Using the Big Weather Web Ensemble to Examine Systemic Biases in WRF-model Parameterizations

The WRF-ARW model is widely used in research and forecasting applications, and one of its strengths is the wide variety of physical parameterizations that can be selected by users. However, this also creates a daunting number (many thousands) of possible configurations. There have been many studies evaluating various aspects of WRF performance, and even more collections of anecdotes about the performance of different model configurations for different applications. Yet there are relatively few analyses of the performance and bias characteristics of different WRF configurations that consider forecasts over long (multi-year) periods.

This study uses the Big Weather Web (BWW) ensemble---an ensemble developed through a collaborative effort distributed across numerous universities---to evaluate biases in different WRF parameterizations, relative to one another and to observations. The full ensemble includes 48 members at 20-km horizontal grid spacing, with a subset of those members altering only one physical parameterization to allow for controlled intercomparisons. At this grid spacing, cumulus parameterizations are required, along with the usual requirements of boundary-layer, microphysical, and radiation parameterizations. This study focuses on systemic biases of the different parameterizations over approximately two full years of forecasts. For example, different cumulus parameterizations are shown to yield very biases in the vertical structure of atmospheric water vapor, and to a lesser extent temperature, when compared to radiosonde observations. Furthermore, this results in bias characteristics in precipitation forecasts that depend on the precipitation threshold being considered. Other biases of this sort, along with speculation as to their underlying causes, will be presented.