Visualizing Forecast Uncertainty in a Model Physics Parameterization Ensemble of the 29 June 2012 Derecho

Effective visualization of forecast uncertainty is an imperative for operational meteorology. With large volumes of numerical weather model data available to them, forecasters require efficient and accurate means of obtaining the information that they need. In predicting severe weather events, the amount and type of variation among forecast ensemble members is crucial to know. This research examines a simple and easily modifiable program, written by the authors in the statistical scripting language R, that allows an individual to view multiple graphical depictions of forecast ensemble variation for any model output parameter. The script presents an ensemble mean plot for the chosen parameter on a map of the domain (an example in Fig. A-1a) and two types of linear cross-section plots (Fig. A-1b and A-1c) to show the solutions of individual ensemble members. The script is scalable to large data sets and any number or type of model ensemble members. It can be run on a typical personal computer through an x-windows system with no resource strain, given appropriate parallel computing and network resources on the server side. Outliers in the model ensemble can be easily determined, and the overall spread of the ensemble forecast for each parameter across the temporal domain can be seen graphically with this program. In this study, the script is tested on a numerical weather model forecast of the 29 June 2012 Ohio Valley/Mid-Atlantic derecho containing 21 members, including 7 microphysics and 3 cumulus physics parameterizations selected for presumed fitness in forecasting severe convective weather events. The event was not considered to be well-forecast for a severe weather event, a critical factor in its selection as the test case for the uncertainty visualization script. For this derecho event, the script was used to visualize modeled virtual temperature, wet-bulb temperature, simulated radar reflectivity, downdraft convective available potential energy (DCAPE), and the Derecho Composite Parameter (DCP). It was found that the cross-section plots enable users to quickly distinguish patterns among the output of the model physics parameterizations. Most of the models had particular difficulty simulating radar reflectivity in a multicellular linear convective form, but the graphical cross-sections enabled good-performing members to be readily distinguished. Significant variations in all the chosen output variables existed, with distinct patterns for certain microphysics becoming apparent in the program's graphical output. The visualizations allowed users to determine that, in this case study, microphysics variation created more uncertainty in selected derecho-related parameters than did cumulus physics variation. This finding indicates the additional usefulness of the program to research meteorologists seeking to examine the uncertainty introduced by varying different types of model physics. The authors believe that this script has significant applicability for operational and research meteorologists in quickly and easily discerning the scope and sources of uncertainty within ensemble forecasts.