2.6.1.4 Calibration

Summary

This section summarises surface water model calibration for the Galilee subregion. The Australian Water Resources Assessment landscape model (AWRA-L) was regionally calibrated at eight unregulated catchments using two calibration schemes: one biased towards high streamflow and the other biased towards low streamflow.

The high-streamflow calibration yields relatively better (i.e. less biased) predictions for annual flow while the low-streamflow calibration yields better predictions for low-flow spells and zero-flow days. The two calibration sets yield predictions with similar levels of bias for most of the remaining hydrological response variables. One variable, daily flow at the 1st percentile (P01), is unable to be predicted by either calibration parameter set because the flow in most catchments is ephemeral so P01 takes a value of zero in most years at most sites.

A consequence of the regionalisation scheme used in this Assessment is that prediction performance in ungauged parts of the subregion is likely to be similar to that in the calibration catchments. This, therefore, provides confidence for applying the AWRA-L model to each model node where there are no streamflow observations. The performance of the calibrated parameters for estimating the hydrological response variables suggests the 10,000 model parameter sets generated using the calibrated parameters can produce a reasonable estimate range for the majority of hydrological response variables. However, both calibrations substantially under-predict the number of low-flow days. This suggests that less confidence may be ascribed to the prediction of this variable in Section 2.6.1.6 than to the prediction of the other variables.

In contrast to most standard surface water modelling approaches, detailed model calibration was not undertaken as part of the surface water modelling in the BA for the Galilee subregion (consistent with the overall BA approach outlined in companion submethodology M06 for surface water modelling (Viney, 2016)). This reflects the focus of the BA modelling on the difference between two possible futures (baseline and CRDP), rather than on making an absolute prediction, as well as the presentation of results within an uncertainty framework. The probabilistic focus means that the model parameters are varied over a wide range of plausible values (i.e. several orders of magnitude) in order to capture the uncertainty inherent in the system. The purpose of model calibration is therefore restricted to ensuring that the model is able to adequately represent the surface water system with optimal parameter values. However, these optimal parameter values are not used further in the modelling, so a detailed and time-consuming optimisation procedure (as commonly undertaken for deterministic modelling) is not followed in the BA. Instead, this calibration methodology means that results are reported for thousands of model runs that cover the range of plausible input parameter values (see Viney, 2016 for further details). This approach, which is not yet widely reported in relevant technical or scientific literature, is different from typical calibration methods used in surface water models which only report results for one optimal model run.

Last updated:
6 December 2018