Predictability of Precipitation in the Southeast United States, Part II: Composites and Case Studies

Forecasting heavy precipitation in the southeastern U.S. is challenging, especially in unstable environments. The strength of synoptic-scale forcing for ascent, mesoscale circulations, moisture-transport feedbacks, and organized convection combine to present difficulties for NWP models during heavy precipitation events. This study utilizes the Earth System Research Lab (ESRL) Reforecast 2 dataset to identify the meteorological characteristics of poorly forecasted and well forecasted precipitation events in the Southeast U.S., with the goals of (i) providing insight into the cause of the forecast errors, and (ii) providing forecasters with tools to aid in recognition of potentially low-skill QPF events. Finally, a representative case study of a badly forecasted event is analyzed in depth, to elucidate potential sources of model error.

The Reforecast 2 dataset, essentially a frozen version of the Global Ensemble Forecast System (GEFS), is used here to identify the best and worst precipitation forecasts from 1985 to 2013. Considerable sensitivity to the skill metric was evident, resulting in a combination of MAE and ETS, with elimination of events that were close to the ETS threshold. Details regarding determination and categorization of the cases will be presented in Part I. Day 5 forecasts in convective environments will be emphasized in this presentation, although other environments and lead times were also considered. Event-relative composites were constructed for both the well forecasted events (“good” cases) and the poorly forecasted ones (“bad” cases). These composites demonstrate that good forecast cases exhibit both stronger synoptic forcing and moisture transport at low levels. Both good and bad events were associated with a lifting trough, suggesting that latent heat release from precipitation just downstream of the trough axis may play a role in forecast error.

Event relative composites were made for Reforecast data as well, and compared to the analyzed data at the corresponding time. Results show that the Reforecast is characterized by approximately the same magnitude and spatial extent of precipitation as was observed. However, the model forecasts indicate a spatial shift to the northeast relative to analyses, indicating that precipitation systems tended to propagate too quickly in the model atmosphere. Additionally, the Reforecast composites feature less low-level moisture compared to observed.

Performing a case study on a poorly forecasted event will be done with the Weather Research and Forecasting model. This should provide further insight into the source of the model error, which can be imparted to forecasters and ideally employed to fix certain model issues. This will lead to better forecasts that will have positive impacts inside and outside the meteorological community.