An Assessment of Convective Forecast Failures with the WRF-ARW Model

During the past four springs and summers, 36 hour realtime forecasts were conducted daily with WRFARW using a 4 km horizontal grid resolution and explicit convection over the central and eastern US, using the 00 UTC NAM model for the initial state and boundary conditions. The goal was to assess the potential benefits of using such resolutions for forecasting convective systems, such as squall lines, bow echoes, mesoscale convective vortices, etc. While surprisingly good forecasts were obtained on many occasions, offering valuable new guidance as to convective system mode and propagation, significant forecast failures were also common. The sources of such forecast failures ranged from the negative feedbacks of prior convection, to the incorrect structure and evolution of the boundary layer, to misrepresenting key mesoscale forcing features. One of the more noteworthy observations was that significantly bad (as well as good) forecasts often corresponded closely with equally bad (or good) guidance from the coarserresolution NAM, which was used for the initial and boundary conditions. Additionally, the larger forecast errors could generally not be remedied via sensitivity testing with model physics, resolution, or simple variations to the initial state (such as RUC versus NAM initializations); e.g., errors in the fine resolution model formulation do not seem to explain these forecast errors. This suggests that resolvable features not currently represented in the initial state or boundary conditions used for such forecasts may be responsible for at least some of these forecast errors. This emphasizes the critical need for mesoscale assimilation systems that can potentially correct for such errors in the largerscale analyses, in addition to assimilation systems that emphasize convectivescale features . In this presentation, the most common modes of forecast failure will be categorized, and initial attempts to remedy such errors will be discussed.