New methods to discriminate forecast skill in mesoscale weather predictions and characterization and application of model error statistics

DTRA operates a suite of high-resolution forecast models for use in atmospheric dispersion prediction. A major problem has been the inability of traditional validation metrics to support the subjective consensus that higher resolution models provide greater skill than coarser models. We believe that the skill scores used over the past 40 years are ill-suited to validating solutions with significant high-frequency components that are seen in both high-resolution model forecasts and observations. WeatherFlow will investigate several spectral techniques to validate high-resolution model wind forecasts. In Phase I we will explore the capability to transform time series of wind observations and forecasts into frequency spectrum's and develop skill scores based on differences in frequency space. These validation schemes will place great emphasis on the correct prediction of observed features and spectral energies. As a result they should not have the effect of traditional schemes which heavily penalize correct solutions with slight temporal/spatial displacements while lightly penalizing forecasts which have very little variability in them. Future extensions for Phase II could include schemes to validate additional forecast variables, use of forecast skill to develop model output statistics corrections to the raw forecast solution, and an improved determination of model forecast uncertainty.