Data assimilation of sea surface temperature and salinity using basin-scale reconstruction from empirical orthogonal functions: a feasibility study in the northeastern Baltic Sea
<p>The tested data assimilation (DA) method based on EOF (Empirical Orthogonal Functions) reconstruction of observations decreased centred root-mean-square difference (RMSD) of surface temperature (SST) and salinity (SSS) in reference to observations in the NE Baltic Sea by 22 % and 34 %, resp...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Copernicus Publications
2021-01-01
|
| Series: | Ocean Science |
| Online Access: | https://os.copernicus.org/articles/17/91/2021/os-17-91-2021.pdf |
| Summary: | <p>The tested data assimilation (DA) method based on EOF
(Empirical Orthogonal Functions) reconstruction of observations decreased
centred root-mean-square difference (RMSD) of surface temperature (SST) and
salinity (SSS) in reference to observations in the NE Baltic Sea by 22 %
and 34 %, respectively, compared to the control run without DA. The method
is based on the covariance estimates from long-term model data. The
amplitudes of the pre-calculated dominating EOF modes are estimated from
point observations using least-squares optimization; the method builds the
variables on a regular grid. The study used a large number of in situ FerryBox
observations along four ship tracks from 1 May to 31 December 2015, and
observations from research vessels. Within DA, observations were
reconstructed as daily SST and SSS maps on the coarse grid with a resolution
of 5 <span class="inline-formula">×</span> 10 arcmin by N and E (ca. 5 nautical miles) and
subsequently were interpolated to the fine grid of the prognostic model with
a resolution of 0.5 <span class="inline-formula">×</span> 1 arcmin by N and E (ca. 0.5 nautical
miles). The fine-grid observational fields were used in the DA relaxation
scheme with daily interval. DA with EOF reconstruction technique was found
to be feasible for further implementation studies, since (1) the method that works
on the large-scale patterns (mesoscale features are neglected by taking only
the leading EOF modes) improves the high-resolution model performance by a
comparable or even better degree than in the other published studies, and (2) the
method is computationally effective.</p> |
|---|---|
| ISSN: | 1812-0784 1812-0792 |