Evaluation of Convective Storm Initiation Nowcasts and Forecaster Involvement
Rita D. Roberts, Amanda R. S. Anderson, Barbara G. Brown
National Center for Atmospheric Research, Boulder, CO U.S.A.

The NCAR Convective Storm Nowcasting System (AutoNowCaster; ANC) has been running at the National Weather Service Dallas/Ft. Worth office for 6 years (2005-2010). A primary motivation for conducting this Forecaster-Over-the-Loop (FOTL) demonstration was to assess the benefit gained by having forecasters provide input into the FOTL-ANC system and to evaluate the impact on nowcast skill.

Historically it has been challenging to show increased skill in automated prediction of convective storm initiation as convective storms typically form over small spatial scales and offsets in time or location can affect the forecast skill significantly. In addition, evaluation of convective-scale storm initiation nowcasts are hampered by the impacts of large-scale storm characteristics and by the substantially larger number of storm extrapolation nowcasts compared to storm initiation nowcasts that dominate the summary statistics, making it difficult to single out the performance measures associated with storm initiation nowcasts. Forecaster input into a mostly automated system created an additional challenge to separate and evaluate their specific contributions toward the accuracy of the nowcasts.

Some new approaches have been taken to address these challenges that include: 1) conducting the evaluation over small convective-scale sub-domains than over the larger-scale ANC domain, 2) analysis of nowcast performance using high-temporal time-series plots from selected cases, 3) objective approach for stratifying nowcast performance with and without forecaster input and 4) use of contingency table performance diagrams to simultaneously investigate changes in multiple forecast performance attributes. The verification results from these different approaches showed that it is possible to single out and evaluate the convective-scale, storm initiation nowcasts and that the performance of automated nowcast products can be improved with forecaster involvement with the nowcast system.