Relating catchment attributes to parameters of a salt and water balance model

Abstract

Salinity is recognised as a global land management issue and the use of appropriate models is vital for the management of salinity-affected areas. A major limitation associated with the modelling of salt and water transport is the heavy reliance on good quality data for model development. Measured salinity data are required for calibration of salt and water balance (SAWB) models; however these data are not available for ungauged catchments. Hence the determination of optimal salinity management strategies for these areas is difficult. A considerable amount of research has been conducted on the development of hydrological models for ungauged catchments; however a similar approach has not yet been developed for SAWB models. In this study Partial Mutual Information (PMI) is used to assess the strength of relationships between readily-obtainable catchment characteristics and the parameters of a SAWB model. This will enable the future development of SAWB modelling for ungauged catchments by eliminating the traditional calibration requirement. In order to use the PMI to assess the strength of these relationships, a set of optimal SAWB model parameters and a set of catchment characteristics are required. The optimal set of model parameters is obtained by calibrating the model for 43 gauged catchments across Australia. CATSALT is chosen as the most appropriate SAWB model for this study due to its parsimony and reliability compared with other commonly available models. All of the inputs for CATSALT are obtained for the 43 catchments, which include values of average soil and groundwater salinity, streamflow and baseflow data. The Australian Water Balance Model (AWBM) is used to obtain the streamflow and baseflow data from measured total runoff data. Calibration of CATSALT for the 43catchments is performed using Differential Evolution, as this has been found to perform favourably compared with other calibration methods. Values for a set of 32 commonly available catchment characteristics are also obtained, including land use, vegetation and spatial attributes of the catchments. After obtaining the set of optimal SAWB model parameters by calibration and the set of catchment characteristics, the PMI algorithm is used. PMI is a technique for input variable selection that can detect linear and non-linear relationships between variables, as well as account for redundancy between variables. It is an improvement on traditional input selection methods, such as partial correlation analysis, which can only detect linear relationships between the variables. Using a bootstrapping procedure with 95% confidence limit as the stopping criterion for the PMI algorithm, six relevant and non-redundant catchment characteristics are found to have a significant relationship with each of the three CATSALT model parameters. This shows that there are relationships between easily obtainable catchment characteristics and the parameters of the CATSALT model. These catchment characteristics could be used in future research to develop models for the prediction of the CATSALT parameters, hence enabling CATSALT to be applied in ungauged catchments. This approach is not limited to the CATSALT model and could be applied effectively to other available SAWB models.