Hybrid data-driven long-term wave analysis in the southern Coral Sea, Australia

Abstract

This paper investigates the long-term statistical characteristics of waves in the southern Coral Sea, Australia. Measured and simulated datasets from three representative sites, including both nearshore and offshore locations, were utilised. The study focuses on wave conditions with highly frequent observations and the forecast of extreme events. To address data gaps, an innovative artificial neural network model is proposed to fill in missing data points. The uncertainty in the statistical results due to seasonal variability is also assessed. The analysed results indicate that the wave climate in the southern Coral Sea shows distinct seasonal differences, but stationary stochasticity can still be applied in data analysis. For the studied sites, the lognormal distribution is suitable for describing the frequently observed wave conditions, while a GEV-GP two-part distribution model may provide an improved and flexible fitting for extreme events, albeit with wide confidence intervals. In addition, under a given recurrence period, as the significant wave height H_s approaches the maxima, the variation in the mean zero up-crossing wave periods T_z tends to concentrate within a narrow range (indicating medium-distance swell dominance), while the variation in T_z spans a wider range (indicating the presence of either wind waves or swells) when H_s is smaller. These findings enhance our understanding of wave statistics and extreme events and provide valuable insights for the design of coastal and offshore structures, as well as other research related to wave climate.