Ensemble Model Forecast Performance of HREF and REFS for 2024 Extreme Precipitation Cases

Ensemble forecasts from numerical weather prediction systems produce a large quantity of information, which is powerful but at the same time challenging to interpret. Some methods of parsing these datasets, including averaging and point probabilities, may obscure important features of the constituent member forecasts. At the same time, a gap exists between big-data-focused verification statistics commonly used for model evaluation and some forecast qualities relevant to users. Ensemble skill assessment is further complicated by the diversity of user needs, encompassing both forecasters concerned with point predictions at specific times and locations, and infrastructure managers interested in long term statistics for the most impactful events over varying domain sizes.

In this presentation, we assess forecast skill and ensemble spread-skill in the accumulated precipitation predictions of two convection-allowing ensemble systems: the operational High Resolution Ensemble Forecast (HREF) system and the prototype Regional Ensemble Forecast System (REFS), on an assortment of flood and other notable extreme precipitation cases that occurred during 2024. Performance is examined for factors including precipitation maximum value and location, areal coverage, and heavy precipitation timing and duration, incorporating aspects of ensemble confidence and dispersion that both indicate the health of the ensemble system and influence user decision-making. The strength of the relationship between these tailored assessments and more generic, bulk skill scores will also be evaluated, to gauge how accurately certain widely-used verification methods translate to practical forecast value for these high-impact events.