INFORMATION COST FUNCTION USED FOR CALIBRATION DATA SELECTION OF RAINFALL-RUNOFF MODELING

摘要

The identification of rainfall-runoff models requires appropriate data to be selected for model calibration. Despite its importance, there is a lack of systematic study on quantitatively selecting the most appropriate data used for calibration and this problem will get worse as more rainfall and flow data are collected by modern telemetry systems. In this study, an entropy-like index, named the Information Cost Function (ICF) based on the discrete wavelet decomposition of flow time series is proposed for the selection of the most appropriate calibration data. With the validation data determined beforehand, we assume that the more similarity does the calibration data set bear to the validation set, the better performance should the calibrated model have. The ICF index can give a good estimation of the information quality of the flow time series. By simply comparing the ICF values of the calibration and validation data sets, the calibration data sets resulting in the best model performances can be successfully identified without taking the parameter optimization procedure. The results also prove the statement that the information quality of the calibration data actually plays the most significant role in determining the model performance, rather than the data length. The idea presented in this paper has shown its potential in enhancing the efficiency of the calibration data utilization, which is especially meaningful for data-limited catchments.