Spatial verification of high-resolution weather forecasts: beyond traditional metrics

A suite of new verification methods has recently emerged to deal with high resolution gridded forecasts. As grid spacing has decreased, forecasts have improved due to less reliance on sub-gridscale parameterization. Even though the forecasts look more realistic, this improvement is not captured well by traditional methods of spatial verification. Traditional methods that rely on a gridpoint to gridpoint comparison between the forecast and observation field will typically show lower skill for smaller grid spacing. This is a fundamental limitation of methods such as critical success index, false alarm ratio, and equitable threat score. Model developers can also artificially increase their skill scores by simply adjusting the bias. These concerns have led to alternate verification methods based on object identification, bias-adjusted CSI, and spatial error decomposition. At a meeting in Boulder in Feb 2007, developers of these new methods convened to show how their unique methods could be applied to common set of gridded observations and forecasts (web page). We will present results from this meeting, including the inherent strengths and weaknesses of the new verification methods.