considerable skill also along the Gulf of Mexico coast and the northeastern US. Comparatively poor skill was found in the interior west, especially where the North American Monsoon produces most of the excessive rainfall. The presentation will include a recap of CSU-MLP performance during representative events, including the atmospheric rivers that led to extreme precipitation and flooding in California in 2022-23. The evaluation of these experimental products in WPC operations will also be summarized.