Comparative assessment of VOS and model wind waves over the global oceans accounting for sampling effects

Validation of the global model long-term wind wave hindcasts against wave observations from Voluntary Observing Ships (VOS) are of a high demand, as VOS observations provide separate estimates of wind sea and swell for a long period of about several decades. However, VOS data suffer from spatial and temporal inhomogeneity of sampling and direct comparisons with model output are difficult as there is no way of quantifying differences associated with model setting and with sampling uncertainties. In this study, we perform a validation of the 40-yr long (1980-2019) wave model hindcast performed with WAVEWATCH-III spectral wave model forced with ERA5 reanalysis against provisionally corrected global VOS wind wave archive, accounting for sampling density in VOS data. For this purpose, model output was subsampled according to the VOS observations using the space-time collocation technique. This allows for estimating the total sampling uncertainty of significant wave heights (SWH), wind sea and swell. Further, this allows for comparison of model data with VOS observations which is not influenced by sampling biases.

Sampling uncertainties in SWH are largely driven by sampling uncertainties in the wind sea, being quite close to each other. Total sampling error variability in SWH and wind sea provide considerable reduction of uncertainties along the major ship routes where sampling errors drop by several times compared to the areas outside of the dense ship traffic. The smallest sampling uncertainties are identified in the North Atlantic subtropics where relatively weak short-term variability of wind waves is collocated with the moderately high sampling associated by dense ship traffic between Europe and the North America. Estimates of sampling errors are separately developed for sea and swell providing hints on the accuracy of wea and swell portioning in spectral wave model. We also estimated the impact of sampling onto extreme wind waves and this impact regionally is not necessarily correlated with the effect of sampling on means. Finally analysis is performed for patterns of interannual variability, including long-term trends.

This study is funded by RSF project # 23-47-00030.