Verification of precipitation in weather systems: determination of systematic errors

Abstract

An object-oriented verification procedure is presented for gridded quantitative precipitation forecasts (QPFs). It is carried out within the framework of “contiguous rain areas” (CRAs), whereby a weather system is defined as a region bounded by a user-specified isopleth of precipitation in the union of the forecast and observed rain fields. The horizontal displacement of the forecast is determined by translating the forecast rain field until the total squared difference between the observed and forecast fields is minimized. This allows a decomposition of total error into components due to: (a) location; (b) rain volume and (c) pattern. Results are first presented for a Monte Carlo simulation of 40,000 synthetic CRAs in order to determine the accuracy of the verification procedure when the rain systems are only partially observed due to the presence of domain boundaries. Verification is then carried out for operational 24-h forecasts from the Australian Bureau of Meteorology LAPS numerical weather prediction model over a four-year period. Forty-five percent of all rain events were well forecast by the model, with small location and intensity errors. Location error was generally the dominant source of QPF error, with the directions of most frequent displacement varying by region. Forty-five percent of extreme rainfall events (>100 mm d −1) were well forecast, but in this case the model's underestimation of rain intensity was the most frequent source of error.