Discussion of ''Classical Extreme Value Model and Prediction of Extreme Winds,'' by Janos Galambos and Nicholas Macri

摘要

We would like to start our discussion by correcting three errors by Galambos and Macri (GM), of material from Simiu and Heckert (1996) (hereinafter referred to as SH).1. Galambos and Macri write: Simiu and Heckert ''report that the best fit for most stations is gamma=12, which would yield c=-1/gamma=-0.085 ... (see Table 1 in SH).'' In fact Table 1 in SH states that for most stations gamma <=12, i.e., c<=-0.085, not c=-0.085.2. More importantly, SH specifically explain that the samples with average 8-day epochs on which the estimates of Table 1 are based are "not likely to be a sound basis for inferences on extremes," and that it is therefore advisable to ''let the tails speak for themselves," i.e., to use the peaks over thresh-old (POT) approach. Winds that occur during 8-day epochs are, in their vast majority, different meteorologically from winds associated with extreme speeds. Analyzing samples that include both winds occurring in storms associated with extremes and winds such as the morning breeze is as likely to be misleading as analyzing samples of the height of the members of a kindergarten class that include the heights of both students and teachers. Differences between the estimates of Table 1 and POT estimates are therefore attributable to inadequacies of the samples on which Table 1 is based, rather than to the superiority of one estimator over another, as suggested by GM. Hence, the contradiction in the paper by SH to which GM allude does in fact not exist.3. GM state: ''On p. 542 the authors (i.e., SH) themselves point out that the data at nine stations do not support their recommended choice c=-0.25."