Assessment of Time-Lagged, Medium-Range, High-Resolution Ensemble Forecasts of Extreme Precipitation Events

In May 2023, the National Science Foundation National Center for Atmospheric Research produced real-time, 204-h, 10-member ensemble forecasts with an FV3-based numerical weather prediction model. These forecasts had 3-km horizontal grid spacing over the conterminous United States (CONUS), were initialized daily at 0000 UTC, and were evaluated with a focus on extreme events as defined by spatially-varying 1-h precipitation accumulation average recurrence intervals (ARIs) from the National Oceanic and Atmospheric Administration’s ATLAS 14 data. Time-lagged ensembles of up to 80 members were constructed from these forecasts, where the maximum time-lagged ensemble size depended on the number of overlapping forecasts that existed at a specified time. For example, 60-member time-lagged ensembles could be produced for 60–84-h forecasts. With a goal of identifying the point of diminishing returns, where more time-lagged members did not improve the forecast quality of extreme precipitation, time-lagged ensembles were assessed over the CONUS east of the Rockies. Reliability statistics and Brier skill scores for events defined by 1- and 5-year ARIs showed time-lagged ensembles of 20–40 members yielded better 3–6-day forecast quality than non-time-lagged 10-member ensembles. However, increasing the size of time-lagged ensembles beyond 40 members provided limited additional benefits. These collective results suggest that 20–40 member time-lagged ensembles may represent the point of diminishing returns for predicting 1- and 5-year extreme 1-h precipitation events.