[ Derived from parent entry - See data hierarchy tab ] Simulated 2D residual velocity fields in the inner German Bight were
subjected to Principal Component Analysis (PCA). Residual currents
were obtained from coastDat2 barotropic 2D simulations with the
hydrodynamic model TRIM-NP V2.1.22 in barotropic 2D mode on a Cartesian grid (1.6km spatial resolution)
stored on an hourly basis for the years 1948 - 2012
(doi:10.1594/WDCC/coastDat-2_TRIM-NP-2d) and later extended until
The present analysis refers to the period Jan 1958 - Aug 2015. The spatial
domain considered is the region to the east of 6 degrees east and to
the south of 55.6 degrees north. All grid nodes with a bathymetry of
less than 10m were excluded.
Residual velocities were calculated in two different ways:
1.) as 25h means,
2.) as monthly means.
Both types of residual current data are available from
The directory contains sub-directories for years and months. Daily
residual currents for the 13th of September 1974, for instance,
are stored in
while monthly mean residual currents for September 1974 are stored in:
All current fields provided were interpolated from the original
Cartesian model grid to a more convenient regular geographical grid
Mean residual currents are stored in:
This data set contains residual velocities both on original Cartesian
grid nodes and interpolated to the geographical grid. An example plot
For PCA, two residual velocity components from each of 12133 Cartesian
grid nodes were combined into one data vector (length 2x12133),
referring to 21061 daily or 692 monthly time levels. Results of two
independent PCAs for either daily or monthly mean fields are stored
Files contain three leading Principal Components (PCs) and
corresponding Emipirical Orthogonal Functions (EOFs). Again EOFs
were also interpolated to a regular geographical grid.
PC time series are also stored in plain ASCII format:
For monthly fields the number N of variables (N=2x12133) is much
larger than the number T of time levels (T=692). Therefore, to reduce
computational demands, the roles of time and space were formally
interchanged. Having conducted the PCA the EOFs were then transformed
back to the original spatial coordinates (cf. Section 12.2.6 in von
Storch and Zwiers (1999), Statistical Analysis in Climate Research,
Cambridge University Press).
A much larger number of time levels made even this approach
prohibitive for the full set of daily data. Therefore, PCAs were
performed for six sub-periods (1958-1965, 1966-1975, 1976-1985,
1986-1995, 1996-2005, 2006-2015(Aug)) independently. EOFs obtained from
these six sub-periods were then averaged to obtain EOFs representative
for the whole period. Corresponding PCs were calculated by
projecting daily fields onto these average EOFs.
IMPORTANT: In contrast with PCA of monthly data, the PCA of daily
data INVOLVES SOME APPROXIMATIONS!
EOFs on the original nodes were normalized to have unit lengths. The
show the first three EOFs obtained from daily data, assuming that
corresponding PCs have the value of one standard deviation. The
following two plots,
show the leading EOFs for monthly mean data. EOF3 is omitted as it
represents just a very small percentage of overall variance (1.7%).