A Reduced Order Model for Understanding Hydrograph Response Downstream of the Flaming Gorge Dam

Summer flow regime recommendations in the Green R. downstream of the Flaming Gorge Dam (FGD) are in place to protect spawning habitats of endangered species' including the Colorado Pikeminnow. The ecological merit of these recommendations can be evaluated with streamflow forecasting tools and serve as an important asset for ensuring future compliance in real time. While physics-based hydrodynamic models accurately assess flow, intense computational requirements and associated costs of assimilating irrigation, climate, groundwater, and ungauged tributaries make lower-cost methods worthy of investigation. This analysis develops a methodology for joint estimation of water travel times from the FGD and Yampa R. tributary to the regulated reach near Jensen, UT. A convolutional neural network (CNN) predicts flow near Jensen using signals produced from discharge-variant hydrograph shifts. Hydrologic records are retrieved from the USGS's stream-gaging program below the dam (USGS 09234500 Greendale, UT), before the Yampa-Green R. junction (USGS 09260050 Deerlodge, CO), and in the Jensen reach (USGS 09261000).

The CNN trained on discharge-dependent lags of upstream inputs can regularize flow predictions with respect to exogenous factors, forecasting 15-min stage height with a summer MAE of 0.55 in. Daily max-min stage height ranges (I.e., daily max stage - min stage) are forecasted with summer MAE of 0.29 in. The produced model is capable of conditional forecasting to show when dam operations are complying with peak stage change recommendations. One experiment simulates a constant discharge input from the Yampa R. to evaluate the distribution of daily peak stage changes based only on summer load energy production. Notably, 37.4% of days in the simulation violate peak stage change recommendations compared to 67.6% of days in the observed hydrologic record when simulating constant Yampa R. discharge equivalent to the 2019 summer base flow (696.1 cfs).