Experiments in very short period forecasting of convective storms using radar extrapolation and numerical weather prediction methods

A wide variety very short period (1-6h) forecast techniques are evaluated for the ability to forecast convective storm events over the central U.S.. The techniques include simple radar extrapolation, numerical (WRF, RUC and MM5), explicit cloud resolving model (Forecast VDRAS), and combined observational and numerical methods. The experiments and evaluation were centered on two areas 1) Texas and 2) Illinois, Indiana and Ohio. In addition particularly significant initiation events in the eastern two-thirds of the U.S are included. The data is obtained from a real-time experiments conducted at NCAR during June 2005.

Summary evaluations of the strengths and weaknesses of the different methods will be provided for the purpose of determining optimum procedures for combining techniques to produce 0-6 h nowcasts of convection. Methods for blending techniques will be suggested based on the synoptic situation, organization and scale of the convective systems. Also presented are human derived forecasts prepared by a forecaster observing all methods.

Radar extrapolation techniques were based on both cell and area trackers. The 4 km NCAR version of WRF was run at 00 UTC each day. MM5 used a nested grid with an inner domain of 3.3 km and it was run every 3 h using observational nudging. Forecast VDRAS is a 4DVAR cloud resolving model that was cycled every 30 min producing 0-1h forecasts of precipitation. The RUC (Rapid Update Cycle) was a 12 km version of the operational RUC that has been modified to produce probability precipitation forecasts. An experimental version of the National Convective Weather Forecast Product called NCWF6 produced 1-6 h probability forecasts and was updated every 10 min, it utilized both echo extrapolation and heuristic methods.