Realtime mesoscale analysis with WRF-DART for explicit convective forecasts

Springtime convective forecasting remains a significant challenge. Both model error and initial condition uncertainty have been shown in the literature to negatively impact convective forecasts. Continuous cycling ensemble data assimilation (CCENDA) systems have been suggested as a candidate for next generation probabilistic forecast systems to address the convective forecast challenge. Through sensitivity studies with CCENDA we have been able to identify sources of analysis error, and in particular sources of model bias. This is achieved by diagnosing the mean misfit between short-term ensemble forecasts and observations over many assimilation cycles (O100 or more). In particular, adjustments in model physics in the cycled analysis system that reduced model bias were found to improve analysis quality and produced more skillful convective forecasts. While this result is not surprising, systematic approaches to identify and remove model bias sources are non-trivial.

During the 2012 spring season, twice daily explicit convective forecasts were made at NCAR in realtime in support of the Deep Convective Clouds and Chemistry (DC3) field campaign. Significant improvement in forecast skill was noted relative to a similar realtime forecast exercise during Spring 2011. At the conference we will: 1) discuss model improvements that enhanced convective event forecast skill during the 2012 season; 2) highlight forecasts of a few significant convective events from the 2011 and 2012 seasons 3) review continuing efforts to improve forecast skill and 4) outline plans for future expansion to realtime explicit ensemble forecasts.