In stratified reservoirs, both dam tailwater discharge and thermal plant intake water quality and temperature can be highly dependent on structure depth. A twodimensional laterally-averaged model allows for better prediction of water quality over time at specific depths. Because high-fidelity models are typically too computationally
expensive for direct inclusion within optimization algorithms, water quality is incorporated using one dimensional models are simple flow requirements. Water quality predictions can be incorporated within the optimization process through using surrogate modeling methods, in this application artificial neural network (ANN) models. ANNs are flexible machine learning tools for function approximation composed of a structure of neurons assembled within a multi-layer architecture. They are capable of handling large amounts of training data and modeling nonlinear dynamic systems, making ANNs a well-suited method for this application. This report illustrates the development of ANN models to emulate the hydrodynamic and water quality modeling capabilities of the high-fidelity, two-dimensional CE-QUAL-W2 (W2) model, as well as a linked riverine reservoir system optimization process which accounts for energy generation, water balance and hydraulics, and compliance point water quality. A process for hourly hydropower generation planning is demonstrated on a pair of reservoirs linked in series. The two reservoirs are U.S. Army Corps of Engineers projects with hydropower capabilities on the Cumberland River near Nashville, Tennessee, USA. The content presented here is largely a combination of technical papers previously presented at the HydroVision International conference (Shaw et al., 2015, 2016).