Climate and statistical models are often independently used to model those phenomena. To better assess performance of the climate models, a variety of spatial forecast verification methods have been developed. However, spatial verification metrics that are widely used in comparing mean states, in most cases, do not have an adequate theoretical justification to benchmark extreme weather events.
We proposed a new integrated threshold selection methodology for spatial forecast verification of extreme events that couples existing pattern recognition indices with high threshold choices. This integrated approach has three main steps: 1) dimension reduction; 2) geometric domain mapping; and 3) thresholds clustering.
We apply this approach to an observed precipitation dataset over CONUS. The results are evaluated by displaying threshold distribution seasonally, monthly and annually. The method offers user the flexibility of selecting a high threshold that is linked to desired geometrical properties.
The proposed high threshold methodology could either complement existing spatial verification methods, where threshold selection is arbitrary, or be directly applicable in extreme value theory.
Adviser: Xin-Zhong Liang