Abstract
A freezing index was estimated from five kinds of probability distribution models. The results showed that it could not be represented by a only distribution model, but good matching was found to a lognormal distribution and a normal distribution. Thus, an estimation method using the largest 1/3 of the figures picked from the data was studied. The freezing index obtained from this method was compared with the actual value, and this demonstrated high accuracy. The effects of the number of data on freezing index estimation were studied. As a result, data for more than 25 years were found to be necessary to estimate the freezing index, because change in that for the period before the 1990s was very large.
