Abstract

Measures of Quantitative Precipitation Forecast (QPF) quality that have been in use up to now suffer from several well-known defects - among these is their inability to diagnose placement and timing errors. This paper discusses an "object-oriented" approach to QPF verification. In this technique,... Show moreMeasures of Quantitative Precipitation Forecast (QPF) quality that have been in use up to now suffer from several well-known defects - among these is their inability to diagnose placement and timing errors. This paper discusses an "object-oriented" approach to QPF verification. In this technique, the forecast and observed fields are broken up into regions of interest - "objects" - which have descriptive attributes and are considered as indivisible units. Differences in the forecast and observed fields can then be described in terms of differences in object attributes. Objects are resolved using a two-dimensional convolution operator. This process often results in shapes very similar to those a human would draw. After object detection, several attributes are calculated for each object, and distinct objects may be merged into one composite object. Object attributes include centroid coordinates, orientation angle, area, first and higher moments and curvature. Differences in any of these attributes between forecast and observed objects can be used to categorize forecast errors by type, e.g., location error, orientation error, etc. Forecast and observed objects can then be matched using various overlap or "nearness" criteria. A variation of this rule set can be used to track individual objects over time. This is somewhat complicated, however, by the fact that objects can split into pieces or even disappear. Future work will focus mainly on addressing these issues. Show less