Abstract

Growing population and socio-economic activities along with looming effects of climate change have led to enormous pressures on water resource systems. To diagnose and quantify potential vulnerabilities, effective tools are required to represent the interactions between limited water availability and competing water demands across a range of spatial and temporal scales. Despite significant progresses in integrated modeling of water resource systems, the majority of existing models are still unable to fully describe the contemplating dynamics within and between elements of water resource systems across all relevant scales and/or variables. Here, a data-driven approach is suggested to represent local details of a water resource system through emulating an existing water resource system model, in which these details have been missed. This is through advising a set of interconnected functional mappings, i.e. integrated emulators, parameterized using the simulation results of the existing model at a common scale and/or variable but can support process representation with finer resolution and/or details. The proposed approach is applied to a complex water resource system in Southern Alberta, Canada, to provide a detailed understanding of the system’s dynamics at the Oldman Reservoir, which is the key to provision of effective water resource management in this semi-arid and already stressed cold region. By proposing a rigorous setup/falsification procedure, a set of alternative hypotheses for emulators describing the local dynamics of local irrigation demand and withdrawals along with reservoir release and evaporation is developed. Findings show that emulators formed using Artificial Neural Networks mainly outperform simpler emulators developed for the variables considered. The non-falsified emulators are then coupled to represent the local dynamics of the water resource system at the reservoir location, considering the underlying interplays with hydro-climatological conditions and human decision on the irrigation area. It is found that emulators with input variables identified through expert knowledge can outperform fully data-driven emulators in which proxies were selected based on an input variable selection method. The top non-falsified coupled models are able to capture the dynamic of lake evaporation, water withdrawal, irrigation demand, reservoir release and storage with coefficient of determination of 0.80 to 0.82, 0.45 to 0.55, 0.52 to 0.59, 0.98 to 0.99 and 0.72 to 0.88, respectively. The practical utility of the proposed approach is demonstrated through an impact assessment study by analysing four performance criteria, corresponding to reservoir’s storage, local irrigation demand, number of spill events and median reservoir release, in three stress-tests. These stress tests asses the local sensitivity of water resource system at the Oldman reservoir at three different levels, corresponding to (1) changing incoming streamflow to the basin in a bottom-up approach; (2) joint scenario of changing streamflow and warming climate, using a coupled bottom-up/top-down approach; and (3) specific changes in incoming streamflow, climate and irrigation area in a heuristic approach. For the first experimentation, weekly realizations for possible water availability are stochastically reconstructed and fed into the top non-falsified integrated emulator. By defining warm/dry, historical and cold/wet flow conditions, we found through alteration from dry to wet regime condition, the expected number of low storage duration is not changed, and expected annual water deficit is declined. Moreover, the expected number of spill events increases whereas median reservoir release increases. In the next impact assessment study, different scenarios of warming climate obtained from NASA-NEX downscaled global climate projections and the joint impact of changing streamflow and temperature on the system’s behaviour is evaluated. This assessment demonstrated that in warmer climate, the expected number of low storage duration in dry condition increases whereas in historical and wet conditions, the low storage duration does not change. In addition, the expected annual water deficit increases while the expected number of spill events decreases in the three flow regime conditions. Moreover, the expected median reservoir release increases in the dry, historical and wet regime conditions. In the final level of assessment, vulnerability of the system under changing streamflow, climate including temperature and precipitation and changing irrigation area is assessed. Results show that increasing irrigation area combined with declining inflow can considerably increase the duration of low reservoir storage in the Oldman Reservoir. Increasing temperature can lead to decline in both reservoir storage and outflow. In addition, when combined with declining inflow, increasing temperature can severely increase the annual water deficit for irrigation sector. Furthermore, it is noted that although the performance of unfalsified models are identical in representing the dynamics of the Oldman Reservoir under the historical data, but assessment can be slightly to moderately different depending on the defined scenarios of change. This is due to the choice of model configuration and can address the uncertainty regarding the system’s behaviour. Our study shows the promise of data-driven emulation approach as a tool for developing more enhanced water resource system models to face emerging management problems in the era of change.