Case study on the use of dynamically downscaled climate model data for assessing water security in the Lower Hunter region of the eastern seaboard of Australia

A key aim of the Eastern Seaboard Climate Change Initiative (ESCCI) is under-standing the effect of climate change on the eastern seaboard of Australia, and the implications for climate change adaptation in this area. The New South Wales (NSW) / Australian Capital Territory (ACT) Regional Climate Modelling project (NARCliM) has produced three dynamically downscaled reanalysis climate datasets along with 12 downscaled general circulation model (GCM) projections of current (1990–2009) and future climate. It is expected that the NARCliM dataset will be used for many climate change impact studies including water security assessment. Therefore, in this study we perform a case study investigation into the usefulness and limitations of using NARCliM data for water security assessment, using the Lower Hunter urban water supply system managed by Hunter Water Corporation. We compare streamflow and reservoir levels simulated using NARCliM rainfall and a gridded historical rainfall dataset (AWAP) and focus our analysis on the differences in the simulated streamflow and reservoir levels. We show that when raw (i.e. not bias-corrected) NARCliM rainfall and potential evapotranspiration (PET) data is used to simulate streamflow and reservoir storage levels, some of the NARCliM datasets produce unrealistic results when compared with the simulations using AWAP; for example, some NARCliM datasets simulate reservoirs at or near empty while the AWAP reservoir simulations rarely drop below 60%. The bias-corrected NARCliM rainfall (corrected to AWAP) produces estimates of streamflow and reservoir levels that have a closer, but still inconsistent, match with the streamflow and reservoir levels simulated using AWAP directly. The inconsistency between the simulations using bias-corrected rainfall and historical AWAP rainfall is potentially because while bias-correction reduces systematic deviations it does not fix temporal rainfall sequencing issues. Additionally, the NARCliM PET is not bias-corrected and using bias-corrected rainfall with uncorrected PET in hydrological models results in physical inconsistencies in the rainfall-PET relationship and simulated streamflow. We demonstrate that rainfall plays a large role in the streamflow simulations, while PET seems to play a large role in the reasonableness of the simulated reservoir dynamics by determining the evaporation losses from the reservoirs. The downscaled GCM datasets that simulate the greatest average PET for 1990–2009 show reservoirs often (unrealistically) near empty. This study highlights the need to assess the validity of all climate data for the applications required, with a focus on long-term statistics for reservoir modelling and ensuring realism and coherence across all projected variables.