Improvement of Hydrologic Model Parameter Estimation Using Hydrograph Section Separation and Uncertainty Analysis

Abstract

ABSTRACT

Improvement of Hydrologic Model Parameter Estimation Using Hydrograph Section Separation and Uncertainty Analysis

by Kim, Chung-Soo Department of Civil and Environmental Engineering Seoul National University Prof. Kim, Young-Oh, Advisor

It is well known that parameter estimation in model simulation is an essential procedure that significantly affects simulation outputs. Nevertheless, there has not been a standardized methodology and procedure for parameter estimation. It is difficult to select and present the most optimal technique and scenarios for parameter estimation because of the uncertainties and variability in parameters. For these reasons, parameter estimation depends on a selected model, rainfall event, or a model user. For estimating the parameters of a rainfall-runoff model, the problem that is of primary concern is to find the optimal solution accurately, but it is difficult to find the solution because an objective function would not be defined clearly or estimated parameters that use only a few number of rainfall events are not reliable. Therefore, a generalization of parameter estimation procedure is necessary, and it should be able to consider the correlation of parameters and provide constant results without reference to model users and used models. In addition, the generalized procedure should consider the different characteristics of each rainfall event, such as the size of rainfall and the features of runoff, because the response modality of each model to each rainfall event is quite distinguished. First, the existing parameter estimation used in practice was examined. In Korea, the initial value of parameters estimated by a trial-and error method based on each event for historical data has been used for flood forecasting. By examining the existing approach, the necessity of an improved parameter estimation method that could secure reliability in flood forecasting was raised. In this study, a problem in parameter estimation for flood forecasting in Korea has been determined. In practical works, a conventional trial-and error method has been used to estimate parameters based on historical data, and they are applied to a model as initial inputs for flood forecasting. The parameter estimation procedure urgently needs to be improved to enhance reliability in flood forecasting. In order to resolve the problems, this study examined the efficiency in parameter estimation by using time-series input data rather than focusing on each event narrowly. Moreover, depending on a trial-and-error method was avoided, and instead, Shuffled Complex Evolution Metropolis of the University Arizona (SCE-UA), a global optimization method that provides the optimal parameter by estimating parameter sets, was used, and its results were analyzed and compared. In particular, a hydrograph analysis that could reflect the characteristics of runoff as parameter estimation was proposed. For a more accurate and reliable parameter estimation, a hydrograph was divided into three sections at each inflection point: rising limb, crest, and falling limb. A proposed method in this study aims to provide a solution to a problem that has fluctuated in parameter estimation in accordance with the characteristics of rainfall events and runoff. In addition, Generalized Likelihood Uncertainty Estimators (GLUE), an uncertainty analysis method, was applied to present a range of parameters, and it was expected to contribute to the improvement of estimation accuracy and to the enhancement of the applicability of estimated parameters. The ultimate objective of this study is to improve the parameter estimation procedure, and the proposed methodology shows finding the optimal parameter that can satisfy the objectives of both peak discharge and discharge volume at the same time in consideration of the size of runoff and the discharge moving behavior.

Keywords: parameter estimation, global optimization, SCE-UA, hydrograph section separation, GLUE, optimal parameter Student Number: 2005-30253