Geoscience Reference
In-Depth Information
Figure 9.6 (a) Pattern
correlation coefficient
between the simulated and
observed precipitation
anomalies for each model
and each season over the
monsoon-ENSO region
(30
8
S-30
8
N, 60
8
E-90
8
W).
(b) Root-mean-square (rms)
of the simulated precipitation
anomalies over the
monsoon-ENSO region,
normalized by the observed
rms. The vertical line in the
bar indicates the range of
the correlation and rms
values of individual runs.
(Adapted from Kang
et al.
2001b
)
(a)
DJF96/97
JJA97
DJF97/98
JJA98
1.0
0.8
0.6
0.4
0.2
0.0
COLA
DNM GEOS GFDL
IAP
IITM
MRI
NCAR NCEP
SNU SUNY Comp
-0.2
(b)
DJF96/97
JJA97
DJF97/98
JJA98
2.0
1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0.0
COLA
DNM GEOS GFDL
IAP
IITM
MRI
NCAR NCEP
SNU
SUNY
Comp
when the El Ni˜o signal is at a peak, suggesting that all models are responsive to
the warm phase of the El Ni˜o. If the eastern portion of the domain (east of the
dateline) is excluded in the pattern correlation calculation, P
cor
reduces drama-
tically (not shown), indicating that most of the good correlation is contributed by
the rainfall directly responding to the El Ni
˜
o SST over the central and eastern
Pacific. The ensemble mean P
cor
for each model tends to be higher than for most
individual ensemble members. Similarly, P
cor
for the model mean (columns to
the extreme right) are generally among the top tier of the better performing
models. The picture is quite similar for R
rms
(Figure
9.6b
). Here, about half of the
models have R
rms
greater than 1.0, and about half less than 1.0. The rms ratios for
the individual members (whose range is indicated by the vertical line) tend to be
larger than the model mean. This is because the model mean tends to smooth
out the spatial features. The all-model R
rms
is less than 1.0, suggesting that
the models collectively underestimate the observed variability of the rainfall
Search WWH ::
Custom Search