Abstract:
|
Methods from spatial statistics, image analysis, and computer vision have been adapted and used for evaluating high-resolution weather forecast systems. While these techniques are useful, a different set of requirements also present challenges. In particular, it is often important to inform specifically about spatial alignment of features within the fields; features that continually change in time and have no easily identified "landmarks" that can be used, for example, in a shape analysis. This talk will show a new binary image measure designed for this purpose, which is compared against existing binary image measures. The new measure is shown to avoid many of the drawbacks faced with the existing measures.
|