Predicting localized extreme precipitation events and their subsequent hydrologic impacts continues to provide a formidable challenge to hydrologic forecasters. Part of the challenge stems from the general inability of traditional, coarse resolution (≥ 12-km horizontal grid spacing) numerical weather predication (NWP) models to well-simulate localized, heavy precipitation events. However, output from explicit, convection-allowing, high resolution (≤ 4-km horizontal grid spacing) computer models in both deterministic and ensemble frameworks has been demonstrated to more skillfully predict smaller-scale, heavy precipitation events. Thus, high resolution NWP forecasts have the potential to provide significantly better guidance for the prediction of precipitation, and therefore, hydrologic impacts, as well.

This study examines model forecasts produced during the 2007 NOAA Hazardous Weather Testbed (HWT) Spring Experiment in the context of optimizing probabilistic precipitation guidance in terms of forecast consistency, quality, and value. During the Experiment, the Center for Analysis and Prediction of Storms (CAPS) at the University of Oklahoma produced daily 10-member 4-km horizontal resolution ensemble forecasts. Each member used the WRF-ARW core with explicitly-resolved convection, was initialized at 21 UTC, and ran for 33 hours over a domain covering approximately 3/4 of the continental United States. Different initial condition (IC), lateral boundary condition (LBC), and physics perturbations were introduced in four of the ten ensemble members, while the remaining six members used identical ICs and LBCs, differing only in terms of microphysics and PBL schemes.

As the first known real-time application of a convection-allowing ensemble, the forecasts allowed us to explore different approaches for generating post-processed probabilistic precipitation guidance. In this regard, a “neighborhood” approach is described and shown to considerably enhance model forecast skill of heavy precipitation events when combined with traditional techniques of producing ensemble probability fields.

These results have important implications for maximizing the utility of probabilistic high resolution model output for heavy rain and hydrologic prediction. Additionally, future hydrologic models may significantly benefit from ingesting these post-processed probabilistic precipitation fields. Practical examples of how to formulate and use this guidance will be discussed at the conference.