Sensitivity of WSR-88D Rainfall Estimates to the Rain-Rate Threshold and Rain Gauge Adjustment: A Flash Flood Case Study

Abstract

A strong thunderstorm produced a flash flood on the evening of 12 July 1996 in Buffalo Creek, Colorado, that caused two deaths and significant property damage. Most of the rain fell in a 1-h time period from 2000 to 2100 MDT. The performance of the WSR-88D rainfall algorithm, Precipitation Processing System, was examined in detail to determine how well it performed. In particular the sensitivity to the algorithm?s rain-rate threshold (hail cap) parameter and the performance of the gauge?radar adjustment subalgorithm on the resulting radar rainfall estimates were examined by comparison with available rain gauge data. It was determined that the WSR-88D rainfall algorithm overestimated the rainfall in general over the radar scanning domain for this event by about 60% relative to the rain gauges although the radar-derived rainfall for the flood-producing storm cell nearly matched the single gauge that sampled it. The derived rainfall over the radar scanning domain was not very sensitive to the setting of the rain-rate threshold parameter. Lowering it reduced the overestimation in general but did not bring the estimates satisfactorily close to unbiasedness. Other error sources were suggested, including use of an inappropriate Z?R relationship and/or radar reflectivity miscalibration. Relative importance of these sources could not be determined. The portion of the rainfall algorithm that adjusts the radar estimates using rain gauge data was tested to determine if it could have satisfactorily reduced the observed overestimation. It was found to have performed suboptimally due primarily to the methodology in the algorithm that forms the gauge?radar pairs. A simpler technique was proposed and tested, and the algorithm?s performance was greatly enhanced as a result. Therefore the performance of the gauge?radar adjustment algorithm depends on the gauge?radar rainfall data that are passed to it, and that data are dependent on the method by which the pairs are formed.