Eric Gilleland, NCAR

Assessing Weather Forecasts

Abstract

In the efforts to produce better weather forecasts, a key issue is in
assessing the performance of new forecast models.  Traditional
verification procedures are based on grid-point to grid-point
comparisons, which can lead to poor measures because of small errors
in location even if the size and intensity of the forecast are
correct.  This is especially a problem as the resolution of forecasts
improves.  Often the usefulness of the forecast is not conveyed by the
traditional verification scores.  Many spatial forecast verification
methods have recently been proposed to provide more informative
assessments of performance.  In this talk, the features-based method
of Davis et al. (2006), now called the Method for Object-based
Diagnostic Evaluation (MODE) tool, is presented.  The method uses a
convolution-threshold procedure to identify weather features of
interest, and then assesses the forecast based on attributes of these
features.