Missing oceanographic data resulting from an instrument that is damaged, malfunctioning, or otherwise non-operational can be an impediment to real-time applications of the data and also to statistical calculations pertaining to climatological studies. The problem of reconstructing lost information by correlating oceanographic data from multiple stations is explored through the use of artificial neural networks. A comprehensive simulation study was performed by developing six buoy networks (each containing several buoys) in locations with diverse geographical and wave properties: the northeastern part of the US, the Gulf of Mexico, and Prince William Sound (Alaska). The simulations demonstrate that the missing significant wave heights at the location of one buoy can, for the most part, be reliably reconstructed using artificial neural networks (ANNs) and data from other buoys. ANN model results in the northeastern part of the US tended to suffer from frequent under-prediction, while networks in the Gulf of Mexico and in Prince William Sound showed greater fidelity to measurements. The potential for redeploying some buoys to other locations and using the neural network model instead is examined. 2007 Elsevier Ltd. All rights reserved.
