Daily sea level forecast at tide gauge Burgas, Bulgaria using artificial neural networks

The elucidation of peculiarities of the sea level fluctuations along the Black Sea coast has an important theoretical and applied significance with respect to the global and regional studies of the climate. The long and short-term forecasts of the water levels are an essential contribution to increasing the monitoring capacity of the natural disasters in order to reduce their impact on society and low-land coastal regions. This study deals with prediction assessments of the daily mean sea levels. The input data set used in the study, contains tide gauge data obtained at the town of Burgas, located at the western Bulgarian Black Sea coast for the period 1990–2003. The forecast is performed applying the conventionally used tide prediction model, multilayer Feed-Forward (FF), Cascade-Feed-Forward (CFF), Feed-Forward Time-Delay (FFTD), Radial Basis Function (RBF), Generalized Regression (GR) neural networks and Multiple Linear regression (MLR) methods. Several tests of different Artificial Neural Network (ANN) architectures and learning algorithms are carried out to assess their applicability as competitive methods to the harmonic analysis. The ANNs offer an effective approach to correlate the nonlinear relationship between an input and output of the sea levels by recognizing the historic patterns between them. The obtained results indicate that the artificial neural technique is suitable for short and long-term forecasts of the sea level parameters.