330 Martin Fischer
u
00 O
ai
'tl
-1
u
00 O
(1)
'tl
-1
c..i
~ O
'tl
-1
$
"
83 84 85 86 87 88 89
. a: control experiment
..
..
.
.
,
83 84 85 86 87 88 89
b: temperature experiment
83 84 85 86 87 88 89
c: sea level experiment
Fig. 16.14 Observed and predicted SST anomalies averaged over the NINO-3 region for the
period 1982 to 1989. Each individual forecast was carried out over one year. From Fischer et
al. 1997.
tionally subsurface temperature measurements were assimilated. The skill is
clearly superior and does only slightly degrade with increasing lead time in the January ensemble.
A point to mention is the local correlation minimum at a lead time of 2 to 3
months in both ensembles of the FDA experiment. Directly after the start of the
forecasts, the correlation skill drops even quicker than persistence, and recovers
after it reached a minimum after 2 months. This behavior was not found in the
other two experiments. The reason is the coupling shock that was already mentioned in section 16.3.1. Since the components ofthe forecast system are initialized
separate1y the ocean and atmosphere are not ba1anced relative to each other. They
u
00 O
ai
'tl
-1
u
00 O
(1)
'tl
-1
c..i
~ O
'tl
-1
$
"
83 84 85 86 87 88 89
. a: control experiment
..
..
.
.
,
83 84 85 86 87 88 89
b: temperature experiment
83 84 85 86 87 88 89
c: sea level experiment
Fig. 16.14 Observed and predicted SST anomalies averaged over the NINO-3 region for the
period 1982 to 1989. Each individual forecast was carried out over one year. From Fischer et
al. 1997.
tionally subsurface temperature measurements were assimilated. The skill is
clearly superior and does only slightly degrade with increasing lead time in the January ensemble.
A point to mention is the local correlation minimum at a lead time of 2 to 3
months in both ensembles of the FDA experiment. Directly after the start of the
forecasts, the correlation skill drops even quicker than persistence, and recovers
after it reached a minimum after 2 months. This behavior was not found in the
other two experiments. The reason is the coupling shock that was already mentioned in section 16.3.1. Since the components ofthe forecast system are initialized
separate1y the ocean and atmosphere are not ba1anced relative to each other. They