PCA¶

module load evaluation_system

analyze --tool PCA --help

PCA (v3.1.0): Principal Component Analysis
Options:
session      (default: 1)
             This defines two possible sessions:
                     1 means the program performs a PCA computing a maximum
             number of EOFs given by EOFS.
                     2 means a PCA has been already performed and a back-
             transformation is computed retaining a certain number of EOFs given
             by EOFS.
eofs         (default: 10)
             Number of EOFs that are computed by the PCA (SESSION=1) or that are
             included in the back-transformation (SESSION=2). If<=0 the largest
             possible value will be used.
input        (default: <undefined>) [mandatory]
             An arbitrary NetCDF file. There are only two restrictions to your
             NetCDF file: a) Time has to be the very first dimension in the
             variable you like to analyze. b) All dimensions in your variable
             need to be defined as variables themselves with equal names. Both,
             a) and b), are usually true.
variable     (default: <undefined>) [mandatory]
             The name of the variable in the NetCDF INPUT file you want to
             analyze.
outputdir    (default: $USER_OUTPUT_DIR)
             The output directory.
outputplots  (default: $USER_PLOTS_DIR)
             Output directory of produced plots
outputtype   (default: eps) [mandatory]
             Output filetype, i.e. eps, pdf, png
pcafile      (default: $input.pca.$variable.nc) [mandatory]
             Filename of the PCA output. If SESSION=1 this is output, if
             SESSION=2 this is input.
projection   (default: $input.pro.$variable.nc)
             Filename of the projection (back-transformation). If SESSION=1 this
             is not applicable, if SESSION=2 this is output.
missingvalue (default: 1e+38)
             The missing value (fill value) of VARIABLE.
principals   (default: True)
             Whether or not you want to have principal components computed.
eigvalscale  (default: False)
             Whether or not you want to have eigenvectors and principal
             components scaled with the square root of the corresponding
             eigenvalues. If true, this gives a physical amplitude to the
             eigenvectors and normalizes the principal components.
normalize    (default: False)
             Whether or not you want to have your data normalized (divided by
             the standard deviation per grid point).
areaweight   (default: False)
             Whether or not you want to have your data area weighted. This is
             done per latitude with sqrt(cos(latitude)).
latname      (default: lat)
             In case you want to use AREAWEIGHT you need to specify the name of
             the dimension representing the latitude.
shiftlats    (default: False)
             Whether or not you want to have the latitudes shifted before
             performing the area weighting. This shift by half of the difference
             between two latitudes avoids the problem that the cosine of
             +/-90deg is zero. This is recommended if AREAWEIGHT is true and if
             there exists a gridpoint with lat=+/-90deg.
testorthog   (default: False)
             Whether or not you want to have the eigenvectors and principal
             components tested for orthogonality. A value close to one indicates
             an orthogonal basis.
bootstrap    (default: False)
             Whether or not you want uncertainties of the eigenvalues computed
             using bootstrap based on case resampling.
boots        (default: 100)
             Number of bootstraps.
threads      (default: 8)
             This lets you control the number of threads you want to use. If you
             want to make use of the full computing power then n is a reasonable
             value with n being the number of cores on your machine.