EOF based predictive model using hybrid Wavelet and NARX network for extreme sea level prediction in the east coast of India

Abstract

A suitable prediction scheme of non-tidal sea level variations due to storm surges is a prerequisite for proper planning and mitigation during extreme events. The variation of sea level due to non-tidal forcing is complicated because of the time-variant and complex interaction of driving factors leading to mutual collinearity between the predictive variables. To address this issue, the present study endeavors to develop an empirical model for a sea level prediction scheme by adopting the Empirical Orthogonal Function (EOF) of the driving parameters and preparing an empirical model using a hybrid Continuous Wavelet Transform (CWT) and Nonlinear Autoregressive Networks with Exogenous inputs (NARX). The model shows very accurate predictions of the CWT coefficients (for instance, prediction during Tropical Cyclone Phailin shows an accuracy of R ≈ 0.99 between the target and the network predictions) and also depicts the relative contribution of different frequency components on sea level extremes. For this reason, the current scheme is applicable for the prediction of the future impact of high-frequency disturbances under the backdrop of varying low-period events like variability in seasonal sea level changes due to global warming and consequently aid in decision-making for designing future coastal structures.