Laserfiche WebLink
<br />Table 1, Panel B, presents the results for two summary <br />verification measures (i.e., TSS and CPOD) for each of the <br />six 2 hour time blocks, each season separately and both <br />seasons combined (number of forecasts are shown in <br />parenthesis). We see clear evidence of the forecasting <br />skill both in the TSS and the CPOD (remember pure guessing <br />would produce a CPOD = .33). However, as expected the <br />values of both summary measures decrease with time from the <br />forecast valid time (0900 or 2100). Interestingly, the TSS <br />(the preferred and more sensitive measure of forecasting <br />ability) shows a rather sharp drop between the 2nd and 3rd <br />two hour period (e.g., .61 to .53 for the TSS in the <br />cor.~ined data). Furthermore, both seasons show evidence of <br />this sharp degradation in performance. Perhaps, f':; ~ ...,.s.f~ <br />extrapol3t~on largely is the mode of forecasting up to four <br />hours, and then a more detailed understanding of the <br />atmosphere is needed to successfully predict future <br />conditions. <br /> <br />B. Onset <br /> <br />The results from forecasting the onset time of SLW are <br />displayed in Table 2. In Panel A we see that 80% of the <br />predictions matched the observations. However, the first <br />two diagonal cells dominate this picture (i.e., 76%), and <br />thus extrapolation again seems to be the most successful <br />mode of prediction for the near-time. Figure 4 presents a <br />plot of the estimated conditional bias for this onset data <br />based on the predicted and observed percentages in Table 2 <br />(note that conditional bias = predicted percentage minus <br />cbse~ed perce~tage, Flueck, 1988). We see that t~ere is a <br />sligh~ tendency for conditional bias to increase as one <br />predicts for later time blocks but it always is less than <br /> <br />3.5%. <br /> <br />9 <br />