Filling gaps in wave records with artificial neural networks

Abstract

This contribution presents a neural data interpolation methodology, which was implemented to restore missing wave measurements. The methodology is based on the ability of artificial neural networks to find and reproduce non-linear dependencies within complex geophysical systems. The data were obtained from a field campaign during July 1985- ecember 1993 near Tasmania. Wave observations from a "Waverider" buoy were broadcasted as a high frequency radio signal via a quarter-wave antenna to a "Diwar" receiver. These measurements were used to train and to validate the neural nets employed. To restore missing data over time periods from 12 to 36 hours, five feed-forward, three-layered, artificial neural networks of a similar structure were implemented. The artificial neural networks' performance was estimated in terms of the bias, root mean square error, correlation coefficient, and scatter index. The methodology demonstrated reliable results with a fairly good overall agreement between the restored wave records and actual measurements.