抄録
Outliers in flood frequency analysis have bothered us to examine the goodness-of-fit to the model of extremes. Too large value of the return period will be thought ridiculous even for the extraordinary magnitude of record maximum. We are just faced to the question: whether the outlier is a special case, or the model is not appropriate. The problem will arise often though the rest of data seems to be fitted very well. We, engineers have tried to do by employing numerous candidates of mathematical functions of population, though there is also the possibility of extrapolation in another point of view of this problem. This study proposes the method of drawing a window with the limits to view the data and models in the Gumbel plot. The outliers beyond the frame of window are inconclusive to judge the goodness-of-fit to the model. The window will release us from the fateful behavior of outliers in the extreme values, as well as the obsession of testing more than enough candidates of population, provided that we make the possible efforts by employing the special techniques, e.g. regional analysis etc. and continuing the observation ordinarily.
