Calibration of rainfall runoff models in ungauged catchments: Regionalization relationships for a rainfall runoff model

Abstract

In many regions where rainfall runoff models are required, there is a lack of streamflow data available to calibrate the model parameters. Along with streamflow data simply not being recorded, there are many reasons for a lack of a suitable data set for model calibration, such as significant modifications to catchment characteristics, or long periods of unseasonable rainfall producing unrepresentative relationships. Generally, for an ungauged catchment, it is desirable to implement a model with as few free parameters as possible, provided the conceptualization of the model is suitable for the catchment under consideration. The Australian Water Balance Model (AWBM) is a rainfall runoff model that is commonly used in Australia. Typically it has 7 parameters, however methods are available to determine the storage size and capacity parameters for the AWBM based on an estimate of the average annual runoff. This is a great advantage when applying the model to ungauged catchments. However, there are two AWBM parameters which cannot be determined by this approach, namely the baseflow index and baseflow recession constant. The aim of this paper is to develop regionalization relationships to allow these two parameters to be estimated for ungauged catchments, based on the characteristics of the catchment that can be easily identified. General Regression Neural Networks are used to identify the relationships between model parameters and catchment characteristics. The results indicate that by using only easily identifiable characteristics of an ungauged catchment, suitable estimates of the unknown AWBM parameter values can be obtained, thereby allowing reasonable rainfall-runoff models to be developed. While the relationships developed in this work are specific to Australian catchments, the methodology used can be easily adapted to develop relationships for other regions.