| description abstract | Although mean or integral properties of wave spectra are typically used to evaluate numerical wave model performance, one must look into the spectral details to identify sources of model deficiencies. This creates a significant problem, as basin-scale wave models can generate millions of independent spectral values. To facilitate selection of a wave modeling technology for producing a multidecade Pacific hindcast, a new approach was developed to reduce the spectral content contained in detailed wave hindcasts to a convenient set of performance indicators. The method employs efficient image processing tools to extract windsea and swell wave components from monthly series of nondirectional and directional wave spectra. Using buoy observations as ground truth, both temporal correlation (TC) and quantile?quantile (QQ) statistical analyses are used to quantify hindcast skill in reproducing measured wave component height, period, and direction attributes. An integrated performance analysis synthesizes the TC and QQ results into a robust assessment of prediction skill and yields distinctive diagnostics on model inputs and source term behavior. The method is applied to a set of Pacific basin hindcasts computed using the WAM, WAVEWATCH III, and WAVAD numerical wave models. The results provide a unique assessment of model performance and have guided the selection of WAVEWATCH III for use in Pacific hindcast production runs for the U.S. Army Corps of Engineers Wave Information Studies Program. | |
