Improving WSR-88D Hourly Rainfall Estimates

Abstract

A statistical objective analysis (SOA) scheme is used to reanalyze the stage III estimate of rainfall, an hourly mosaic of digital precipitation arrays produced by a network of WSR-88Ds. The technique also uses rainfall measurements from the Oklahoma Mesonetwork that are taken over the Lake Altus area in southwest Oklahoma. The Lake Altus area is monitored by four WSR-88D radars: Frederick, Oklahoma; Twin Lakes, Oklahoma; Amarillo, Texas; and Lubbock, Texas. A total of 185 hourly maps of precipitation accumulation between June 1995 and July 1996 are used in the reanalysis. The results indicate that the stage III analysis underestimates total rainfall accumulations by as much as 40% when compared to the SOA reanalysis. Furthermore, the largest discrepancies between the stage III analysis and the SOA reanalysis coincide with overlapping areas of coverage between WSR-88D umbrellas. Some stage III precipitation fields used in this study clearly show fictitious high gradients of rainfall that exactly coincide with the maximum range ring of adjacent WSR-88Ds. Currently, stage II precipitation fields are mosaicked by averaging nonzero rainfall accumulations, regardless of their respective distance from a WSR-88D, to generate the stage III analysis. It is shown that wherever WSR-88D surveillance areas overlap, analysis errors, introduced solely by the radar-range effect, will adversely affect the accuracy of the stage III estimate. An error-weighted averaging method is proposed to eliminate this problem.