have larger growth rate and more coherent spatial ENSO-like
structure between events.
oBred vectors of CZ model have the largest growth rate in
season => “spring barrier”.
in data assimilation:
oThe ENSO forecast errors can be reduced as much as 30% when bred vectors are removed
from initial errors.
oWithin data assimilation cycles, this reduction of errors would accumulate to an even a larger
in ensemble forecasts:
oThe ensemble forecasts with a pair of
“positive/negative” bred vectors
improve the skill significantly.