Hydrodynamic model of the lower Columbia River, Oregon and Washington, 2017-2020

A three-dimensional hydrodynamic model of the lower Columbia River (LCR) constructed using the Delft3D Flexible Mesh (DFM) modeling suite (Deltares, 2020) was used to simulate water levels, flow, and seabed stresses between January 1, 2017 and April 20, 2020. The model application was adapted from van der Steeg (2016) and Wherry and others (2019). The model domain extends between the Bonneville Dam at its upstream limit to about 130 km offshore of the river mouth in the Pacific Ocean. Ten equally spaced vertical sigma layers were used to simulate 3D effects within the model domain. A high resolution digital elevation model constructed from recent topographic lidar and bathymetry data was used for the model bathymetry (http://www.estuarypartnership.org/lower-columbia-digital-terrain-model). Offshore bathymetry was derived from the General Bathymetric Chart of the Oceans (General Bathymetric Chart of the Oceans, 2018). Over 250 levees, stone jetties, and pile dikes were prescribed as sub-grid weirs with 0 transmission throughout the model domain. Roughness in the model was prescribed using a Chezy value of 65 for the ocean and lower estuary and a value of 70 farther upstream. Oceanic boundaries of the LCR model were forced using astronomic tidal constituents derived from the satellite-derived FES 2012 global tide model (Lyard and others, 2006). A vertical offset of 1.21 m (positive up) was applied at the oceanic boundary to account for the difference between local mean sea level and NAVD88 based on the tidal datum at nearby Toke Point, WA (NOAA station 9440910). Oceanic subtidal variations were imposed at the oceanic open boundary as a time-varying correction to the astronomic tides. The subtidal time-series was derived from observations of water levels at NOAA stations 9440910 (Toke Point, WA) and 9437540 (Garibaldi, OR). Water level time series from the two stations were low-pass filtered with a 66-hr cutoff to remove fluctuations at tidal frequencies. The low pass filtered values from both stations were highly correlated and an average for the two stations was applied to the oceanic model boundaries. Upstream, fluvial discharges from the main stem Columbia River, Cowlitz, Lewis, Willamette, and Sandy Rivers were prescribed based on observations made at various USGS gauge stations. The oceanic and fluvial boundaries were prescribed constant salinity values of 33 and 0 PSU, respectively. The effects of variations of temperature, wind, and waves on circulation and enhanced bed stresses were neglected in the present model application.

Simulated water levels and discharges were compared against time-series measurements of observed water levels from locations throughout the study area. All water level observations were converted to a common vertical datum (NAVD88) and water level and discharge data were interpolated in time from the native measurement interval (between 6 and 15 minutes depending on site) to the 10-minute output interval that was available for model simulations. Bulk error metrics describing the comparisons between model and observations over the entire simulation time frame (January 2017 to April 2020) were calculated following Jolliff and others (2009). Total root-mean-square error (RMSE) was less than 20 cm for water levels at all of the measurement sites. However, peak water levels were underestimated by as much as 0.75 m during high river discharge at Vancouver despite slightly overestimating total discharge at that location.

Harmonic tidal analysis on the simulated and observed water levels (Pawlowicz and others, 2002) over the analysis time frame was performed to validate model predictions of tidal propagation into the estuary and tidal river. The model accurately predicts the initial amplification of the M2 tidal constituent between the river mouth and Astoria tide gauge and subsequent decay farther upstream. The predicted phase of the M2 tidal constituent initially agrees well with observed values, but the simulated tidal wave is slower than observed upstream of the Wauna tide gauge. Further investigation is needed to determine the cause of the discrepancy and improve model predictions. Other major tidal constituents including K1, O1, and N2 were adequately characterized in the model simulation.