River Water Balance Accounting to Evaluate Model Adequacy and Uncertainty in Climate and Development Scenario Assessment

The climate and development scenarios assessed in the Murray-Darling Basin Sustainable Yields project largely made use of existing river models. Some effort was made to assess the adequacy of the models for the project and to identify major sources of uncertainty, both in the scenarios assumed (external uncertainty) and in the river models used to estimate the resulting hydrological changes (internal uncertainty). The assessment addressed data availability, prior model review and testing, and new analysis. In particular, monthly river water balance terms simulated by the baseline river model (representing past climate and current development) were compared reach by reach with largely independent river water balance accounts. Monthly river water balance accounts were developed for the period 1990-2006 wherever possible, by integrating direct streamflow measurements with diversion records, remote sensing based estimates of wetland/floodplain and irrigation water use, and ungauged inflows estimated by the SIMHYD rainfall-runoff model. Statistical and visual data exploration helped to attribute apparent ungauged gains and losses to these terms, with the remainder consisting of unattributed gains and losses or measurement noise. Water accounts were developed for 14 out of 18 regions at the time of writing. These results indicate that on average, about 75% of inflows, outflows, gains and losses is gauged, and about half of the remaining water balance can be accounted for with additional data and modelling. However, large unattributed losses and noise remain, amounting to about 12% of the water balance on average. Among the largest uncertainties in the river water balance were ungauged losses in the lower part of river systems associated with ungauged distributaries, floodplain and wetland inflows, gauge bypass flows and flow harvesting. Remote sensing provides valuable information on the overall magnitude and spatial distribution of these losses, but not how they translate into monthly losses from the main river channel. Generally a major fraction of total inflows was gauged before entering the section of rivers where diversions, regulation and floodplain and wetland losses occur (though there were exceptions). Exchanges of water between the river and groundwater systems could not be estimated and were therefore not included in accounts. Linked river-groundwater modelling suggested that these exchanges are mostly modest. Some common themes emerged when evaluating the implications for the use of the projected changes in river flow regime. For a given change in river inflows, uncertainty in consequent river flows increased towards the end of the system. River models generally performed well in reproducing medium and high flows, but typically less well in simulating low flows. The necessary use of empirical loss functions in river models contributed to uncertainty. Differences between current or potential diversions (assumed in the baseline model) and historic diversions led to discrepancies with the water accounts, with implications for the interpretation of modelling results. The implications for future water accounting efforts are briefly discussed.