The regional model calibration results (Table 5 and Figure 9) suggest that the AWRA-L model performs well in estimating high streamflow and low streamflow in the and its surrounding area when it is calibrated against in situ high streamflow and low streamflow, respectively.
The Nash–Sutcliffe efficiency (NSE) of daily streamflow obtained in this study is about 0.10 to 0.15 better than the predicted NSE obtained from traditional hydrological modelling carried out in south-east Australia (; ), which first calibrates a hydrological model against streamflow observation at each catchment, then regionalises the model parameters from a nearby catchment to a target ungauged catchment for streamflow prediction. The median Ed(0.1) is about 0.07 to 0.15 higher than the predicted NSE of log-transformed daily streamflow obtained in south-east Australia (; ).
It is noted that when the regional model is calibrated against observations from the 16 streamflow gauges it does not generate a uniform model performance. Though the AWRA-L model performs well overall, its performance is modest in some catchments. For instance, the high-streamflow model calibration reproduces daily streamflows poorly at catchment 210017 and the low-streamflow model calibration exhibits a poor model performance at catchment 210080 (Table 5). Both are tributaries of the Hunter River (Figure 8) and the model noticeably overestimates at the two catchments. For the four calibration catchments contributing to the Gloucester subregion (208005, 208020, 209002 and 209018), the model performs well in terms of model efficiency and shows a slight tendency toward overestimation.
A key characteristic of a regional calibration approach is that, unlike with local calibration, there is little degradation in prediction performance between model calibration and model prediction (; ). This means that prediction performance in calibration provides a good guide to the expected performance in ungauged parts of the modelling domain. In other words, it is reasonable to expect that at all the Ed(1.0) values will be of the order of 0.55 to 0.75 and the biases will be of the order of –0.17 to 0.22. This, therefore provides confidence in the prediction quality of the AWRA-L model outputs in each receptor location where there are no streamflow observations.
The results from the simulated (Figure 10) show that in the Gloucester subregion the AWRA-L model, calibrated against high streamflow, performs well for estimating hydrological response variables reflecting high-streamflow metrics, and the AWRA-L model, calibrated against low streamflow, performs well for estimating most hydrological response variables reflecting low-streamflow metrics (except for ).
Product Finalisation date
- 188.8.131.52 Methods
- 184.108.40.206 Review of existing models
- 220.127.116.11 Model development
- 18.104.22.168 Calibration
- 22.214.171.124 Uncertainty
- 126.96.36.199 Prediction
- Currency of scientific results
- Contributors to the Technical Programme
- About this technical product