A methodology for data gap filling in wave records using Artificial Neural Networks. (May 2020)

• Record Type: Journal Article
• Authors: Vieira, Filipe; Cavalcante, Georgenes; Campos, Edmo; Taveira-Pinto, Francisco
• Is Part Of: Applied ocean research. Volume 98(2020)
• Publication Date: 2020-05
• Subjects: SWAN -- Model validation -- Measurement gap -- Artificial neural network -- Time series reconstruction; Ocean engineering -- Periodicals; 620.416205
• DOI: 10.1016/j.apor.2020.102109
• Languages: English
• ISSNs: 0141-1187

Abstract

Highlight: Alternative method for data gap filling in wave records using Artificial Neural Networks (ANN) Two-one week gaps in a long term ADCP deployment were filled using wave modelling and ANN Comparison between wave modelling results and ANN show similar performance ANN is simpler, faster and does not require detailed knowledge of numerical modelling Abstract: Wave measuring equipment are subject to malfunction that can be caused by various reasons such as maintenance, navigation accidents, errors in communications and biofouling of sensors. This can eventually result in loss of valuable data before the equipment is fixed. A continuous data set is often critical for modelling and/or analysis of wave conditions. Data gaps in wave records can be filled using numerical modelling. This study presents an alternative method for filling missing data based on publicly available wind and wave information and artificial neural networks (ANN). The ANN developed in this study uses offshore hindcast wave and wind information (significant wave height, peak wave period and direction, wind speed and direction) from a data set publicly available for any region of the globe. The results of the application of this method to two one-week gaps in wave measurements are compared against a spectral wave model developed and implemented to estimate wave conditions at the measurement location. The ANN results for the validation period show slightly better statistical performance when compared to the wave model results in terms of the correlation coefficient, root mean square error, bias and scatter index. ANN are viable alternative methods to wave modelling for filling gaps in data and have several advantages since specific knowledge of wave models is not required, input data to feed the neural network is available for any region of the globe and the processing time is highly reduced when compared to numerical modelling.