No abstract

Object-oriented verification methodology is becoming more and more common in the evaluation of model performance on high-resolution grids. The research herein describes an advanced version of an object-oriented approach that involves a combination of object identification on multiple scales with Procrustes shape analysis techniques. The multiscale object identification technique relies heavily on a novel Fourier transform approach to associate the signals within convection to different spatial scales. Other features of this new verification scheme include using a weighted cost function that can be user defined for object matching using different criteria, delineating objects that are more linear in character from those that are more cellular, and tagging object matches as hits, misses, or false alarms. Although the scheme contains a multiscale approach for identifying convective objects, standard minimum intensity and minimum size thresholds can be set when desirable. The method was tested as part of a spatial verification intercomparison experiment utilizing a combination of synthetic data and real cases from the Storm Prediction Center (SPC)/NSSL Weather Research and Forecasting (WRF) model Spring Program 2005. The resulting metrics, including error measures from differences in matched objects due to displacement, dilation, rotation, and intensity, from these cases run through this new, robust verification scheme are shown.

No abstract

No abstract