Pattern recognition for imputation of missing radial surface current data - a Malta-Sicily Channel case study

Abstract

Surface currents can accurately be measured remotely using high-frequency radars, with the drawback that those measurements are susceptible to external interference resulting in frequent gaps in data. This paper compares the gap-filling accuracy of four pattern recognition machine learning methods - k-means clustering, self organising maps, growing neural gap and a Generative Adversarial Network. Several experiments are demonstrated using data from Malta-Sicily Channel, exploring the possibilities of applications of feature engineering to reduce the dimensionality of the problem. The methods are also compared to numerical models. Findings indicate how classical pattern recognition algorithms result in an average relative error of around 15%, while the Generative Adversarial Network halved that error.