Non data-driven reservoir outflow and storage simulations in hydrological models
by J. de Vos
In the last half century, many dams have been constructed on a global scale to help people get control over water resources. The effects of the reservoirs that are impounded after construction of the dams are that downstream river discharges cannot be classified as natural flows any more. Different water related interests make that many dams are operated in order to contribute to one or more operational goals such as irrigation water supply, hydropower generation and flood control. Data on individual reservoirs are difficult to obtain, especially operational rules and historical time series on inflow, storage and outflow. The consequence is that exact reservoir behaviour is often hard to simulate. Simulation of reservoir operations is important to investigate the effects of climate change and changes in human water demands on water availability in the future. There are multiple reservoir simulation models available that do their own assumptions in order to simplify reservoir outflow and storage simulations in the best possible way.
In this study, a literature review is performed first to find out how several existing hydrological models take into account reservoirs in terms of outflow and storage simulations. Furthermore it is investigated how easy-to-obtain parameters and datasets can contribute to reservoir simulations. The sensitivity and performance of the existing reservoir modules in hydrological models with low data demands are tested for a sample of sixteen reservoirs for which data is available. The time series data of the sixteen reservoirs in combination with a reservoir property named the Impoundment Ratio and a drought index named the Standardized Precipitation Evapotranspiration Index are used to develop a new reservoir simulation model. This newly developed model is also subjected to a sensitivity and performance test. The sample of tested reservoirs includes reservoirs in the USA, Central Asia and Southeast Asia. The sensitivity and performance assessments are based on two goodness of fit parameters, the normalized root-mean-square error (NRMSE) and the coefficient of determination (r2), and a visual inspection of results.
Overall performance of all tested reservoir simulation models is bad to moderate in terms of NRMSE and r2. It seems that a target release approach implemented in the Soil and Water Assessment Tool performs generally worst for simulation of both discharge and storage. All other tested models perform better, however it is difficult to observe simulations that are very distinctive in a positive way. A natural lake outflow scheme seems to be the best choice for over-year American reservoirs. Regarding storage simulation it is often a reservoir scheme designed by Hanasaki et al. (2006) or the newly developed reservoir simulation model that shows the best results, since they take into account over-year storage fluctuations. The natural lake outflow scheme and the scheme implemented in the Soil and Water Assessment Tool generally perform worse because they are not designed to simulate over-year fluctuations. For the within-year reservoirs in Central and Southeast Asia, it is often least bad not to model individual reservoirs and use actual inflow as a proxy for outflow. Within-year storage simulation is often best represented by the newly developed reservoir simulation model. Over-year storage fluctuations for two over-year Asian reservoirs are best represented by the newly developed reservoir simulation model or a reservoir scheme designed by Hanasaki et al. (2006).
This study shows that reservoir modules with low data requirements are not always significantly beneficial for the simulation individual reservoirs. The Impoundment Ratio and the Standardized Precipitation Evapotranspiration Index are considered as useful data for the simplification of reservoir operation simulations. For further research, efforts have to be made to retrieve more time series data so that the reservoir modules can be tested for multiple situations. It is also advised to build all reservoir modules in one hydrological model so that error propagation analyses can be performed for reservoir systems that are connected in series. Finally, further research should be done on worldwide representation of reservoir water demands since it affects the outflow distribution of relatively large reservoirs and many models need water demand as model input.
Student:
Joris de Vos
Committee: Prof. dr. ir. N. C. van de Giesen (TU Delft), Dr.ir. M.M. Rutten (TU Delft), Dr.ir. P. Droogers (FutureWater), Dr.ir. C.J. Sloff (TU Delft)