Long-term Reservoir Routing Simulations Using Data-Driven Approaches

Monday, 14 December 2015
Poster Hall (Moscone South)
Hamed Ashouri1, Hemant Chowdhary1, Kiran Chinnayakanahalli1 and Boyko Dodov2, (1)Research, AIR Worldwide, Boston, MA, United States, (2)AIR Worldwide Boston, Boston, MA, United States
Flood is a highly complex natural hazard that accounts for major losses to human societies worldwide. Dams built with the aim of mitigating the flood risk significantly modify river flow regimes but unavailability and/or inaccessibility of proper information about reservoir operational rules impose a big hurdle to global flood modeling. This is specifically critical for flood-prone regions where lack of proper representation of reservoir operation can lead to significant under- or overestimation of the flood magnitude, risk, and losses. With the availability of longer in-situ observational data records, as well as advancements in satellite altimetry techniques for measuring reservoir levels, operational rules can be indirectly deduced. In this study, the observed reservoir levels as well as the historical and forecast time series of inflows are incorporated into a stochastic autoregressive moving average statistical modeling scheme to simulate the releases from the dam at each time step. The resulting operational rule curve is used in a reservoir simulation model to simulate the outflows from the reservoirs. The efficiency of the model is examined for three case studies in the United States, including John Martin Reservoir (CO), Coralville Lake (IA, and specifically for the devastating 2008 flood in the state), and Boca Reservoir (CA). Statistical measures are derived and tested to evaluate the accuracy of the simulated hydrographs against USGS streamflow gauge observations. The results prove the capability of the developed model in simulating reasonably accurate outflows from dams and will be presented at the meeting.