Executive Summary
The Climate Prediction and its Application to Society (CliPAS) team is an
international research project of the Asia-Pacific Economic Cooperation (APEC)
Climate Center (APCC). Its goals is to provide APCC with frontier research in
climate predictability and prediction and to facilitate APCC’s effort in
developing first-rate model tools and technologies and continuously
improving APCC operational forecast system. The strategy of APCC/CliPAS is
to coordinate leading climate scientists in 12 institutions through well designed
research projects and to share their expertise in climate prediction and its
application.
The project in 2007 has been devoted to improving APCC operational multimodel ensemble (MME) seasonal prediction system through (1) providing one-tier
predictions using coupled models developed by three non-operational institutions in
APCC/CliPAS team and (2) implementing new MME scheme (MME-SPPM) to
APCC. Three coupled model predictions from FRCGC, SNU, and UH was delivered
to APCC, targeting 6-month lead coupled predictions, initiated from Nov 1, 2007.
The MME-SPPM was implemented to APCC. In addition, the APCC/CliPAS team
completed the coordinated multi-institutional retrospective forecast experiments for
advancing our understanding of climate predictability and determining the capability
and limitations of the MME prediction.
1
1. The Summary of 2007 Achievements
Plan
Progress
Achievement
Improvement of MME Test and Evaluation of the MME Kug, Lee, Kang, Wang,
method
3.1 (the MME method based on and
SPPM
v2)
using
Park
(2007,
APCC submitted to GRL)
operational predictions.
Draft
manuscript
APCC
for
operational
prediction
Case Study of the Case study was done for DJF Draft
causes
of
manuscript
the 1989/90, MAM 1994, SON 2003, APCC
seasonal forecast for DJF
2003/04
using
for
operational
APCC prediction
which most models operational prediction
failed
Evaluating
operational
with
a
APCC A
hierarchy
of
models designed
metrics
to
was Report on a hierarcy of
evaluate metrics,
newly intraseasonal-to-seasonal
Technical
designed hierarchy of prediction.
metrics
report
evaluating
on
APCC
APCC operational prediction was operational prediction
evaluated using metrics on mean
states (annual mean and cycle)
and
interannual
precipitation.
modes
of
variability
of
The
dominant
A-AM
monsoon
variability in APCC MME and
individual
models
were
also
investigated.
One-Tier predictions The
CliPAS
multi-institutional Three
coupled
with coupled models hindcast experiments consist of 7 predictions
model
from
from non-operational two-tier and 6 one-tier predictions FRCGC, SNU, and UH
center
for 1981-2004. Among the one- was delivered to APCC,
tier prediction models, two are targeting 6-month lead
operational
(NCEP
2
CFS,
and coupled
predictions,
BMRC). Among 5 non-operational initiated
coupled
models,
3
from
Nov
1,
models 2007.
implemented to APCC.
Multi-institutional
The
multi-institutional Wang,
retrospective forecast retrospective
experiments
forecast Shukla,
Lee,
Park
Kang,
and
co
experiments were updated and authors (2007, will be
completed
for
four
seasons. submitted to J. Climate)
APCC CliPAS team collected 7
one-tier and 7 two-tier predictions
APCC
operational The APCC operational two-tier Draft
two-tier
MME
manuscript
and MME prediction was compared comparison
APCC/CliPAS one-tier with APCC/CliPAS one-tier MME one-tier
MME prediction
and
for
between
two-tier
prediction for the Boreal summer MME prediction
precipitation
and
atmospheric
circulation
Predictability
of Predictability of ENSO and global Jin, Kinter, Wang, and co
ENSO
and precipitation in coupled models authors (2007, accepted
predictability
of were investigated.
to Clim. Dyn.)
global precipitation in
Jin and Kinter (2007,
coupled models
submitted to Clim. Dyn.)
Wang,
Lee,
Kang,
Shukla, Hameed, Park
(2007,
CLIVAR
Exchanges 12, 4, 17-18)
Experimental
Experimental hindcasts of MJO Fu, Wang, Bao, Liu, and
prediction of ISO with and Boreal summer ISO have Yang (2007, submitted to
a
hybrid
model
coupled been produced using UH hybrid GRL)
coupled GCM
3
4
2. Highlighting Achievements
(1) Improvement of MME method
The MME-SPPM, which developed in 2006 project, was tested on prediction
of 850 hPa temperature and precipitation using APCC hindcast data for the period
1983-2003 and operational forecast data for 2006 and 2007. The MME-SPPM code
has been transferred to APCC for operational use. The major advantage of the new
MME system is to improve the skill of seasonal precipitation over extratropical
continental regions. Compared to other MME predictions, MME-SPPM offers better
skill over the regions in which the average of individual model skill is poor.
(2) Case Study of the causes of the seasonal forecast for which
most models failed
Anomaly correlation skill for seasonal precipitation using APCC operational
prediction system reveals several interesting points. First, although the MME-SPPM
prediction has better skill in seasonal precipitation over the globe (0-360E, 50S50N) than the equal weighting MME prediction (MME 1), its interannual skill varies
in a very similar way as that of MME 1. Second, the interannual skill variation is
season-dependent. Third, the year-to-year variation in overall skill depends on
ENSO variability. The worst years tend to occur during transition or normal ENSO
phase. The major failures in prediction of DJF precipitation tended to occur one
year before (after) the mature phases of El Nino (La Nina). Fourth, it is noted that in
some seasons, majority of models failed prediction (correlation skill of MME 1
prediction below 0.1. For these seasons, it is likely that the MME-SPPM is not able
to improve the forecast skill regardless of application of statistical correlation
5
methods. Finally, the prediction skill in winter season is strongly related to that in
previous fall season during recent decade.
The case studies for those seasons reveal that most of coupled models failed
to predict SST anomalies over Tropical Oceans as well as extratropical Oceans,
resulting in the failure in predicting atmospheric circulation and precipitation. For all
cases, large temperature anomalies were found over the subtropical and
extratropical region rather that tropical Eastern Pacific in observation, while models
cannot capture the observed anomalies. For all cases, the spatial pattern of
predicted anomalies was significantly different among models, resulting in very
weak anomalies of MME prediction all over the globe.
(3) Evaluating APCC operational models with a hierarchy of
metrics
A hierarchy of metrics have been designed (Appendix I) for gauging
climate models performance on intraseasonal-to-seasonal prediction and a
selected set of metrics has been applied to evaluate APCC climate models’
performance (Appendix II), which can offer feedbacks to the participating
institutions to facilitate further improvement of the participating models and the
APCC operational prediction system. The metrics have been used for assessing
the annual modes of precipitation include long-term annual mean, the two
leading modes of annual variations, and monsoon domain. Temporal correlation
skill was calculated to verify model’s general performance on interannual
variability of precipitation for four seasons. In addition, Season-reliant EOF
6
(SEOF) analysis is used as the metrics for gauging model’s performance on
interannual variability of Asian-Australian monsoon (A-AM) precipitation.
(4) One-tier predictions from non-operational centers
The CliPAS multi-institutional hindcast experiments consist of 7 two-tier and 6
one-tier predictions for 1981-2004. Among the one-tier prediction models, two are
operational (NCEP CFS, and BMRC). Among 5 non-operational coupled models, 3
models implemented to APCC for real-time 6-month lead prediction initiated from
Nov 1, 2007. These three models are from FRCGC in Japan, UH in USA, and SNU
in Korea. It is expected that the APCC operational MME prediction, which is
currently mainly consisting of two-tier predictions can be improved through including
more one-tier predictions.
(5) Multi-institutional retrospective forecast experiments
The APCC/CliPAS team has completed the four-season multi-institutional
retrospective forecast experiments for the period 1979-2004. We collected 7 twotier and 7 one-tier predictions from 12 institutions in Korea, USA, Japan, China, and
Australia. This is an extremely valuable dataset for advancing our understanding of
climate predictability and determining the capability and limitations of the MME
forecast.
The APCC/CliPAS team has made comprehensive analyses of 21 state-ofthe art climate models’ two-decade long hindcast datasets. These 21 models
7
include 14 from APCC/CliPAS and 7 from DEMETER. The analysis results have
significantly advanced our knowledge on what can be expected to be predicted and
what the current level of forecast skills is. The analysis also revealed some forefront
climate prediction issues to be addressed by the climate prediction community.
(6) APCC operational two-tier MME vs APCC/CliPAS one-tier
MME prediction
The APCC operational two-tier MME prediction was compared with the
APCC/CliPAS one-tier MME prediction for JJA precipitation. The APCC two-tier
MME consists of CWB, GCPS/SNU, GDAPS/KMA, METRI/KMA, JMA, and MGO
and the CliPAS one-tier MME consists of NASA, CFS/NCEP, SINTEX-F, SNU, UH,
GFDL, POAMA/BMRC. In JJA, the CliPAS one-tier MME system has better skill
than two-tier MME for seasonal climate prediction as well as simulation of mean
and annual cycle. On the contrary, the skill difference between two MME systems is
very small in DJF.
(7)
Predictability of
ENSO
and
predictability
of
global
precipitation in coupled models
Predictability of ENSO-monsoon relationship has been studied using the
simplified numerical experiments in order to understand the role of possible sources
to degrade the predictability of ENSO-monsoon relationship in CGCM forecast. In
CGCM retrospective forecasts, the predictability of lead-lag ENSO-monsoon
relationship drops with respect to lead month. Systematic errors of couple models is
major factor of limiting predictability (mean error, phase error, amplitude error,
8
seasonal cycle). In this study, we investigate the model capability in long simulation
is the one key to understand the behavior of forecast error.
The relative role of systematic errors of remote forcing and other sources
including air-sea coupling, ocean dynamics, and SST outside the tropical eastern
Pacific on ENSO-monsoon relationship are investigated by performing idealized
experiments. The pacemaker, control, coupled, and AMIP run are performed and
compared. Pacemaker run shows the higher predictability comparing to the control
and AMIP run suggesting the air-sea coupling can improve to reproduce the
atmospheric bridge coming from the remote forcing. However, the western North
Pacific summer monsoon index associated with low-level anticyclone does not
change much regardless of existence of air-sea coupling.
How to measure the predictability in coupled climate system, where no
atmospheric lower boundary forcing given, is an open issue. We have shown that
the prediction skill of the coupled model MME basically comes from the skill in
prediction of the first three major modes of interannual variations in the global
tropical precipitation. The three modes together account for about 54% of the total
interannual variance averaged over the tropics in observation. This portion of the
variation may be considered as practically predictable part of the precipitation
variability, because the MME can capture these four major modes reasonably well
but cannot capture the rest higher modes. This result leads to a new approach to
estimate the practical predictability of the tropical seasonal precipitation in the
coupled climate models; i.e., we can quantify the “predictability” by the fractional
variance that is accounted for by the “predictable” leading modes in observation.
9
Such “predictable” modes can be determined by examining models’ hindcast results
such as the performance.
(8) Experimental prediction of ISO with a hybrid coupled model
Experimental hindcasts of MJO and ISO have been produced using UH
hybrid coupled GCM. For MJO experiments, the model was initialized with TOGACOARE observations from January 1, 1993, and allowed to run freely for 2 months.
A comparison of daily rainfall from the observation and from a 100-ensemble-mean
model output reveals that the model was able to “forecast” the eastward movement
and associated rainfall of the MJO beyond one month fairly accurately. For boreal
summer monsoon ISO experiments, the model was initialized with NCEP reanalysis
data on Jun 11, 2006. The model also was able to predict the northward movement
and associated rainfall of the ISO quite realistically.
3.
Achievement
from
APCC/CliPAS
Multi-Institutional
Retrospective Forecast Experiment
(1) Prediction of Equatorial SST
The equatorial sea surface temperature (SST) anomalies are the primary
sources of climate predictability worldwide. The MME one-month lead hindcast can
predict, with high fidelity, the spatial-temporal structures of the first two leading EOF
modes for both JJA and DJF seasons, which accounts for about 80-90% of the total
variance. The major bias is a westward shift of the SSTA between the dateline and
120E, which results in a significant error in the western Pacific SST and potentially
10
degrades the teleconnection associated with the western Pacific SSTA. The
temporal correlation coefficient (TCC) skill of the MME forecast of Nino 3.4 index at
a 6-month lead reaches 0.81 and 0.85 for boreal summer and winter seasons,
respectively. The TCC for SST predictions over the equatorial eastern Indian Ocean
(EIO) reaches about 0.68 at a 6-month lead forecast. However, the TCC skill for
Indian Ocean Dipole (IOD) index drops below 0.4 at the 3-month lead forecast for
both the May and November initiations. There exist IOD prediction barriers across
January and July.
(2) Prediction of Atmospheric Circulation, Precipitation, and
Temperature
Prediction of 850 hPa streamfunction field shows high skills over Western
Pacific and Asian continents in JJA, and in the eastern Pacific (east of 180E) and
North America, as well as maritime continent in DJF. The 200 hPa streamfunction
shows good or very good TCC skill almost everywhere between 40S and 60N
except in the equatorial region. At 500 hPa, the geopotential height shows a high
prediction skill confining to the global tropics with a north-south seasonal migration.
The DJF skills are considerably higher than JJA for all three levels.
The prediction of atmospheric circulation shows higher temporal correlation
skill than precipitation and may provide more reliable large scale signals for
downscaling to regional scale precipitation. The season-dependence and spatial
patterns of the circulation forecast skills can be well explained in terms of the ENSO
impacts. The variations in the spatial patterns and the seasonality of the correlation
skills strongly suggest that ENSO variability is the primarily source of the global
11
seasonal prediction skill.
Prediction of air temperature is considerably superior to the persistence skill
in the warm pool oceans, but not over the continental areas. The precipitation
prediction in Asian-Pacific monsoon region has moderate skill in cold seasons but
little skill over the continental summer monsoon regions. The seasonal march of the
thermal equator seems to add predictability to local summer hemisphere and
change of the westerly jet location can, to some extents, provide prediction skill by
influencing Rossby wave activity.
(3) Comparative assessment of the one-tier and two-tier MME
prediction
One-tier and two-tier MME predictions have been compared using 7 one-tier
and 7 two-tier predictions in APCC/CliPAS project. In JJA, the one-tier MME system
has better skill than two-tier MME for seasonal climate prediction as well as
simulation of mean and annual cycle. On the contrary, the skill difference between
two MME systems is very small in DJF. NCEP two-tier prediction was forced by
predicted SST using NCEP one-tier system. The comparison of NCEP one-tier and
two-tier prediction supports the necessity to use one-tier system for predicting
summer rainfall. NCEP one-tier prediction shows increased feedback from local
SST to some extent, although it bears similar systematic error as two-tier, especially
over East China Sea and Western North Pacific. NCEP one-tier prediction shows
improved ENSO-monsoon teleconnection over Indian Ocean, while it remains
unrealistic in JJA precipitation prediction over the Western North Pacific in the
following SON and DJF.
Two-tier MME shows distinctive difference from one-tier prediction during El
12
Nino onset and decaying summers. Precipitation error is large over South Asia in
one-tier prediction during El Nino onset summers. Two-tier MME has large errors
over the same regions during El Nino decaying summers.
(4)
Impacts of the systematic errors on ENSO-monsoon
relationship
Previous studies have shown that improvements in a coupled model’s mean
climatology generally lead to a more realistic simulation of ENSO-monsoon
teleconnection. Using 13 coupled retrospective forecasts, it is shown that the
systematic bias of SST mean state degrades the skill in predicting SST anomaly
and the ENSO-monsoon relationship. The errors in El Nino amplitude, phase, and
the location of the maximum variability in the coupled models are associated with
the mean state errors such as colder equatorial Pacific SST and stronger easterly
wind over western equatorial Pacific. The breaking relationship between ENSO and
Indian monsoon is evident in observation, whilst the MME produces robust negative
relationship, which is mainly due to the SST anomaly bias. The anomalous
precipitation and circulation are predicted better in the ENSO decaying JJA than
ENSO developing JJA.
13