Ocean hydrodynamics model with tidal forcing derived using an Artificial Neural Network

Abstract

The prescription of open boundary conditions is one of the greatest challenges in ocean and coastal hydrodynamic modeling. It requires extensive information on current velocity, tidal elevation, temperature and salinity at specific points in time and space. Once boundary conditions are known and correctly prescribed, one can solve a set of governing equations to simulate the dynamics of the state variables in a domain of interest. In this work, we compare the performance of a hydrodynamic model using two different sets of tidal boundary conditions derived from harmonic analyses in one case and an artificial neural network (ANN) in another. The hydrodynamic model used in this study is TMSI's TMH (Tropical Marine Hydrodynamics). The study domain includes the entire Singapore Strait and has six open boundaries. Since no consistent current velocity measurements are available at the moment, only tidal elevations are prescribed at the boundaries. In one approach, tidal boundary conditions are predicted by applying harmonic analysis on a set of known harmonic constituents. These harmonic constituents are extracted from JTS, ATT tide tables, or TotalTide. This technique requires information from only tidal stations nearest the open boundaries. In the ANN method, historical data from 25 tidal stations in the domain are used to derive the required information at the model boundary points. The tide elevation, H(t), at time t at a particular location (x, y) is expressed in terms of nearby TotalTide stations as follows: H <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i</sub> (t)=f(x <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i</sub> ,y <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i</sub> , H <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> (t),H <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> (t-1),H <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> (t),H <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> (t-1), ... The ANN is applied to compute the parameters in the above relationship. The ANN is first trained, and its performance is judged by the goodness-of-fit measures resulting from the training and the cross validation datasets. Upon achieving satisfactory performance, the trained ANN is then used to generate tide elevation data at specific boundary grid points. TMH is then run for the two cases of boundary conditions, and the results are compared between each others, and both against measurements. The results from ANN methods are capable to improve accuracy of tidal hydrodynamic predictions.