Nonlinear Optical Flow for Verification

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Wednesday, 20 January 2010
Exhibit Hall B2 (GWCC)
Caren Marzban, APL, Seattle, WA; and S. Sandgathe

Handout (1.2 MB)

The verification of spatial fields such as sea-level pressure or precipitation involves the comparison of two gridded fields, one for the forecast and another representing the observed (e.g. analysis). Numerous methods have been recently developed to address the existence and matching of spatial features (or objects) in the two fields. Among them is a class of methods based on an idea - generally referred to as Optical Flow - where the two fields are mapped to one another in some fashion to align synoptic features, and different verification/summary measures are computed from the map. Most optical flow methods are linear in the parameters of the map; but it is shown that the application of optical flow technique to the verification problem requires extension to the nonlinear realm. In this talk, linear and nonlinear regression models are employed for performing the optical flow analysis. One benefit of using these traditional techniques is that standard regression results can be utilized to "explain" the results.