A winter hydrometeor classification algorithm

Over the past two winter seasons, the National Severe Storms Laboratory (NSSL) has collected observations of precipitation type from anonymous, volunteer public observers though the Winter Precipitation Identification Near the Ground (W-PING) project. To date, nearly 3,000 observations of winter precipitation type have been logged through a web based form hosted by the NSSL. Analysis of these reports shows that the current hydrometeor classification algorithm (HCA), which was developed with warm season convection in mind, performs poorly when confronted with winter precipitation. This project proposes to use these data, along with new capabilities within the warning decision support system integrated information (WDSSII) system, to build a data driven winter surface precipitation classifier and implement it within WDSSII.

The WDSSII system has the ability to extract vertical profiles of environmental data from rapid update cycle (RUC) model grids at designated times and locations. In addition, these profiles may be tilted at angles from the vertical to take into account the fact that precipitation source and generating regions are upwind from the surface observation location. Environmental data is on a rectilinear grid, but WDSSII can now interpolate the rectilinear data onto the spherical grid used by radar data. Hence, the radar range gates and the environmental data can be colocated.

With these capabilities it is possible to construct a complete set of radar and model based environmental attributes associated with any given observation within the W-PING data base. With these capabilities also comes the ability to build a hybrid winter surface precipitation classifier using a mixture of rule-based and data driven, statistically principled methods. The algorithm will generate classifications for surface precipitation type only; it will not create classifications for precipitation aloft because there are no verification data aloft.

The resulting classifier is evaluated statistically through various performance scores such as POD, FAR and CSI along with various other skill scores, such as the Heidke and Peirce skill scores. As envisioned, the classifier will define at least four basic categories: liquid, freezing, frozen and none. The frozen category may be divided into two subcategories: “snow” and “not snow,” depending upon the quality and quantity of appropriate W-PING data.