A combination of field monitoring and numerical modelling systems are increasingly applied by water resource managers to provide information to inform their management activities. Determination of the optimal configuration of monitoring locations in water supply reservoirs is challenging and there is no generally accepted method for designing and assessing these monitoring systems. With the increased application of complex physically-based models that simulate three-dimensional (3D) coupled hydrodynamic and biogeochemical processes there is need to develop improved methods for optimising the design of monitoring systems to better support this monitoring-modelling approach. This project contributes to this area of research through the presentation of a novel method that applies numerical models to explore the influence of different numbers and spatial locations of coupled meteorological and water quality monitoring systems on the performance of water quality simulation outcomes.
The numerical models used in this project were informed by field measurements from an Australian subtropical water supply reservoir, Little Nerang Dam (LND). Under Australian subtropical settings reports of physical limnological investigations of water supply reservoirs are rare. This study presents the first comprehensive multi-season field measurements of physical and chemical limnology of LND and more broadly contributes much needed information on the limnology of a typical subtropical reservoir. Results show that LND is a warm monomictic reservoir, which occasionally exhibits meromictic tendencies. Thermal stratification and periodic wind forcing generate internal waves which influence mixing in the deeper zones of the reservoir and thus influence water quality. Major inflow-outflow events of short duration were found to cause sudden alternations of the thermal stratification and water quality. The influence of these events are thought to be as important in the management of the water resource as the seasonal thermal stratification and associated internal waves on water quality. These findings greatly enhance the current conceptual model of the processes that influence water quality in these subtropical systems.
The ELCOM-CAEDYM modelling platform was applied to simulate coupled 3D hydrodynamic and biogeochemical processes in LND. This model was setup and calibrated based on the field monitoring data from LND and represents the first 3D coupled hydrodynamic and biogeochemical model developed for this reservoir. The model was able to realistically simulate the annual cycling of temperature and dissolved oxygen (DO) selected state variables as observed in LND. The novel monitoring system optimisation method that was developed and applied in the project utilised the ELCOM-CAEDYM model for LND. While the approach could have been developed using synthetic data, the LND model was used to ensure realistic events were captured. Simulations were modified by introducing a synthetic set of meteorological forcing fields (created by interpolation of data from multiple points within a regional numerical weather simulation model) that allowed heterogeneous meteorological conditions to be applied. This allowed 95 meteorological stations and 271 initial condition profiles to be applied to create a "benchmark model" (BM). The BM forms a system where data can be sampled at any spatial location and at any frequency (though limited to the grid cell size and time step of the BM). Via a simple root mean square error (RMSE) method outcomes from the BM simulations for water temperature and DO are compared to those from less complex "engineering models" (EM). The EMs differ from the BM in that they only use a limited number (progressively increased from 1-5) of initial condition profiles and meteorological forcing data which are sampled from the BM simulation results or BM meteorological forcing field. In this way the EMs follow the approach used in most numerical modelling applications where monitoring information is limited to a few observation points.
The RMSE comparison between the BM and EM simulation results were used to identify locations of greatest error in simulation results. The EM was then re-run with an additional initial condition profile and meteorological forcing station (sampled from the BM simulation data) at the locations of greatest error and simulation results then re-evaluate to identify changes in simulation performance. This approach was also compared with EM simulation results that used either expert opinion or a pseudo-random approach to locate a progressively increasing number of monitoring points (initial condition profiles and meteorological forcing locations).
Results from all simulations (i.e., BM-EM approach, expert opinion and pseudo-random) showed simulation performance improved with an increasing number of monitoring stations. This improvement was exponential with the addition of monitoring stations most beneficial for simulation performance for low numbers of monitoring stations and the amount of improvement decreased as the number of stations increased. The results also showed that the BM-EM approach was capable of spatially distributing a progressively increasing number of monitoring stations so the RMSE systematically decreased and in some instances resulted in better performance compared to the expert opinion and pseudo-random approaches. The BM-EM approach shows potential for assisting in the determination of the number and spatial distribution of monitoring stations within a water reservoir so simulation performance improves.
While the BM-EM approach developed in this project shows promise, further investigation is required before the method can be readily applied in water resource management applications. In particular there is a need for exploration of the influence of additional monitoring station data (i.e., > 5 addition locations) as well as investigation of the validity of the method for other state variable (i.e., beyond water temperature and DO that were the focus of this project). The method should also be applied and tested on other systems. In the longer term experimentation with simulation-driven adaptive placement of monitoring systems on real-world reservoir systems would also be beneficial.
As part of the development of the BM-EM approach a search for a useful measure to quantify the potential for mixing across the thermocline and related water quality changes due to the influence of physical processes was conducted. It was envisaged that such measure could then be used to spatially locate monitoring stations. A critical exploration of the Lake number (LN) was conducted and results concluded that, for the application of the BM-EM approach, the LN was not the most useful measure, due to the presence of relatively large, short term variations in the LN. Instead the water temperature difference between the epilimnion and hypolimnion was found to be more suitable for this purpose. The outcomes of this component of the research project have implications for future application of the LN with a number of issues and suggested improvements to the methodology for estimating the LN identified.