Monday, 14 January 2002: 11:29 AM
Improved approaches for measuring the quality of convective weather forecasts
Traditional approaches to the verification of convective forecasts are inadequate for a number of reasons. These methods generally rely on overlaying a forecast grid over an observation grid and computing standard statistics based on the 2x2 verification table. A major flaw with this type of verification is that it is insensitive to the size of the forecast error in terms of intensity, areal coverage, and location. A second problem associated with verification of these types of forecasts is related to the inadequacy of the observations, which generally are based on rain gages (which poorly sample the regions of interest) or radar observations (which provide an indirect measure of the phenomena of interest). This paper will review alternative verification methods that have been developed for various applications and will consider how they might be applied for convective forecast verification. Additionally, new methods developed specifically for verification of these types of forecasts and for verification of precipitation forecasts will be demonstrated. Results obtained from the application of a prototype approach, which involves measuring the translation and rotation errors associated with large-scale operational convective forecasts, will be presented. The additional diagnostic information that can be obtained from these types of approaches will be considered in detail. Finally, the implications of directly considering the uncertainty associated with verification observations will be discussed.