A comparison of two operational wave assimilation methods

TL;DR

This paper compares two operational wave assimilation methods for the North Sea, analyzing their differences, impacts, and performance using real and synthetic data over a 39-day period.

Contribution

It provides a detailed comparison of the WAM/OIP and DASWAM wave assimilation schemes, highlighting their differing correction approaches and impacts on forecast accuracy.

Findings

WAM/OIP corrections are localized near observations.

DASWAM adjustments are more global in nature.

Wave data assimilation improves forecast accuracy up to 12 hours.

Abstract

A comparison is carried out between two operational wave forecasting/assimilation models for the North Sea, with the emphasis on the assimilation schemes. One model is the WAM model, in combination with an optimal interpolation method (OIP). The other model, DASWAM, consists of the third generation wave model PHIDIAS in combination with an approximate implementation of the adjoint method. In an experiment over the period February 19 - March 30, 1993, the models are driven by the same wind field (HIRLAM analysis winds), and the same observation data set is assimilated. This set consists of a) spectra from three pitch-and-roll buoys and b) Synthetic Aperture Radar (SAR) spectra from the ERS-1 satellite. Three analysis/forecast runs are performed: one without assimilation, one with assimilation of buoy measurements only, and one with all data assimilated. For validation, observations from four buoys, altimeter data from ERS-1 and Topex-Poseidon, and scatterometer data from ERS-1 are used. A detailed analysis of the "Wadden Storm" (February 20-22) shows the very different nature of the two assimilation schemes: the wave and wind field corrections of the WAM/OIP scheme are all in the vicinity of the observations, whereas the DASWAM adjustments are more of a global nature. The impact of some individual buoy and SAR observations is visualized. A comparison of the performance of the two schemes is somewhat obscured by the very different behaviour of the two first-guess runs. A statistical analysis over the whole 39-day period gives the following results. In a comparison with buoy observations it is shown that a positive impact of wave data assimilation remains until about 12 hours in forecast in