Variability in calibration and validation data lengths in relation to obtaining the best parameter set of a hydrological model

The success of a hydrological model is primarily governed by the model parameters that are generally obtained through calibration and validation of the model against historical rainfall and runoff data. Therefore, calibration and validation are critical steps in hydrological modelling, which assist minimisation of uncertainty in runoff estimates. Calibration and validation of a hydrological model depends on various factors such as availability of necessary data, length and quality of the data and choice of optimisation technique. In this study, impact of data length to obtain the best set of calibrated parameters of a hydrological model, known as Australian Water Balance Model (AWBM) is investigated. Several combinations of calibration and validation data lengths have been considered in the AWBM modeling to calibrate the model for four Australian catchments. Four different tests based on the calibration and validation data lengths have been applied for each site to evaluate the model performance in simulating the observed runoff time series. Each test has four different combinations of data periods such as, "peak", "wet", "dry", and "normal". A total of sixteen independent comparisons have been adopted to address the model calibration and validation performance of the AWBM model to identify the best set of parameters and to evaluate the impacts of the variations in data length on the model output. The results show that the validation of the AWBM model for dry periods are unsatisfactory, and gives poor outcomes in both the calibration and validation periods as NSE values are found to be less than 0 in most of the cases. The better NSE values for the calibration and validation of the AWBM model are obtained when "peak" and "wet" periods are used as calibration data element. The results of the study indicate that a notable degree of uncertainty can be resulted into the runoff estimates by the AWBM model if the appropriate calibration and validation data elements have not been selected.