Assessing Numerical Weather Prediction Uncertainty in Warm-Season Rainfall Ensemble Simulations

It is well known that numerical weather prediction models have inherent uncertainty associated with them. The use of ensemble modeling partially accounts for this uncertainty by providing multiple model solutions, but traditional ensemble approaches use the average of ensemble members in the formulation of model output. It is of value to researchers and forecasters to determine the optimal ensemble member or members that best diagnose the simulated weather event. In the Southeast United States, warm-season rainfall that is primarily convectively driven is a well-documented phenomenon that has little predictability, particularly with numerical weather prediction models. Little work has been done in model parameter selection that best characterizes these warm-season rainfall events. In order to assess ensemble model uncertainty for warm-season rainfall events, 16-member parameterization ensemble simulations of 5 events using the Weather Research and Forecasting (WRF) model were formulated using the North American Regional Reanalysis (NARR) data as input. Through using multiple uncertainty quantifications, including the bootstrap mean confidence interval width, bootstrap max-min width, standard deviation, variance, and inter-quartile range, locations of ensemble model simulation uncertainty were identified and diagnosed. Additionally, resampling statistical methods were utilized to determine which ensemble member best characterized the rainfall patterns for each event, validated with the NEXRAD multi-sensor precipitation estimator (MPE) dataset. By identifying the optimal model parameterization combination for warm-season rainfall, it will be possible to improve the diagnosis and predictability of these events in operational forecasts.