Empirical Estimation of Nearshore Waves From a Global Deep-Water Wave Model

Global wind-wave models such as the National Oceanic and Atmospheric Administration WaveWatch 3 (NWW3) play an important role in monitoring the world's oceans. However, untransformed data at grid points in deep water provide a poor estimate of swell characteristics at nearshore locations, which are often of significant scientific, engineering, and public interest. Explicit wave modeling, such as the Simulating Waves Nearshore (SWAN), is one method for resolving the complex wave transformations affected by bathymetry, winds, and other local factors. However, obtaining accurate bathymetry and determining parameters for such models is often difficult. When target data is available (i.e., from in situ buoys or human observers, empirical alternatives such artificial neural networks (ANNs) and linear regression may be considered for inferring nearshore conditions from offshore model output. Using a sixfold cross-validation scheme, significant wave height
$H_s$
and period were estimated at one onshore and two nearshore locations. In estimating
$H_s$
at the shoreline, the validation performance of the best ANN was
$r = 0.91$
, as compared to those of linear regression (0.82), SWAN (0.78), and the NWW3
$H_s$
baseline (0.54).