Abstract

Faster and more powerful computers can handle weather models with less sub-gridscale parameterization and smaller grid spacing. At the same time, advances in model verification are needed to keep pace with these advances in computing resources. As finer-scale features are resolved by today's mode... Show moreFaster and more powerful computers can handle weather models with less sub-gridscale parameterization and smaller grid spacing. At the same time, advances in model verification are needed to keep pace with these advances in computing resources. As finer-scale features are resolved by today's models, valuable information about storm structure and organization may be inferred. Unfortunately, traditional verification measures based on collocated points between observation and forecast (e.g. Gilbert Skill Score, root mean squared error, probability of detection) are insensitive to this potential added quality. Other information about weather features that may help diagnose model deficiencies (such as scale-dependent bias, or displacement error) are not forthcoming from the traditional methods. In this paper, we attempt to compile the latest and most relevant approaches to this problem from a broad segment of the verification research community. We divide the approaches into four broad categories: neighborhood, scale-decomposition, feature, and field verification, and apply them to specific weather events from spring 2005. We also explore how these metrics compare to a subjective evaluation of forecast quality based on a survey of 30 individuals. Some of the goals of this intercomparison are to guide the user to the proper verification method based on his or her needs and to show some of the inherent strengths and weaknesses of the various methods. Show less