A New Look at Stratospheric Sudden Warmings. Part II: Evaluation of Numerical
Model Simulations
ANDREW J. CHARLTON,*
,@@
LORENZO M. POLVANI,
⫹
JUDITH PERLWITZ,
#
FABRIZIO SASSI,
@
E
LISA MANZINI,
&
KIYOTAKA SHIBATA,** STEVEN PAWSON,
⫹⫹
J. ERIC NIELSEN,
⫹⫹
AND DAVID RIND
##
* Department of Applied Physics and Applied Mathematics, Columbia University, New York, New York
⫹
Department of Applied Physics and Applied Mathematics, and Department of Earth and Environmental Sciences,
Columbia University, New York, New York
#
Cooperative Institute for Research in Environmental Sciences, Climate Diagnostics Center, University of Colorado, and Physical
Sciences Division, NOAA/Earth System Research Laboratory, Boulder, Colorado
@
National Center for Atmospheric Research, Boulder, Colorado
&
Istituto Nazionale di Geofisica e Vulcanologia, Bologna, Italy
** Meteorological Research Institute, Tsukuba, Ibaraki, Japan
⫹⫹
Global Modeling and Assimilation Office, NASA GSFC, Greenbelt, Maryland
##
NASA Goddard Institute for Space Studies, New York, New York
(Manuscript received 13 October 2005, in final form 28 March 2006)
ABSTRACT
The simulation of major midwinter stratospheric sudden warmings (SSWs) in six stratosphere-resolving
general circulation models (GCMs) is examined. The GCMs are compared to a new climatology of SSWs,
based on the dynamical characteristics of the events. First, the number, type, and temporal distribution of
SSW events are evaluated. Most of the models show a lower frequency of SSW events than the climatology,
which has a mean frequency of 6.0 SSWs per decade. Statistical tests show that three of the six models
produce significantly fewer SSWs than the climatology, between 1.0 and 2.6 SSWs per decade. Second, four
process-based diagnostics are calculated for all of the SSW events in each model. It is found that SSWs in
the GCMs compare favorably with dynamical benchmarks for SSW established in the first part of the study.
These results indicate that GCMs are capable of quite accurately simulating the dynamics required to
produce SSWs, but with lower frequency than the climatology. Further dynamical diagnostics hint that, in
at least one case, this is due to a lack of meridional heat flux in the lower stratosphere. Even though the
SSWs simulated by most GCMs are dynamically realistic when compared to the NCEP–NCAR reanalysis,
the reasons for the relative paucity of SSWs in GCMs remains an important and open question.
1. Introduction
In Part I of this study (Charlton and Polvani 2006,
henceforth CP06), we constructed a new climatology of
major midwinter stratospheric sudden warmings
(SSWs) and proposed benchmarks for their simulation
in general circulation models (GCMs). In this study we
will analyze the simulation of SSWs by a series of
stratosphere-resolving GCMs. The GCMs will be
evaluated in two ways. First, the number, type, and
climatology of SSWs in the models will be compared to
the climatology established by CP06. Second, process-
based benchmarks of SSWs, introduced by CP06, will
be used to assess the performance of each GCM.
Previous studies have examined the simulation of
SSWs by individual stratosphere-resolving GCMs (e.g.,
Butchart et al. 2000; Manzini and Bengtsson 1996;
Erlebach et al. 1995), but as far as we are aware there
has been no comprehensive intercomparison of the per-
formance of a series of GCMs in this respect. Most of
the recent intercomparisons of stratosphere-resolving
GCMs (e.g., Austin et al. 2003; Shine et al. 2003) have
touched only briefly on the simulation of SSWs.
The occurrence of SSWs is crucial to the chemistry of
ozone, since the low temperatures that occur in undis-
turbed winters are an important prerequisite for deni-
@@
Current affiliation: Department of Meteorology, University
of Reading, Reading, United Kingdom.
Corresponding author address: Andrew J. Charlton, Depart-
ment of Meteorology, University of Reading, Reading, Berkshire,
RG6 6BB, United Kingdom.
E-mail: a.j.charlton@reading.ac.uk
470 JOURNAL OF CLIMATE VOLUME 20
© 2007 American Meteorological Society
JCLI3994
trification and subsequent catalytic ozone loss, if the
vortex remains intact into the spring. The importance
of warmings was recognized early in the GCM Reality
Intercomparison Project for SPARC (GRIPS) model
evaluation (Pawson et al. 2000), but the restricted
length of model simulations available until quite re-
cently has precluded detailed examination of the fre-
quency of occurrence of simulated SSWs. With longer
runs of coupled Chemistry–Climate Models (CCMs)
now possible, the CCM Validation (CCMVal) project
(Eyring et al. 2005) will examine SSWs in more detail.
This study, along with CP06, should complement and
inform CCMVal, both by assessing the performance of
some GCMs and by suggesting process-based dynami-
cal benchmarks to test other GCMs.
Part of the interest in validating stratosphere-
resolving GCMs is also due to the potential interactions
between greenhouse gas–induced climate changes,
stratospheric ozone depletion, and dynamical coupling
between the stratosphere and troposphere (Hartmann
et al. 2000). There is little consensus about future
changes to variability of the Arctic stratospheric polar
vortex (Rind et al. 1998; Schnadt and Dameris 2003).
We surmise that a necessary though not sufficient con-
dition for the suitability of GCMs to accurately simu-
late future stratospheric variability is that they produce
a credible simulation of the current SSW climatology.
Other factors, such as the simulation of future tropo-
spheric variability, should also be considered when de-
termining the suitability of a GCM for this task.
The paper is structured as follows. The GCMs to be
analyzed and the methods used are described in section
2. In section 3 we compare the stratospheric climatol-
ogy of the GCMs. In section 4 we examine the number,
type, and climatology of SSWs. In section 5 we compare
process-based benchmarks of SSWs between the
GCMs. In section 6 we provide further discussion and
comparison of the stratospheric dynamics of each
GCM. In section 7 we present conclusions.
2. Methodology and GCM runs
This section briefly describes the methodology used
to identify and classify SSWs and gives brief details of
the GCMs used in the study. Neither discussion is in-
tended to be exhaustive and readers should consult rel-
evant references for further details.
The methodology for identifying and classifying
SSWs is described in full by CP06. We confine our study
to SSWs that occur during the extended winter season,
November to March. First, SSWs are defined to occur
when the zonal mean zonal wind at 10 hPa and 60°N
becomes easterly, in line with the WMO definition. An
additional criterion, that the zonal mean zonal winds
return to westerlies for 10 or more consecutive days
following the SSW, is used to remove events that are
final warmings. Second, the algorithm classifies SSWs
into vortex splits (in which the stratospheric polar vor-
tex breaks into two comparably sized pieces) and vor-
tex displacements (in which the vortex remains largely
intact). The algorithm uses absolute vorticity to identify
the vortex edge and then compares the size and
strength of cyclonically rotating vortices in the flow to
determine if events are vortex splits or vortex displace-
ments.
In CP06, data from both the National Centers for
Environmental Prediction–National Center for Atmo-
spheric Research (NCEP–NCAR) reanalysis (Kistler et
al. 2001), and its Climate Data Assimilation System
(CDAS) extension, and the 40-yr European Centre for
Medium-Range Weather Forecasts (ECMWF) Re-
Analysis (ERA-40) dataset (Kallberg et al. 2004) were
used to establish a climatology of SSWs events between
the winter seasons of 1957/58 and 2001/02. The results
from the two reanalysis datasets were found to be very
similar, as should be expected given their largely com-
mon source of observations. Therefore in the present
study we use only the NCEP–NCAR data to evaluate
the GCMs. This also makes the construction of many of
the statistical tests much simpler.
The CP06 algorithm was used to test the simulation
of SSWs in a series of GCM simulations. We study
GCMs that are explicitly designed to resolve the strato-
sphere, which we call stratosphere resolving. We define
stratosphere-resolving GCMs as those with a model top
close to or above the stratopause (approximately 50 km
or 0.8 hPa) and with a meaningful number of model
levels (10 or more) in the stratosphere. One major con-
straint in choosing and obtaining GCM integrations in
order to examine the intra-annual variability of SSWs is
that daily or finer time resolution of diagnostic fields is
required. We found that the archiving of daily output is
by no means a standard practice among the modeling
centers and groups that run stratosphere-resolving
GCMs.
The GCMs used in this study are summarized in
Table 1, and the forcings used in each model are shown
in Table 2. In the following subsections we briefly dis-
cuss each GCM. We have attempted to restrict our at-
tention in this study to GCM runs that are forced by sea
surface temperatures (SSTs) from the same time period
as the NCEP–NCAR reanalysis. It was not possible to
obtain runs with observed SSTs for the Meteorological
Research Institute/Japan Meteorological Agency 1998
Model (MRIJMA; run with climatological SSTs), and
this may be a potential source of bias.
We have also attempted to examine the longest avail-
1FEBRUARY 2007 C HARLTON ET AL. 471
able runs of each GCM, to try to avoid spurious dis-
agreement between the GCMs and reanalysis resulting
from potential decadal variability of SSWs (Butchart et
al. 2000). Except for the Middle Atmosphere ECHAM
Model (MAECHAM; which is run for 29 full winter
seasons), all of the GCM runs used here are over com-
parable or longer time periods than the reanalysis data,
typically 50 yr.
a. NASA Goddard Space Flight Center,
finite-volume GCM (FVGCM)
The National Aeronautics and Space Administration
(NASA) Goddard Earth Observing System (GEOS-4)
GCM is a middle-atmosphere GCM based on the finite-
volume dynamical core of Lin (2004), with gravity wave
drag, cloud, and cumulus parameterizations originally
based on those in the Community Atmosphere Model
version 3 (CAM3). The model has 55 levels in the ver-
tical and the model top is at 0.01 hPa (approximately 80
km); the average vertical spacing of levels in the strato-
sphere is 1.2 km. The model has a flexible horizontal
resolution and is run in this case at 2° ⫻ 2.5°. The model
is forced with observed SSTs and sea ice between 1949
and 1997 using the Rayner et al. (2003) dataset. The
model runs are described in more detail in Stolarski et
al. (2005).
b. NASA Goddard Institute for Space Studies,
Global Climate/Middle Atmosphere Model, new
version (GISSL53)
The new NASA Goddard Institute for Space Studies,
Global Climate/Middle Atmosphere Model 3 is an up-
date from the previous version of the Global Climate–
Middle Atmosphere Model (GCMAM; Rind et al.
2002). The update includes new boundary layer and
turbulent schemes, convective and cloud cover param-
eterizations, and atmospheric radiation code. The
broad nature of the changes to these schemes is shared
with the new GISS model-E (Schmidt et al. 2006). The
gravity wave drag in this model utilizes the formula-
tions discussed in Rind et al. (1999, 1988) except that
much smaller values are used. A major difference with
the other models is that the nonorographic gravity wave
drag components are a function of resolved processes in
the troposphere. The model has four different vertical
and horizontal resolutions; the version used here is 4° ⫻
5°, with 53 layers in the vertical and model top at 0.002
hPa (approximately 85 km). The vertical spacing is 500
m in the middle to upper troposphere, 0.5 to 1 km in the
lower stratosphere, and 2 to 2.5 km in the upper strato-
sphere. The model is forced with observed SSTs and
sea ice between 1951 and 1997 using the Rayner et al.
(2003) dataset. A more complete description of all the
versions of the model is given in Rind et al. (2006,
manuscript submitted to J. Geophys. Res.).
c. NASA Goddard Institute for Space Studies,
Global Climate/Middle Atmosphere Model,
legacy version (GISSL23)
The NASA GISS Global Climate/Middle Atmo-
sphere Model is a middle-atmosphere GCM based on
the climate model of Hansen et al. (1983) and the
middle-atmosphere version outlined by Rind et al.
TABLE 2. Forcings used in each GCM run.
GCM Sea ice extent
CO
2
conc./ppmv O
3
climatology
Solar forcing
TOA/W m
⫺2
FVGCM Obs 1949–97 (Rayner et al. 2003) Fixed 355 Monthly variance (Langematz 2000) Fixed 1367
GISSL53 Climate 1975–84 (Rayner et al. 2003) Fixed 311 Monthly and yearly variance; multiple
sources (see text)
Fixed 1365.5
GISSL23 Climate 1975–84 (Rayner et al. 2003) Fixed 311 Monthly variance (London et al. 1976) Fixed 1367.6
WACCM Obs 1951–2000 (NCEP–NCAR, Reynolds) Fixed 355 Monthly variance (Liang et al. 1997) Fixed 1367
MAECHAM Obs 1970–98 (Rayner et al. 2003) Fixed 348 Monthly variance (Fortuin and Kelder 1998) Fixed 1365
MRIJMA Climate 1978–98 (Rayner et al. 2003) Fixed 348 Monthly variance (Liang et al. 1997)
(⬍0.4 hPa) CIRA (⬎0.4 hPa)
Fixed 1365
T
ABLE 1. GCM experiments used in the study.
GCM
Run
length/winters SST forcing
Horizontal
resolution
Vertical
levels Model top Reference
FVGCM 49 Obs 1949–97 2° ⫻ 2.5° 55 0.01 hPa Stolarski et al. (2005)
GISSL53 47 Obs 1951–97 4° ⫻ 5° 53 0.002 hPa Rind et al. (2002)
GISSL23 46 Obs 1951–96 8° ⫻ l0° 23 0.002 hPa Shindell et al. (1998)
WACCM 50 Obs 1951–2000 T63 66 150 km Sassi et al. (2004)
MAECHAM 29 Obs 1970–98 T42 39 0.01 hPa Manzini et al. (2006)
MRIJMA 60 Climate 60 years T42 45 0.01 hPa Shibata et al. (1999)
472 JOURNAL OF CLIMATE VOLUME 20
(1988). The model has 23 levels in the vertical and the
model top is at 0.002 hPa (approximately 85 km). The
vertical spacing of levels is 0.2 km near the surface, 3.8
km in the upper troposphere, and 5 to 5.8 km in the
stratosphere. The model has a horizontal resolution of
8° ⫻ 10°. The model is forced with observed SSTs and
sea ice between 1951 and 1997 using the Rayner et al.
(2003) dataset. GISSL23 has a much coarser horizontal
and vertical resolution than most of the other models in
the study. We include it because it has been used in a
number of high-profile studies that examined the re-
sponse of the stratosphere to changing greenhouse gas
concentrations and the impact of these changes on the
tropospheric flow (e.g., Shindell et al. 1999). Note that
this version of the model differs from that used in Rind
et al. (1988) and subsequent publications in that it has
greatly reduced orographic drag (Shindell et al. 1998).
d. NCAR Whole Atmosphere Community Climate
Model (WACCM)
The NCAR Whole Atmosphere Community Climate
Model version 1b is an extended version of the NCAR
Community Climate Model version 3 (CCM3; Kiehl et
al. 1998). The model has 66 levels in the vertical and the
model top is at 150 km (approximately 0.000002 hPa).
The average vertical spacing of levels in the strato-
sphere is 1.5 km. The model has a spectral formulation,
with resolution of T63 (approximately 1.875° ⫻ 1.875°).
The model is forced with observed SSTs from 1950 to
2000 using the NCEP Reynolds observed dataset
(http://podaac.jpl.nasa.gov/reynolds). The model runs
are described in more detail in Sassi et al. (2004).
e. Max Planck Institute for Meteorology
(MPI)/Middle Atmosphere ECHAM Model
(MAECHAM)
The MPI MAECHAM model is an extended version
of the MPI ECHAM5 model (Roeckner et al. 2003).
The model has 39 levels in the vertical and the model
top is at 0.01 hPa (approximately 80 km). The vertical
spacing of levels in the stratosphere varies from 1.5 to 3
km. The model has a spectral formulation, with resolu-
tion of T42 (approximately 2.8° ⫻ 2.8°). The model is
forced with observed SST and sea ice forcings, from the
Atmospheric Model Intercomparison Project II (AMIP
II; Gates et al. 1999). The model is described in more
detail by Manzini et al. (2006).
f. The Meteorological Research Institute/Japanese
Meteorological Agency 1998 Model (MRIJMA)
The MRIJMA 1998 Model is a hybrid version of the
Meteorological Research Institute model (Chiba et al.
1996) and the operational global model (GSM9603) of
the Japan Meteorological Agency (JMA 1997). The
model has 45 levels in the vertical and the model top is
at 0.01 hPa (approximately 80 km). The average verti-
cal spacing of levels in the stratosphere is 2 km. The
model has a spectral formulation, with resolution of
T42 (approximately 2.8° ⫻ 2.8°). The model is forced
with climatological SSTs and run for 60 yr. The model
setup is described in more detail in Shibata et al. (1999).
The climatological SSTs are 21-yr averages between
1978 and 1998, based on the Hadley Centre SST dataset
(HadSST).
3. Climatology of GCMs
In this section, the stratospheric climatology of the
GCMs is briefly examined. An indication of the
strength and size of the stratospheric polar vortex in
each GCM can be gained by examining the strato-
spheric climatology, with the caveat that it is often dif-
ficult to separate the time and zonal mean state of the
stratosphere and its time-varying component. We re-
strict our analysis to the zonal mean zonal wind at 10
hPa and the meridional heat flux at 100 hPa both for
sake of brevity and because of the limited amount of
data available to us.
a. Zonal mean zonal wind at 10 hPa
The zonal mean zonal wind on the 10-hPa pressure
surface as a function of latitude and time for each of the
GCMs and the NCEP–NCAR reanalysis is shown in
Fig. 1. The top-middle and top-right panels show line
plots of the winter mean zonal mean zonal wind as a
function of latitude for the various GCMs (colored
lines) and for the NCEP–NCAR reanalysis (black line).
There is large variability between the GCMs, both in
the seasonality of the zonal mean zonal wind and in its
maximum and winter mean values.
In terms of the winter mean zonal mean zonal wind,
three GCMs have a zonal jet either within or close to
one standard deviation from the NCEP–NCAR re-
analysis. Only GISSL23 and WACCM have zonal wind
speeds noticeably different from the reanalysis. Both
have very strong winter mean zonal mean wind maxi-
mum—GISSL23 has a maximum of 43.4 m s
⫺1
while
WACCM has a maximum of 44.8 m s
⫺1
compared to
the NCEP–NCAR reanalysis maximum of 21.9 m s
⫺1
.
The extremely strong jets in these GCMs are reminis-
cent of the “cold pole” problem prevalent in many
stratosphere resolving GCMs (Pawson et al. 2000). A
further curiosity is the easterly zonal mean zonal wind
values close to the pole in GISSL23. The extremely
strong jets in the GISSL23 model are a direct result of
the reduced orographic drag used by Shindell et al.
1FEBRUARY 2007 C HARLTON ET AL. 473
(1998) and are not a characteristic of the model as nor-
mally used (e.g., Rind et al. 1988, their Figs. 2 and 3).
Only GISSL53 has a weaker winter mean zonal mean
wind maximum (17.6 m s
⫺1
) than the reanalysis.
The seasonal cycle [Fig. 1 and further analysis (not
shown)] varies markedly between the different GCMs.
The reanalysis shows peak zonal mean zonal winds in
the extratropics between days 50 and 80 of the winter
season (late December to early February). Three of the
models (FVGCM, WACCM, and MRIJMA) simulate
this seasonality correctly, although the absolute values
of the zonal mean zonal wind in WACCM are on av-
erage 15 m s
⫺1
larger than the NCEP–NCAR reanaly-
sis. GISSL53 has a seasonality shifted toward early win-
ter, with peak zonal mean zonal winds between days 10
and 30 (November). MAECHAM has a seasonality
shifted toward late winter, with peak zonal mean zonal
winds between days 80 and 110 (late January to Feb-
FIG. 1. Zonal mean zonal wind climatology at 10 hPa for GCMs that resolve the stratosphere in this study. (top left) Climatology from
NCEP–NCAR reanalysis for the years 1958–2002. Contour interval is 5 m s
⫺1
. (top middle), (top right) Line plots show winter mean
for each GCM: NCEP–NCAR climatology in thick black line, FVGCM in red line, GISSL53 in green line, GISSL23 in blue line,
WACCM in magenta line, MAECHAM in the cyan line, and MRIJMA in yellow line. Gray shading shows ⫾one interannual standard
deviation from the mean.
474 JOURNAL OF CLIMATE VOLUME 20
Fig 1 live 4/C
