Wave Prediction and Data Supplementation with Artificial Neural Networks

Abstract

Successful development and management of coastal zone infrastructure is to a great extent based on proper estimates of current and future wave conditions. In this paper, a site-specific artificial neural network methodology is proposed to serve as a basic tool for both now and forecasting of wave parameters in variable coastal environments. This tool is an alternative to data and computational effort demanding deterministic wave models. The data used were collected during a field campaign near the Tasmanian coast. The campaign was undertaken to estimate sea states and temporal variability of wave conditions in the Tasmanian region. This study is considered to be a logical extension of that campaign. These data were used in a neural network study for the first time. The periods under consideration differ for different sites and extend from September 1985 to December 1993. The quality of the neural simulations was assessed vs. the independent measurements in terms of the bias, root mean square error, correlation coefficient, and scatter index. Neural networks of three different architectures perform in a similar way and produce reliable predictions 3 and 6 hours ahead. The accuracy drops for larger warning times. Meanwhile, the tracking and retrieval of wave parameters exhibit stable, high quality. Thus, the results show the feasibility of the methodology for wave forecasting and data supplementation in essentially different topographic conditions.