長期流出タンクモデルの実用的同定法とその考察

Abstract

Several problems have been examined to propose the practical methods for applying themathematical optimization technique to the identification of the series tanks model with sixteenunknown parameters for the long term runoff analysis. The data used are the daily ones ofprecipitation and runoff for the period of 1973-1979 in the Eigenji Dam basin of 132 km2 and of1975-1976 in the Ichinotani basin of 0.13 km2. On the basis of numerical experiments, the SDFP(Standardized DFP) method has been chosen as the optimum identification technique to be usedhere under the Chi-Square criterion of errors to be minimized.The results of practical examination are summarized as follows:1) The water balance constraint is indispensable for identifying the model, by which goodaccuracy is expectable in the period of prediction, too.2) The data of drought years, even if their length is short, are effective for the model identificationof good accuracy.3) Excluding the data of runoff greater than a given level from the evaluation of errors is fairlyeffective if no data of drought period are available for the model identification.4) Permitting the one-day shift back of observed runoff is effective for improving the model, because the influence of night rainfall cannot be neglegible.5) Excluding the data of runoff in snow season from the error evaluation may be permittedbecausc of the uncertainty in the estimation of snowfall.