Numerical Weather Prediction (NWP) forecasts of precipitation are noncontinuous (i.e., clustered or featured) fields. Therefore, their verification must be done in a feature oriented framework. Meanwhile, Cluster Analysis (CA) is aimed at identifying clusters or features in any data. As such, it is natural to utilize CA for precipitation verification. The objectives of this research are to develop 1) a CA methodology to identify clusters/features in spatial forecast and observation fields, taking into account multiple scales and attributes, and 2) a verification methodology for automatically matching the observed and forecast features.
Methods: To devise CA methods suited to the identification of the types of clusters/features appearing in meteorological fields (e.g., precipitation), model-based CA will be examined. The matching (i.e., verifying) of the clusters identified in the forecast field to those in the observation field will be performed within the framework of CA by computing different measures of the distance between the clusters. These methods will be compared with more conventional approaches, and to a "hand-verified" benchmark. The methodology will be tested on a series of mesoscale NWP precipitation forecasts acquired from the University of Washington mesoscale short-range ensemble system, and from the an advanced Department of Defense numerical mesoscale ensemble system.
Broader Impacts The Applied Physics Laboratory and the department of statistics at the University of Washington are partners in a 5-year Multi University Research Initiative (MURI). This has provided for a ripe environment for exploring all matters dealing with verification, and has attracted a wide range of students interested in statistical aspects of weather forecasting. As such, the research will promote teaching, training, and learning through involvement and participation of a graduate student focusing in statistical aspects of weather forecasting. The research will also enhance infrastructure for research and education by developing and making freely available a web-based verification system for assessment of mesoscale numerical prediction systems that can be easily utilized by students, forecasters and model developers.
