Image Warping for Forecast Verification

Complete details for the approach outlined here are given in Lindström et al. (In Prep). The main idea of the approach is to apply image warping to the forecast field to make traditional verification comparisons of the raw fields, the fields after having applied an affine transformation to the forecast field, and finally to the fields after having additionally applied a nonlinear transformation to the affinetransformed forecast field. The approach is similar to that of Dickinson and Brown (1996) and Alexander et al. (1999), with some important differences; particularly with respect to: (i) in determining the landmarks used to control the warp, and (ii) the warping function. Some other similar, but not as closely related, methods are described in Keil and Craig (2007); Nehrkorn et al. (2003). The primary idea of the image warp is to morph one image to look as much like another image as possible. In doing this, the pixel coordinates are rearranged rather than a smoothing or spectral decomposition of the fields (e.g., Ebert, 2007; Casati et al., 2004; Briggs and Levine, 1997). That is, for the forecast field represented by, F (s) where s for our purposes here represent the grid coordinates