Study On Automatic Optimization Of Hydrologic Models' Parameters

The parameters of hydrologiccal models, whether conceptual models or distributed models, are based on conceptual representations of the physical processes. However, due to difficulty in measuring the parameters directly, we can only depend on the system methods to determine them. The emphasis of this paper is automatic calibration of Hydrologic models.There are two kinds of methods to calibrate the parameters : manual calibration and traditional optimization algorithms. The manual calibration requires detailed understanding of the model, which can only be obtained through many calibration experiences. Moreover, this method is quitely subjective. Most of the traditional optimization algorithms are local optimization methods, so how to find a global optimization method becomes the hotspot of Hydrologic model calibration.Studies on calibrating the parameters of the Xinanjiang model, Sacramento model and the Topmodel model are made by using the genetic algorithm and SCE-UA(shuffle complex evolution ) mothod. It includes three main parts. Firstly, manual method is used to calibrate the Xinanjiang model, Sacramento model and Topmodels' parameters manually, compare the calculated results and the observed results, and determine a set of parameters. Secondly, the range of each parameter is given on the basis of the results by trial-and-error method and provided for the genetic algorithm and SCE-UA ways to calibrate the models. Finally, the comparison is made among the results of the three methods .The results show that the automatic calibration can conquer the disadvantages of the manual calibration, which are time-consuming and subjective. According to the determined coefficient of each flood, the precision of SCE-UA is the highest among the three methods. The precision of the genetic algorithm Method is second and the presicion of the manual calibration is lowest. The parameters gained by the genetic algorithm and SCE-UA methods can accord with the physical meaning well, and have instructional meaning.