Close Window

University of Virginia: "Probabilistic quantitative precipitation forecasting"

Final Report

By the end of 1990, the Pittsburgh WSFO had been preparing quantitative precipitation forecasts (QPF) for eight hydrologic river basins, twice a day for two years. The forecasts provided a single estimate of the 24-hour precipitation amount. However, the degree of confidence in these QPFs could not be conveyed by a single estimate. Consequently, when a QPF was input into a river forecast model, the resultant forecast of the river stage did not provide a sound basis for issuing flood watches and warnings because its reliability was unknown. Quantification of uncertainty in QPFs is of utmost importance so that the reliability of flood watches and warnings (defined by the probabilities of detection and false alarms) can be estimated in real time.

The probabilistic QPF system developed by the University of Virginia under this Cooperative Project consists of: 1) a forecasting methodology, 2) local climatological guidance (LCG), and 3) a verification method. The forecasting methodology produces a forecast of the areal average precipitation amount accumulated over a river basin during the 24 hours from the forecast time and a forecast of the temporal disaggregation of any 24-hour precipitation amount into four 6-hour precipitation amounts. The LCG is used to aid the forecaster in calibrating his/her judgment. For each basin and each month, the LCG provides: 1) climatological exceedance fractiles of the 24-hour basin average precipitation amount, and 2) climatological expected temporal disaggregation of the 24-hour precipitation amount. The verification of the probabilistic forecast consists of several statistics which compare the forecast against the actual 24-hour basin average precipitation amount and its temporal disaggregation. A subsequent Cooperative Project (begun January 1994) involves developing the method for incorporating probabilistic QPFs into a probabilistic river stage forecasting methodology.

Several training workshops for forecasters have been held to acquaint them with judgmental tools and estimation procedures used in probabilistic QPF. The computer software used in the forecasting is currently being tested. A prototype of the LCG system (which includes technical documentation, software, data files with historical records for two test river basins, user's manuals, and hard copies of main displays) is currently undergoing operational testing. Implementation of an integrated data base for verification has been completed, and forecasters have been trained in interpreting verification statistics. The potential use of the probabilistic QPF technique is being studied for its wider application in the NWS.