Evaluating spatial quantitative precipitation forecasts in the form of binary images

Development of alternative verification approaches for spatial quantitative precipitation forecasts is a rapidly growing area of research. There is a strong need for information that is meaningful from an operational context that can be derived from traditional gridbased verification approaches and that can capture the specific sources of forecast error. Many new methods have been proposed for evaluating the results of these forecasts.

One promising approach is to define objects that represent meaningful areas of forecast and observed precipitation. These objects can be represented as binary images (with the actual intensity values also preserved for later analysis). The challenge is then reduced to comparing two sets of binary images in an intelligent and meaningful way. Here, a new technique is proposed to both merge objects within a forecast (or analysis) image, and then match the objects from the forecast image to the analysis image in order to subsequently verify the forecast image. The method for merging, matching and verifying makes extensive use of an image comparison metric instigated by Baddeley (1992). This approach is demonstrated using precipitation forecasts from the Weather Research and Forecasting (WRF) model, In addition, the method is compared to other recently proposed methods for matching and comparing such forecast images to analysis images (e.g., a recently developed fuzzy logic approach).