Laserfiche WebLink
<br />Section B, Table 2, presents the results for the two <br />verification summary measures. Sizable forecaster skill is <br />suggested for each field season and the combined data (e.g., <br />TSS = .76 and CPOD = .70 for the combined seasons). <br />Comparing these results with those from concentration <br />suggests that onset is the easier forecasting problem. <br /> <br />C. Duration <br /> <br />The contingency table of counts and percentages for the <br />duration of liquid water on the barrier (i.e., length of <br />the storm period) is given in Table 3. About 70% of the <br />counts reside on the diagonal with the two extreme cells <br />greatly dominating the results (i.e., a sub-total of 68.8%). <br />The conditional POD's for these two categories are .86 and <br />.92 respectively whereas the other five categories have <br />values from .29 to .06. <br /> <br />The conditional bias plot is shown in Fig. 5, and one <br />now sees a clear tendency to under-forecast the longer <br />duration events. The overall, or unconditional, bias is <br />very small (i.e., < 1%). <br /> <br />Section B., Table 3, presents the results of the two <br />suw~ary meas~res for the duration forecasts, and one can see <br />that some skill is present in predicting the duration of <br />liquid water at KGV (e.g., TSS = .64 for the combined two <br />seasons). Comparing concentration, onset, and duration <br />predictions, we see that forecasters are best able to <br />predict onset, then d~ration, and finally concentration <br />(e.g., TSS = .76, .64 and about .60 respectively). The <br />presumed extrapolation mode of prediction for onset seems to <br />aid this result. <br /> <br />10 <br />