Using Gridded Multi-mission Sea Surface Height Data to Estimate Tidal Heights at Columbia River Estuary

Abstract

Measured tidal height time-series are critical for establishing initial and boundary conditions for hydrodynamic models of estuaries. The inexistence of tidal stations in the developing world is more evident than in other parts of the world. The lack of tidal height time-series data oftentimes forces modelers to interpolate or extrapolate these critical data, introducing important uncertainty bounds in the output of hydrodynamic models of estuaries. This paper assesses the feasibility of using gridded multi-mission sea surface height data for estimating tidal heights for the Columbia River estuary located in Northwestern USA. Ocean surface anomaly and geostrophic velocities gridded data, along with historical measured tidal heights are used for training and calibration of the ANN model. A nonlinear autoregressive exogenous neural network is used to predict tidal heights at a 14-min time step. Statistical comparison between measured and predicted data showed that the quality of predicted values was good. Regression analysis for goodness of fit showed the applicability of the proposed method: R2 = 0.93, significance-F = 4.57 * 10−18, F = 370.80, P-value for the intercept = 0.13, standard error = 0.26 m. Overall, the neural network provides good estimations of tidal heights, considering that it uses coarse-spatial-resolution ocean surface and water velocity data with daily temporal resolution.