Status of nowcasting thunderstorm initiation: where do we go from here?

The NCAR automated, short term, convective storm nowcasting (Auto-nowcaster) system is a fairly unique nowcasting system combining radar, satellite, surface station, sounding, and numerical model data. It incorporates not only the detection and extrapolation of existing storms, but provides storm initiation, growth and decay nowcasts as well. Arguably, the greatest challenge is in providing accurate spatial and temporal nowcasts of new storm initiation. Several years have been spent attempting to obtain accurate storm initiation nowcasts using a fuzzy logic-based engine to combine predictor fields from observational datasets. More recently, predictor fields derived from frequently-updated numerical model output representing the stability of the atmosphere have been included in the suite of fields used to assess the likelihood for new convective storms. A key ingredient in accurately predicting storm initiation is locating and entering low-level convergence boundaries into the Auto-nowcaster System; boundaries have been shown to play an important role in triggering new convection. Initially automated algorithms have been used to detect convergence boundaries in radar data. However these algorithms have difficulty detecting all portions of convergence boundaries that extend between several radars spaced 300 km or more apart.

In the last few years, NCAR has been collaborating with the FAA, the National Weather Service forecasters and WSI Corporation meteorologists to include forecaster and meteorologist entered, low-level convergence boundaries into the ANC system in real-time to provide more complete representation of the convergence boundaries in the automated system and provide more accurate thunderstorm initiation nowcasts to the aviation community for enroute traffic management planning. In this paper we will present results from two recent Auto-nowcaster demonstrations conducted at the U.S. National Weather Service (NWS) Forecast Office in Ft. Worth Texas and with WSI over the northeastern section of the U.S. during the spring and summer of 2006. In particular, the accuracy of the storm initiation nowcasts will be examined in detail. Standard methodologies for computing PODs, FARs, and CSIs over very large domains fail to illustrate the true skill of most nowcasting systems. We will present an alternative approach (and results) to validating storm initiation nowcasts and assess if further improvements in accuracy can be achieved.