PowerPoint Presentation - Dimension reduction in large data sets

ĄSEOF1 gives loadings of the linear function of 1008 variables that has maximum variance, subject to an extra constraint which ÔsimplifiesŐ loadings, including forcing some to zero.
ĄSEOF2 similarly, except that the linear function also has an orthogonality constraint relative to the first.

ĄMuch of the white space corresponds to exactly zero loadings.
ĄAdditional simplicity is Ôpaid forŐ by a loss of variance (17%, 13% of total compared to 23%, 13%)
ĄSEOF1 contrasts high and mid-latitudes in Atlantic/Europe sector (North Atlantic Oscillation – NAO)

ĄPacific is dominant in SEOF2