Long-term tidal forecasting and hindcasting using QuickTIDE tidal simulation package

Abstract

A tidal simulation package built on MATLAB neural network toolbox with user-friendly graphical user interface (GUI) is presented. The application, named QuickTIDE, is modeled using a back-propagation neural network (BPN) with single hidden layer and five input constituents. Twenty-one built-in network models trained on one-year hourly tidal observations are able to generate reliable tidal simulation from year 1994 to 2004 corresponding to their respective locations in Malaysia. Further tests on the historical data of station Boston Harbour from year 1930 to 2000 show that good forecast and hindcast can be produced up to 70 years if the number of neuron in the hidden layer and the number of input argument are properly selected. Using 30- day hourly observations, the application is capable of producing satisfactory simulation with average correlation coefficient R in the order of 0.96 from year 1994 to 2004 for station Port Klang. For forecasting and hindcasting from year 1930 to 2000 for station Boston Harbour, R-values average at 0.92 and 0.90 respectively, with better results in the month the training data originates.