84th AMS Annual Meeting

Thursday, 15 January 2004: 2:30 PM
An Object Oriented Approach to the Verification of Quantitative Precipitation Forecasts: Part II—Examples
Room 6A
Michael Chapman, NCAR, Boulder, CO; and R. Bullock, B. G. Brown, C. A. Davis, K. W. Manning, R. Morss, and A. Takacs
Poster PDF (216.2 kB)
In Part I of this study, a new verification technique for Quantitative Precipitation Forecasts (QPFs) was introduced. This "object-oriented" method is used to identify regions of interest that would most likely be relevant to a human observer. These regions are first identified by the use of convolution with a shape, such as a cylinder. After the convolving takes place the field is thresholded. Thresholding helps to "smooth" out the boundaries around the area making the shape's appearance look more human-rendered. After the smoothing takes place separate objects may be "merged" if it is determined that the regions are related to one another. As the objects are identified, attributes of the forecast and observed objects, such as location, shape, orientation, and size, are compared.

In Part II of this study, a critical look at the performance of this technique is provided. The 22km Weather Research and Forecasting (WRF) model is used as the forecast data set, and it was restricted to areas within the borders of the CONUS. Stage 4 data are utilized as the observation data set. The higher resolution Stage 4 data were smoothed to allow a more valid comparison to the lower resolution WRF model output. The verification technique is applied to several years worth of data, generally in the summer months. Over the course of this analysis, the convolution radius and threshold were varied to assess the results of how each affects the sensibility of the technique. Impacts of the many changes are evaluated and compared. Several cases were identified for both favorable and non-favorable results. In some instances the objects identified by this technique were meteorologically sound, and an accurate matching of the forecast and observed regions by an objective means seems possible. However, for several cases a matching the forecast and observed regions was not possible. These unfavorable cases made evident the need for a more complex rule set for matching objects to be applied to the technique. Several examples will be presented to provide a more complete look at the effects changing the radius and threshold variables have on the quantity, shape, and character of the objects, as well as the basic problems and successes of this verification method.

A secondary study with different data sets is under way. The forecast data are the National Convective Weather Product (NCWF), the Auto Nowcaster, and 4km WRF data from the BAMEX project. These products are "regionalized" and should provide some valuable comparisons of this approach over a smaller spatial scale.

Supplementary URL: