A Morphing-based Technique for the Verification of Precipitation Forecasts

The talk describes a morphing-based precipitation verification approach. This approach uses a pyramid matching algorithm to morph the forecast image into an image that resembles the analyzed (observed) precipitation field. The algorithm computes an optical flow (vector field) that maps the original forecast into the morphed forecast. The displacement error and the error in the structure of the forecast precipitation field are quantified based on the information provided by the optical flow. There are three novel aspects of the proposed approach compared to the morphing-based strategies published in the literature. First, it imposes a constraint on the pyramid matching algorithm to prevent over-convergence of the precipitation to a few locations of large errors during the morphing process. Second, it introduces an objective criteria for the selection of the sub-sampling parameter, which is the key free parameter of the pyramid matching algorithm. Third, the proposed definitions of the displacement error and the structure error are new. The performance of the proposed approach is demonstrated for both idealized examples and high-resolution numerical model forecasts of the precipitation for tropical cyclones and extratropical frontal systems.