Laserfiche WebLink
<br />OOJ396 <br /> <br />The seven year (1979-1985) verification statistics for the Denver <br />F2-P2 are summarized in Figure 1. The verification presented is for <br />the prediction of excessive convective rainfall (ECR) days. An ECR <br />day is defined as a day (DOOO-2400L) on which at least one ECR event <br />occurs within District boundaries. An fCR event is the reported <br />occurrence of excessive rainfall, generally from thunderstorms, which <br />causes flash flooding, stream flooding or street flooding. A correct <br />ECR day forecast required that all ECR events occur in areas that <br />received internal alerts. <br /> <br />FIGURE 1 <br />EXCESSIVE CONVECTIVE RAINFALL FORECAST <br />VERIFICATION 1979-1985 <br /> <br /> Forecast Not Forecast <br />Observed 194 3 FAR 21% <br /> Probabil ity of <br />Not 52 829 Detect i on 98% <br />Observed Accuracy 95% <br /> <br />.. The Denver F2-P2 statistics show an overall accuracy of 95 percent, <br />a false alarm rate of 21 percent and a probability of detection of 98 <br />percent. The probability of detection statistic measures the number of <br />days ECR events occurred in the District and were correctly predicted. <br />In the past six years only three events went unforecast, all in 1979. <br />Additionally lead times exceeded gO minutes for major events of 2 inches <br />or greater rainfall in all cases. Lead times were calculated from time <br />of forecast issuance to approximate time of event occurrence. These <br />results and the low false alarm rate has promoted a strong sense of <br />cooperation for HKA from its F2-P2 customers who believe in the forecasts <br />and use them to make operational decisions. <br /> <br />Table 1 shows the seasonal verification of the F2-P2 ECR day <br />forecasts. Verification for 1979 and 1980 was changed to conform to <br />more stringent verification rules imposed since 1981. In general the <br />false alarm rate has continued to be decreased while all but 3 of <br />197 event days were predicted in the seven year period. The percent of <br />"blanket" six county message decreased from about 75 percent in 1979/80 <br />to about 50 percent in 1984/85 period as expertise in storm-scale <br />prediction increased. It is doubtful the false alarm rate can be <br />dropped further without seeing a decrease in the probability of <br />detection. <br /> <br />-3- <br />