Content uploaded by Rodrigo J. Bombardi
Author content
All content in this area was uploaded by Rodrigo J. Bombardi on Oct 14, 2014
Content may be subject to copyright.
Content uploaded by Leila M. V. Carvalho
Author content
All content in this area was uploaded by Leila M. V. Carvalho
Content may be subject to copyright.
IPCC global coupled model simulations of the South America
monsoon system
Rodrigo J. Bombardi Æ Leila M. V. Carvalho
Received: 16 May 2008 / Accepted: 24 October 2008 / Published online: 15 November 2008
Ó Springer-Verlag 2008
Abstract This study examines the variability of the South
America monsoon system (SAMS) over tropical South
America (SA). The onset, end, and total rainfall during the
summer monsoon are investigated using precipitation
pentad estimates from the global precipitation climatology
project (GPCP) 1979–2006. Likewise, the variability of
SAMS characteristics is examined in ten Intergovernmen-
tal Panel on Climate Change (IPCC) global coupled
climate models in the twentieth century (1981–2000) and
in a future scenario of global change (A1B) (2081–2100). It
is shown that most IPCC models misrepresent the inter-
tropical convergence zone and therefore do not capture the
actual annual cycle of precipitation over the Amazon and
northwest SA. Most models can correctly represent the
spatiotemporal variability of the annual cycle of precipi-
tation in central and eastern Brazil such as the correct
phase of dry and wet seasons, onset dates, duration of rainy
season and total accumulated precipitation during the
summer monsoon for the twentieth century runs. Never-
theless, poor representation of the total monsoonal
precipitation over the Amazon and northeast Brazil is
observed in a large majority of the models. Overall, MI-
ROC3.2-hires, MIROC3.2-medres and MRI-CGCM3.2.3
show the most realistic representation of SAMS’s charac-
teristics such as onset, duration, total monsoonal
precipitation, and its interannual variability. On the other
hand, ECHAM5, GFDL-CM2.0 and GFDL-CM2.1 have
the least realistic representation of the same characteristics.
For the A1B scenario the most coherent feature observed in
the IPCC models is a reduction in precipitation over cen-
tral-eastern Brazil during the summer monsoon,
comparatively with the present climate. The IPCC models
do not indicate statistically significant changes in SAMS
onset and demise dates for the same scenario.
Keywords South America monsoon system
Climate change Global change
IPCC global coupled climate models A1B scenario
1 Introduction
Tropical South America (SA) is under the influence of a
monsoon regime. Unlike other monsoon systems, easterly
winds dominate during the entire year over northern SA
and tropical Atlantic. Zhou and Lau (1998) demonstrated
that when the annual mean is removed from winter and
summer composites, a clear reverse in the low-level cir-
culation monthly anomalies becomes evident, which
supports the existence of the South America monsoon
system (SAMS). The beginning of the summer monsoon in
SA is characterized by the increase in convective activity
over northwest Amazon that progressively intensifies over
southeast SA (Kousky 1988; Marengo et al. 2001; Lieb-
mann and Marengo 2001; Gan et al. 2004; Vera et al.
2006a). Over central and southeast Brazil the onset of the
rainy season is observed between September and Novem-
ber whereas the demise of the rainy season is observed
between March and April (Gan et al. 2004; Silva and
Carvalho 2007).
Another prominent feature during the rainy season in
tropical SA is the presence of a northwest–southeast
R. J. Bombardi (&) L. M. V. Carvalho
Department of Atmospheric Sciences, University of Sao Paulo,
Rua do Mata
˜
o, 1226, Sao Paulo, SP 05508-090, Brazil
e-mail: bombardi@model.iag.usp.br
L. M. V. Carvalho
Institute for Computational Earth System Science, University
of California Santa Barbara, Santa Barbara, USA
123
Clim Dyn (2009) 33:893–916
DOI 10.1007/s00382-008-0488-1
oriented band of clouds and precipitation that originates in
the Amazon and runs toward the subtropical Atlantic,
which is known as the South Atlantic convergence zone
(SACZ) (Kodama 1992). The SACZ is, therefore, an
important component of SAMS and plays a significant role
in the rainfall variability during the rainy season over
central and southeast Brazil (Liebmann et al. 2001; Carv-
alho et al. 2002a, 2004). Moreover, the monsoon regime in
SA is characterized by large spatial (Carvalho et al. 2002a,
2004) and temporal variability from intraseasonal to
interannual timescales (Kayano and Kousky 1992; Lenters
and Cook 1999; Grimm et al. 1998; Jones and Carvalho
2002; Carvalho et al. 2002b; Vera et al. 2006a; Silva and
Carvalho 2007).
The Earth’s average surface temperature has increased
by 0.6 ± 0.2°C since the late nineteenth century with
global impacts. The response of the climate system to the
rapid increase of greenhouse gases remains uncertain
Intergovernmental Panel on Climate Change (IPCC 2007).
Nevertheless, recent studies have indicated that climate
changes resulting from the increase of CO
2
may affect the
intensity and frequency of extremes temperature and pre-
cipitation in several regions over the globe with large
socio-economical implications (Kharin et al. 2007). Over
SA, modifications in the probability of extremes will have
significant impacts on water resources, endangered eco-
systems, agriculture, and human health (IPCC 2001). In
recent years, the implications of global warming to the
spatiotemporal variability of precipitation in monsoon
regimes have received further attention. Ashrit et al.
(2003), for instance, examined transient climate change
simulations of the CNRM ocean–atmosphere coupled cli-
mate model (CCM) with increase in greenhouse gases.
Their focus was on the Indian summer monsoon and El
Nin
˜
o southern oscillation (ENSO) teleconnections. They
found no clear strengthening of the monsoon circulation
but an increase in the monsoon precipitation likely linked
to large increase in precipitable water over India due to
global warming.
Labraga and Lopez (1997) and Carril et al. (1997) are
some of the earliest studies that have investigated the
impacts on SA precipitation in a future scenario with
double the present CO
2
concentration. Labraga and Lopez
(1997) examined simulations of five general circulation
models coupled to an oceanic model with a single mixing
layer, whereas Carril et al. (1997) evaluated four simula-
tions of early versions of the IPCC coupled models. Both
studies indicate an increase in precipitation over west
Pacific inter-tropical convergence zone (ITCZ) and west
coast of SA as a response to the increase in CO
2
. Giorgi
and Francisco (2000) examined models from the third
assessment report (IPCC 2001), which are an earlier gene-
ration of models. They evaluated five IPCC coupled
models in four distinct future scenarios and verified an
increase of about 10% in precipitation over tropical SA
during December–February season. According to Meehl
et al. (2005), a warmer climate implies a larger availability
of water vapor in the atmosphere and a larger capacity of
the air to retain humidity. With more humidity in the
atmosphere, relatively more intense rainfall and/or poten-
tially strong snowstorms can occur.
Although most IPCC models have improved since the
third assessment report (IPCC 2001), there are still many
uncertainties and discrepancies among models in some
regions. For instance, a large majority of the IPCC models
underestimate precipitation over tropical SA including the
Amazon (Sun et al. 2005; Dai 2006; Vera et al. 2006b). On
the other hand, Vera et al. (2006b) investigated seasonal
precipitation over SA using seven global coupled IPCC
models and observed that most models reproduce the mean
basic characteristics of the annual cycle of precipitation,
such as the seasonal migration of convection over tropical
SA and the maximum precipitation observed over southern
Andes. Nevertheless, models diverge in the location and
intensity of that maximum. Other remarkable discrepancies
discussed in Vera et al. (2006b) are associated with the
SACZ. Some models (GFDL, MIROC ad MRI) represent
the SACZ intensity and location similar to observations,
whereas for a few others the SACZ is displaced north-
eastward of its climatological position or is even absent.
Most previous studies discussed here have focused
mainly on the skill of the global climate models in repre-
senting the mean seasonal intensity, frequency and large-
scale patterns of precipitation in the present climate and
future scenarios of global change due to increase in
greenhouse gases. However, some characteristics of the
monsoon regime such as the onset and duration of the rainy
season and total precipitation, which are essential for water
resource management and agriculture, have not been
properly examined yet. Large variations in these charac-
teristics due to global warming will imply in a significant
socio-economic impacts for all SA counties. In the present
study we examine the ability of ten IPCC global coupled
models in realistically simulating SAMS onset and demise
dates, as well as the total summer monsoon precipitation in
the present climate (1981–2000). Likewise, we investigate
future projections of these models for the A1B scenario
with twice the present CO
2
concentration (2081–2100).
The identification of the performance of individual IPCC
models in simulating SAMS characteristics will be useful
to detect regions of large and poor reliability for the
interpretation of projections in future scenarios of climate
change.
This study is organized as follows: data and models are
presented in Sect. 2. Section 3 discusses the method
applied to define the monsoon onset, demise and total
894 R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations
123
precipitation. Sections 4–7 discuss the IPCC simulations of
the monsoon characteristics for the present climate. Sec-
tion 8 shows the IPCC projections for the A1B scenario.
The main conclusions are presented in Sect. 9.
2 Data
Precipitation data used in this study are 5-day mean (pen-
tad) rainfall from the global precipitation climatology
project (GPCP) from 1979 to 2006. The GPCP pentad is
based on station gauges and satellite estimates with spatial
resolution 2.5° 9 2.5° lat/lon (Xie et al. 2003). The
advantage of using GPCP is its global coverage, including
oceanic areas. In addition, Muza and Carvalho (2006) have
shown that GPCP shows a good correspondence with
gridded precipitation from stations (Liebmann and Allured
2005) in areas over tropical and subtropical Brazil. The
domain examined in the present study extends from 5.0°N–
35.0°S to 30.0°W–80°W.
South America monsoon system characteristics are
examined in ten IPCC CCMs (Table 1) in the same domain
and in two distinct periods: the present climate (i.e. 1979–
2000) and for the A1B scenario (2080–2100). Data were
obtained from the archives of the world climate research
programme’s (WCRP’s) coupled model intercomparison
project phase 3 (CMIP3) multi-model data set. In the A1B
scenario the maximum greenhouse gases emission is
reached in the middle of the twenty-first century and cor-
responds to approximately a double of the present
concentration (i.e. 700 ppmv). Concentrations of CH
2
and
N
2
O in the same period are about 2.0 and 0.37 ppmv,
respectively (e.g. Kharin et al. 2007).
All IPCC models discussed here are coupled models with
precipitation on daily resolution and with integrations
available in the two distinct periods indicated above. In
addition, all models have at least 2.8° spatial resolution. In
order to compare observations and simulations in the present
climate, pentad precipitation was calculated for all models.
There is more than one simulation available for FGOALS-
g1.0, MIROC3.2-medres, and MRI-CGCM2.3.2 models in
both scenarios (Table 1). Therefore, all simulations avail-
able for these three models are considered in this work.
All IPCC models examined here have atmospheric and
oceanic components. However, no model has dynamical
vegetation. Only CNRM-CM3 has ozone transport and
simplified atmospheric chemistry reactions (Cariolle and
De
´
que
´
1986; Cariolle et al. 1990). With the exception of
GFDL-CM2.0 and GFDL-CM2.1 that do not include
aerosols, all other models include some type of aerosol, in
general sulfates. MIROC3.2-hires and MIROC3.2-medres,
and MPI-ECHAM5 include indirect effects of the aerosols.
CGCM3.1(T63) and MRI-CGCM2.3.2 have global flux
adjustment for heat and water and the MRI-CGCM2.3.2
has momentum adjustment between 12°N and 12°S. The
only difference between MIROC3.2-hires and MIROC3.2-
medres is the resolution, whereas GFDL-CM2.0 and
GFDL-CM2.1 differ in the numeric scheme for atmo-
spheric advection (Dai 2006).
3 Method to estimate SAMS onset, end and total
precipitation
The onset, end and duration of SAMS are determined based
on the method adapted from Liebmann and Marengo
(2001). For every grid point, the following summation is
computed:
S pentadðÞ¼
X
pentad
n¼pentad0
RðnÞ
RðÞ ð1Þ
where R(n) is the mean precipitation for the pentad n
(mm/day), and
R is the climatological annual mean daily
precipitation (mm/day). The initial pentad (pentad
0
)is
Table 1 Model description: name, country, spatial resolution, number of simulations for both scenarios twentieth century (20CM) and A1B, and
key reference for each model
Model name Center country Resolution lat 9 lon 20CM runs A1B runs Key reference
CGCM3.1(T63) Canada *2.8 9 2.8 1 1 Flato et al. (2000)
CSIRO-Mk3.0 Australia *1.9 9 1.9 1 1 Gordon et al. (2002)
CNRM-CM3 France *2.8 9 2.8 1 1 Salas-Me
´
lia et al. (2005)
ECHAM5 Germany *1.9 9 1.9 1 1 Roeckner et al. (2003)
FGOALS-g1.0 China *2.8 9 2.8 3 3 Yu et al. (2004)
GFDL-CM2.0 USA 2.0 9 2.5 1 1 Delworth et al. (2006)
GFDL-CM2.1 USA 2.0 9 2.5 1 1 Delworth et al. (2006)
MIROC3.2-hires Japan *1.125 9 1.125 1 1 Hasumi and Emori (2004)
MIROC3.2-medres Japan *2.8 9 2.8 2 2 Hasumi and Emori (2004)
MRI-CGCM2.3.2 Japan *2.8 9 2.8 5 5 Yukimoto et al. (2006)
R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations 895
123
always taken within the dry season such that the onset is
never missed (Liebmann and Marengo 2001). In the pres-
ent study the initial pentad is computed for every grid point
as the pentad in which the minimum climatological daily
rainfall is observed. Figure 1 shows an example of the
application of Eq. 1 to estimate the onset and end of the
rainy season. During the dry season S (Eq. 1) shows neg-
ative slope. As the wet season starts rainfall becomes more
regular and, consequently, the S slope becomes steadily
positive. When daily precipitation decreases and/or
becomes less regular the S slope becomes steadily negative
again, which indicates that the dry season has begun. In
order to accurately estimate the onset and demise of
SAMS, S is further smoothed
SðÞby applying a moving
average filter and the first derivative d
S
=
dt is then com-
puted (Fig. 1). SAMS onset (demise) is assumed to be the
pentad when d
S
=
dt ¼ 0; followed by the pentad when
d
S
=
dt [ 0d
S
=
dt\0ðÞ: The total precipitation during the
summer monsoon is the total precipitation that is accu-
mulated from the onset to the demise dates.
4 Summer daily precipitation
We start our discussion about the performance of the IPCC
models in simulating the characteristics of SAMS in the
present climate (hereafter referred to as 20CM) by showing
the observed and simulated daily mean and standard
deviation rainfall (mm/day) during the peak of the rainy
season (December–February, DJF) (Fig. 2). We recall that
all simulations available for FGOALS-g1.0, MIROC3.2-
medres, and MRI-CGCM2.3.2models (Table 1) are con-
sidered in the analysis. The GPCP (Fig. 2a) indicates the
presence of a NW–SE oriented band with daily precipita-
tion above 3 mm/day extending from the Amazon toward
subtropical Atlantic, which has been characterized as the
SACZ (e.g. Carvalho et al. 2002a). Over the continent, the
maximum precipitation (*9 mm/day) is observed over
central Amazon, consistent with Carvalho et al. (2004).
Another region with maximum daily precipitation is
observed in association with the Atlantic ITCZ. In addition,
the SACZ and the ITCZ (Fig. 2
a) are both related to large
precipitation variance as the result of the interplay of
phenomena occurring in a broad range of timescales (e.g.
Grimm et al. 2000; Carvalho et al. 2002b; Silva et al. 2006).
The IPCC models (Fig. 2b–k), in general, capture the
main spatial patterns of the summer (DJF) mean daily
precipitation, such as the precipitation maxima over the
continent and in association with the ITCZ and SACZ.
Nevertheless, most IPCC models simulate the continental
maximum displaced southeastward of its actual position,
approximately co-located with the Brazilian highland.
These results are consistent with the patterns of seasonal
mean precipitation shown in Lambert and Boer (2001) and
Vera et al. (2006b). The displacement is particularly
noticeable for ECHAM5 (Fig. 2e) that misplaces the
maximum precipitation toward west tropical Atlantic, as an
extension of the Atlantic ITCZ. In addition, ECHAM5
(Fig. 2e) does not reproduce an NW–SE oriented band of
precipitation observed in all other IPCC models examined
here. These characteristics are particular of this version of
the model in contrast with previous versions (e.g.
ECHAM4.5) that show a much more realistic representa-
tion of SAMS (Liebmann et al. 2007). Another issue in the
representation of the seasonal precipitation in the IPCC
models is the unrealistic double ITCZ pattern observed for
GFDL-CM2.0 (Fig. 2g), GFDL-CM2.a (Fig. 2h) and MI-
ROC3.2-hires (Fig. 2i) (Dai 2006). MIROC3.2-medres
(Fig. 2j) and MRI-CGCM2.3.2 (Fig. 2k) (both with 2.8°
resolution) show an unrealistic wide ITCZ. The other
feature simulated in all models (Fig. 2b–k) that is not
observed in GPCP (Fig. 2a) is the maximum precipitation
along the Andes.
Nevertheless, the IPCC models that realistically simu-
late the mean daily DJF precipitation over the convergence
zones simulate fairly well the observed precipitation vari-
ability, represented here by the standard deviation of daily
DJF precipitation and indicated by shades in Fig. 2.
Overall, MIROC3.2-hires (Fig. 2i) has the best represen-
tation of the SACZ in regard to its mean summer daily
Fig. 1 Example of application of the method adapted from Liebmann
and Marengo (2001) to determine the onset and demise of the rainy
season. Dashed line represents the S estimate (Eq. 1); continuous line
indicates the smoothed S estimate (three point moving average passed
over 40 times); doted line indicates the first derivative of the
smoothed S estimate. Curves are shown as a function of time starting
from the initial pentad (pentad
0
) and the ordinate represents the
accumulated precipitation anomaly (mm/pentad). Changes in the
slope of S (dS/dt) define the onset (dS/dt [ 0) and demise (dS/dt \ 0)
of the rainy season
896 R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations
123
Fig. 2 Daily mean precipitation
(contours) during the peak of
the rainy season (December–
February) and standard
deviation (mm/day) (shading).
Contour interval is equal 3 mm/
day. Dark gray indicates
standard deviation above 5 mm/
day and light gray shows
standard deviation between 3
and 5 mm/day. Regions with no
shading indicate standard
deviation below 3 mm/day.
a GPCP; b CGCM3.1(T63);
c CNRM-CM3; d CSIRO-Mk3.0;
e ECHAM5; f FGOALS-g1.0;
g GFDL-CM2.0; h GFDL-
CM2.1; i MIROC3.2-hires;
j MIROC3.2-medres;
k MRI-CGCM2.3.2
R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations 897
123
precipitation and standard deviation features. For instance,
MIROC3.2-hires is the only model among the ten models
that simulates mean daily DJF precipitation between 6 and
9 mm/day over southeast Brazil, with standard deviation
comparable to observations. Moreover, the maximum daily
precipitation in MIROC3.2-hires (*9 mm/day) is dis-
placed toward central Amazon in agreement with GPCP.
These features are reasonably well represented in GFDL-
CM2.0 (Fig. 2g), GFDL-CM2.1 (Fig. 2h), and MIROC3.2-
medres (Fig. 2j). FGOALS-g1.0 (Fig. 3f) and ECHAM5
(Fig. 3e) on the other hand, show a very poor representa-
tion of the SACZ and ITCZ with respect to both mean
precipitation and standard variation.
5 Precipitation annual cycle
Realistic simulations of SAMS characteristics depend on
the ability of models to reproduce the observed precipita-
tion annual cycle. Figure 3 shows the climatological annual
Fig. 3 Observed and simulated
annual cycle of precipitation for
the 20CM run. The mean annual
cycle is obtained in an area
corresponding to 5° latitude/
longitude for the regions
a northwestern SA, b western
Amazon, c Amazon mouth,
d southern Amazon, e central
Amazon, f central Brazil and
g SACZ over South Atlantic
898 R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations
123
cycle of rainfall for observation and models in distinct
areas over tropical SA and over the Atlantic Ocean as
indicated by boxes in Fig. 3a. Areas were selected with the
objective to verify the ability of the models in simulating
distinct precipitation regimes in the following regions:
northwest SA (2.5°N, 72.5°W), Amazon mouth (0.0°S,
50.0°W), central Amazon (5.0°S, 60.0°W), western Ama-
zon (70.0°W, 5.0°S), southern Amazon (10.0°S, 55.0°W),
central Brazil (17.5°S, 50.0° W) and the oceanic portion of
the SACZ (30.0°S, 37.5°W). The latter, indicated by the
box over subtropical Atlantic (Fig. 3a), corresponds
approximately to the region where high variance in daily
outgoing long-wave radiation (OLR) is observed in asso-
ciation with the SACZ (e.g. Carvalho et al. 2004). For this
discussion, the average annual cycle in each box preserved
the original spatial resolution of the models such that we
can identify the role of increasing resolution in different
versions of the models (as in the case of MIROC3.2-hires
and MIROC3.2-medres).
The observed (GPCP) mean annual cycle of precipita-
tion over northwest Amazon (Fig. 3b) shows maximum
precipitation in June (*pentad 31) and minimum in Jan-
uary (*pentad 1). Figure 3b indicates a large dispersion
among IPCC models for the simulation of the mean annual
cycle in that region. GFDL-CM2.1 and GFDL-CM2.0, for
instance, simulate an annual cycle about 6 months out-of-
phase with respect to observations, with minimum pre-
cipitation in June and maximum in January. An unrealistic
spring and fall peak in the annual cycle is also observed for
many models and is more pronounced for MRI-
CGCM2.3.2. MIROC3.2-hires and MIROC3.2-medres
capture the correct phase of the annual cycle, although
MIROC3.2-hires overestimates (underestimates) precipi-
tation in the rainy (dry) season comparatively to
observations. With the exception of MIROC3.2-hires all
other models underestimate precipitation in the rainy sea-
son. The large dispersion in the results and also the double
peak in the annual cycle are likely related to the misrep-
resentation of the Pacific and Atlantic ITCZ and their
seasonal variability in most models investigated here (see
Fig. 2).
Near the Amazon mouth (Fig. 3c), western (Fig. 3d) and
central (Fig. 3f) Amazon a large dispersion among models
for the simulations of the annual cycle of precipitation is
also observed. The dispersion decreases and the simulated
phase of the annual cycle approaches observation over
southern Amazon (Fig. 3g), perhaps as the result of a
weaker influence of the ITCZ. Over the Amazon mouth
(Fig. 3c) all other models show an out-of-phase maximum
of precipitation and underestimate the observed precipita-
tion during the peak of the rainy season.
The dispersion among model simulations of the pre-
cipitation annual cycle decreases dramatically over central
Brazil (Fig. 3g) and SACZ (Fig. 3h). All models are
capable of correctly identifying the wet and dry seasons
over central Brazil (Fig. 3g) where SAMS has its large
signal (Silva and Carvalho 2007). However, all models
maintain a negative bias during the dry season. During the
wet season, different models present either positive or
negative bias with respect to observations. Over the oce-
anic SACZ (Fig. 3h), observations show a low amplitude
annual cycle, with minimum precipitation during SH
winter. All models correctly simulate the low seasonal
variability, but with amplitudes that are lower than
observed. Some models such as GFDL-CM2.0 and GFDL-
CM2.1 show no seasonal variation in the annual cycle in
that region.
6 SAMS onset, end, duration and total precipitation
The onset, end and duration of SAMS were estimated for
every grid point and for every season by applying the
Liebmann and Marengo (2001) method (Eq. 1). The total
summer monsoon precipitation was also computed for
every season as the total rainfall accumulated between the
onset and demise of SAMS. Medians were used to describe
the features of the 20CM and A1B experiments to avoid
assumptions regarding the actual distributions of the vari-
ables and provide information about the central value that
is less influenced by extreme values (e.g. Wilks 2006). The
onset, demise, duration and total precipitation were com-
puted separately for each simulation of FGOALS-g1.0,
MIROC3.2-medres and MRI-CGCM2.3.2, whereas the
statistical analyses were performed considering all model
simulations together.
6.1 Total summer monsoon precipitation
The spatial variability of the median total precipitation
during the summer monsoon for GPCP data is shown in
Fig. 4a. Large median total precipitation ([1,600 mm) is
observed over the Brazilian Amazon and over the Atlantic
and Pacific ITCZ. A local maximum ([1,800 mm) is
observed near the Amazon mouth (Fig. 4a). These obser-
vations are consistent with Marengo et al. (2001) and
Liebmann and Marengo (2001) that show large total annual
rainfall over northeast Amazon and near the Amazon
mouth. Liebmann and Marengo (2001) using rain gauge
data showed that the annual mean precipitation varies more
than 50% within the Amazon basin with totals of
2,000 mm over south, east and north of the Amazon, and
3,000 mm over northwest Amazon, where relatively high
topography is observed. They also identified a secondary
maximum near the Amazon mouth, which they have rela-
ted to the nocturnal convergence of the trade winds with
R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations 899
123
Fig. 4 Median total
precipitation during the rainy
season for a GPCP;
b CGCM3.1(T63); c CNRM-
CM3; d CSIRO-Mk3.0;
e ECHAM5; f FGOALS-g1.0;
g GFDL-CM2.0; h GFDL-
CM2.1; i MIROC3.2-hires;
j MIROC3.2-medres and
k MRI-CGCM2.3.2. Contour
interval equals 200 mm. Light
gray shading shows regions
with 1200 mm B median \
1,600 mm and dark gray
shading shows regions with
median C 1,600 mm
900 R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations
123
the sea breeze. Over central Brazil, the median precipita-
tion decreases, with totals between 1,000 and 1,400 mm.
The southeastward displacement of the 1,200 and 800 mm
isohyets indicates the importance of the SACZ.
For the 20CM simulations, large discrepancies are
observed over the Amazon (Fig. 4b–k) as diagnosed in
previous studies (Sun et al. 2005; Dai 2006; Vera et al.
2006b). For instance, CGCMT63 (Fig. 4b) underestimates
in about 400 mm and CSIRO-Mk3.0 (Fig. 4d) in more than
600 mm the total median precipitation over the entire
Amazon. In addition CSIRO-Mk3.0 displaces the maxi-
mum precipitation (*1,200 mm) toward central Brazil.
CNMR (Fig. 4c) and ECHAM5 (Fig. 4e) are not capable of
correctly identifying the precipitation regime over north-
east Brazil (known as ‘‘nordeste’’) and distinguish it from
SAMS and ITCZ, which results in unrealistic large summer
precipitation in that region. Moreover, CNRM-CM3
(Fig. 4c) overestimates precipitation over central Brazil by
about 600 mm. Both GFDL-CM2.0 (Fig. 4g) and GFDL-
CM2.1 (Fig. 4h) show a very poor representation of
the total summer monsoon precipitation. MIROC3.2-
hires (Fig. 4i) simulates the maximum precipitation
([1,600 mm) over western Amazon around its actual
position (Fig. 4a) and shows a secondary maximum over
northern Brazil, displaced eastward from the maximum
observed with GPCP over the Amazon mouth. MIROC3.2-
medres (Fig. 4j) shows a good representation of the actual
pattern of the summer monsoon precipitation. MRI-
CGCM2.3.2 (Fig. 4k) realistically represents the total
precipitation over northeast Brazil and captures the main
spatial features of SAMS.
6.2 Variability of the summer monsoon precipitation
The ability of each model to capture the actual variability
of the rainfall is essential to assess possible climate chan-
ges in future scenarios. This issue was investigated here by
computing the median absolute deviation (MAD), which
measures the residuals (deviations) from the data median
and is defined as:
MAD ¼ median y
it
~
y
i
jj
ðÞ; ð2Þ
where y
it
is the total summer monsoon precipitation for the
year t at a given grid point i and
~
y
i
is the median of the
summer monsoon precipitation at the same grid point.
MAD provides a good measure of the scale parameter of
unknown distributions in substitution to standard devia-
tions and is less affected by extremes (Wilks 2006).
Figure 5 shows MAD for the simulations of the total
summer monsoon precipitation discussed in Fig. 4. Large
(small) MAD can be interpreted as regions with high (low)
interannual variability of SAMS precipitation. Large
interannual variability in precipitation ([200 mm) is
observed with GPCP data (Fig. 5a) over northern SA and
southern Brazil, Uruguay, and northeast Argentina. This
pattern of interannual variability is consistent with the
modulation of precipitation by ENSO phenomenon (e.g.
Grimm et al. 2000; Coelho et al. 2002; Magan
˜
a and Am-
brizzi 2005) on interannual timescales. Relatively low
interannual variability (\200 mm) is observed over central
SA, where large seasonal variations of precipitation, low-
level circulation and humidity are observed in association
with SAMS (Silva and Carvalho 2007). In addition, low
interannual variations occur over subtropical Atlantic and
Pacific near the west coast of SA.
CNRM-CM3 (Fig. 5c) is the only model that simulates
high interannual variability of total precipitation
([200 mm) in association with the model’s SACZ (com-
pare with Fig. 2c). All other models show relatively low
variability of precipitation in association with the SACZ
and over the SAMS core in center Brazil (Silva and
Carvalho 2007). MIROC3.2-hires (Fig. 5i) shows the most
realistic pattern of MAD over SA. MIROC3.2-medres
(Fig. 5j) shows less interannual variability (\200 mm) in
basically all SA, with the exception of a small area over the
Amazon and central east Brazil. All models tend to
increase the interannual variability of precipitation over
northern SA and ITCZ likely as a response to the model’s
ENSO. Nevertheless, only ECHAM5 (Fig. 5e) and MI-
ROC3.2-hires (Fig. 5i) show a large MAD over southern
Brazil in agreement with observations (Fig. 5a).
6.3 SAMS onset
The present analysis investigates the spatial variability of
the median SAMS onset over tropical SA. For this purpose,
the onset is schematically indicated with arrows in Fig. 6.
The direction of the arrows indicates the period of the year
when the onset occurs for a given grid point according to
the circle at the bottom of Fig. 6. The GPCP data indicates
that over central and southeast Brazil, the median onset is
observed between pentads 58–61 (approximately mid to
end of October, respectively), consistent with Silva and
Carvalho (2007) (Fig. 6a). The median onset date varies
rapidly toward northern SA, with the rainy season begin-
ning between early November to late December, in
agreement with observations in Liebmann and Marengo
(2001), Lincoln et al. (2005) and Seth et al. (2007). Most
IPCC models simulate correctly the median onset between
pentad 58 and 61 over central Brazil (Fig. 6b–k). The
overall agreement among models is consistent with the
observation that the phase of the annual cycle is well
captured by all models over this region (Fig. 3g). Never-
theless, over southeast Brazil, where the SACZ manifests,
there is a larger dispersion among models, with simulated
onsets varying from early October to late November
R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations 901
123
Fig. 5 MAD of total
precipitation during the rainy
season for a GPCP;
b CGCM3.1(T63); c CNRM-
CM3; d CSIRO-Mk3.0;
e ECHAM5; f FGOALS-g1.0;
g GFDL-CM2.0; h GFDL-
CM2.1; i MIROC3.2-hires;
j MIROC3.2-medres and
k MRI-CGCM2.3.2. Contour
interval is equal to 100 mm.
Light gray shading shows
regions where
100 mm B MAD \ 200 mm
902 R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations
123
Fig. 6 Median onset of the
rainy season for a GPCP;
b CGCM3.1(T63); c CNRM-
CM3; d CSIRO-Mk3.0;
e ECHAM5; f FGOALS-g1.0;
g GFDL-CM2.0; h GFDL-
CM2.1; i MIROC3.2-hires;
j MIROC3.2-medres; k MRI-
CGCM2.3.2. Arrows indicate
the time of the year when the
onset occurs, according to the
circle indicated at the bottom of
the figure. The size of the
arrows varies only due to the
resolution of the model and has
no meaning. Shading indicates
regions where the median onset
date is greater than pentad 58
(13–17 October) and less than
pentad 62 (2–6 November)
R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations 903
123
(i.e. pentad 56–64). Furthermore, this analysis emphasizes
the misrepresentation of the actual onset of the monsoon over
north and western Amazon and near the Amazon mouth as
suggested by the annual cycles in Fig. 3b–d, respectively.
A good performance of some models in simulating the
actual median precipitation does not imply a good repre-
sentation of SAMS onset. For instance, MIROC3.2-hires
shows one of the most realistic representations of SAMS
median precipitation features (Fig. 4i). However, this
model shows early onsets (between pentad 56 and 58) from
south Amazon toward southeast Brazil, in addition to a
fairly unrealistic representation of onsets over northern
Amazon (Fig. 6i). On the other hand, ECHAM5 does not
correctly simulate the SACZ and underestimates SAMS
precipitation (Fig. 4e), but captures the onset of the rainy
season over central Brazil (Fig. 6e). FGOALS-g1.0
(Fig. 6f), and MRI-CGCM2.3.2 (Fig. 6k) show the earliest
onsets for the rainy season over central Amazon and
southeastern Brazil (between pentad 56 and 58).
A remarkable unrealistic representation of the onset of
the rainy season over north Amazon by all IPCC models is
clearly evident (compare Fig. 6a with Fig. 6b–k). For
instance, the out-of-phase onsets of the rainy season over
north Amazon (Fig. 3b) and over the Amazon mouth
(Fig. 3c) in models such as GFDL-CM2.0 and GFDL-
CM2.1 appears as a large feature over northern SA
(Fig. 6g, h).
The interannual variability of onsets is examined here by
computing MAD (Eq. 2) and replacing total precipitation
with the onset dates (Fig. 7). The GPCP data (Fig. 7a)
indicate that from 1979 to 2006 the onset of the monsoon
over central Brazil showed median variability of about two
pentads. MAD increases to four pentads over northern
Amazon, southern Brazil and Atlantic SACZ. Large MAD
is observed over northern Amazon and southern Brazil,
Uruguay and northeastern Argentina (Fig. 7a), likely as a
response to ENSO. Most IPCC models show small MAD
over central Brazil (Fig. 7b–k). In general, most IPCC
models tend to increase MAD over north–northwest SA in
association with interannual variations in the model’s
ITCZ. Large variations are also observed over the sub-
tropics of SA. Among all models, MIROC3.2-hires
(Fig. 7i) and MRI-CGCM2.3.2 (Fig. 7k) show patterns of
the onset MAD that are the most consistent with observa-
tions over central Brazil (compare with Fig. 7a).
6.4 SAMS demise and duration
The median end of the rainy season (Eq. 1) over central
and southeast Brazil occurs between pentad 18–21 (end
March to mid-April) as intense precipitation gradually
migrates from south Amazon and central Brazil toward the
equator (Kousky 1988; Marengo et al. 2001; Gan et al.
2004; Vera et al. 2006a; Silva and Carvalho 2007) (not
shown). During SAMS demise, convection associated with
the Atlantic ITCZ weakens (Vera et al. 2006a). Due to the
simulation of a stronger than observed and/or double ITCZ,
the IPCC models tend to simulate early demises of the
rainy season over northeast Brazil (between 4 and 6 pent-
ads, not shown). CSIRO-Mk3.0, for instance, indicates the
demise of the rainy season as early as four pentads over
central Brazil. GFDL-CM2.0 and FGOALS-g1.0 show
relatively less difference, between -1 and 1 pentad over
the same region, although both GFDL-CM2.0 and GFDL-
CM2.1 do not realistically simulate the pattern of demise
dates over SA, showing basically the same dates for the
north and central Brazil regions.
The median duration of SAMS observed with GPCP
(Fig. 8a) is between 32 and 36 pentads over most SA,
with long durations toward the center of the continent.
The IPCC simulations for the present climate tend to
overestimate the duration of the rainy season over west
SA and underestimate over central Brazil (Fig. 8). Mod-
els such as FGOALS-g1.0 (Fig. 8f), MIROC3.2-hires
(Fig. 8i) and MRI-CGCM2.3.2 (Fig. 8k) are skillful in
representing the patterns of duration of the rainy season
over tropical SA, with less difference with respect to
observation over Central Brazil (between 0 and 2 pent-
ads). On the other hand, GFDL-CM2.0 (Fig. 8g) and
GFDL-CM2.1 (Fig. 8h) show the worst performance in
simulating the patterns of duration of SAMS, underesti-
mating the duration of the monsoon in about four pentads
over central Brazil.
Median absolute deviation computed for SAMS dura-
tion is shown in Fig. 9. We recall that the duration of the
rainy season depends on the onset and demise of the season
and MAD quantifies the yearly variations of these quanti-
ties. The duration of the rainy season based on GPCP data
(Fig. 9a) shows less variability over central tropical SA (2–
4 pentads). In general, MAD over the core of monsoon
region (Silva and Carvalho 2007) is below four pentads for
all models. MIROC3.2-hires (Fig. 9i), MIROC3.2-medres
(Fig. 9j) and MRI-CGCM2.3.2 (Fig. 9k) show the most
realistic spatial patterns of MAD over central Brazil com-
pared to observations. The largest differences between
observation and simulations of SAMS duration is simulated
by ECHAM5 (Fig. 9e) and FGOALS-g1.0 (Fig. 9f), which
underestimate the interannual variability over tropical SA,
whereas GFDL-CM2.0 and GFDL-CM2.1 (Fig. 9h) over-
estimate the variability.
7 Summary of SAMS Simulation in the 20CM scenario
To objectively evaluate the skill of the IPCC models in
reproducing SAMS spatial pattern of the median and
904 R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations
123
Fig. 7 Same as in Fig. 5, but
for MAD of onset of the rainy
season. Line interval is equal to
two pentads. Light gray shading
shows regions where two
pentads \ MAD B four
pentads and greater than two
pentads and dark gray shading
shows regions where MAD B 2
pentads
R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations 905
123
Fig. 8 Same as in Fig. 4, but
for median of duration of the
rainy season. The line interval is
equal to two pentads. Light gray
shading shows regions where 32
pentads B median \ 36
pentads and the dark gray
shading shows regions with
median C 36 pentads
906 R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations
123
Fig. 9 Same as in Fig. 5, but
for MAD of duration of the
rainy season. Contour interval is
equal to two pentads. Light gray
shading shows regions where
two pentads \ MAD B four
pentads dark gray shading
shows regions where the
MAD B two pentads
R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations 907
123
respective MAD of onset, demise, duration and total pre-
cipitation in the 20CM simulations, spatial correlation (SC)
and spatial root mean square error (RMSE) between model
simulations and GPCP were computed. The SC is used as a
measure of the spatial co-variability between IPCC model
simulations and GPCP. The spatial RMSE indicates the
average absolute error between the median and MAD
simulations and GPCP and complements the SC analysis.
For this purpose, the 20CM maps of median and MAD
onset, demise, duration, and total precipitation for both
GPCP and IPCC models were re-gridded to a common
latitude/longitude resolution. In the cases where models
have finer resolution than GPCP, models median and MAD
were re-gridded to a 2.5° latitude/longitude resolution;
otherwise, GPCP median and MAD were re-gridded to a
2.8° latitude/longitude resolution. The domain used in this
analysis extends from 23.5°S–7.5°S to 62.5°W–25.0°W,
which corresponds approximately to the region where
SAMS and SACZ have their maximum variability (Silva
and Carvalho 2007). This domain does not include the
Andes to avoid comparisons with the unrealistic precipi-
tation patterns simulated in these high elevations by all
models (Marengo 2003). In addition, it was intentionally
positioned southward of the climatological position of the
ITCZ indicated by GPCP (e.g. Fig. 3a) due to the mis-
representation of this convergence zone by most models.
The total number of grid points in these analyses is 72 (91)
for the coarser (finer) resolution. Table 2 summarizes the
results.
High SC (i.e. SC above 0.7) is observed for the large
majority of models for the median onset and demise.
However, lower SC is observed for duration, with only two
models (MRI-CGCM2.3.2 and ECHAM5) indicating values
above 0.7 (SC * 0.71 and 0.74, respectively). High SC
([0.7) is observed for total precipitation for most models,
with the worst performance observed for ECHAM5
(SC * 0.15). MIROC3.2-hires shows the best skill in
reproducing the spatial patterns of total precipitation
(SC * 0.92) followed by MIROC3.2-medres (SC * 0.90)
and MRI-CGCM2.3.2 (SC * 0.83). Nevertheless, large
Table 2 Spatial correlation (top) and spatial RMSE (bottom) between IPCC models simulation and GPCP for median (right) and MAD (left) of
onset, demise, duration, and total precipitation
Model Spatial correlation (median) Spatial correlation (MAD)
Onset
(pentads)
Demise
(pentads)
Duration
(pentads)
Total precipitation
(mm)
Onset
(pentads)
Demise
(pentads)
Duration
(pentads)
Total precipitation
(mm)
CGCM3.1(T63) 0.834 0.441 0.660 0.798 0.531 0.517 0.577 0.269
CNRM-CM3 0.788 0.434 0.415 0.762 0.546 0.314 0.609 0.087
CSIRO-Mk3.0 0.785 0.671 0.577 0.830 0.252 0.486 0.621 0.322
ECHAM5 0.926 0.673 0.738 0.153 0.163 0.269 0.232 -0.151
FGOALS-g1.0 0.788 0.765 0.509 0.841 0.581 0.269 0.656 0.180
GFDL-CM2.0 0.872 0.741 0.432 0.631 0.394 0.398 0.649 0.128
GFDL-CM2.1 0.884 0.835 0.275 0.706 0.445 0.368 0.582 0.004
MIROC3.2-hires 0.806 0.720 0.561 0.921 0.609 0.500 0.816 0.591
MIROC3.2-medres 0.903 0.805 0.375 0.903 0.547 0.553 0.787 0.342
MRI-CGCM2.3.2 0.820 0.821 0.711 0.839 0.641 0.698 0.826 0.346
Spatial RMSE (median) Spatial RMSE (MAD)
Onset
(pentads)
Demise
(pentads)
Duration
(pentads)
Total precipitation
(mm)
Onset
(pentads)
Demise
(pentads)
Duration
(pentads)
Total precipitation
(mm)
CGCM3.1(T63) 10.3 9.9 3.4 298.5 2.7 4.9 2.7 300.5
CNRM-CM3 8.0 10.3 4.6 489.7 3.0 5.4 3.2 592.8
CSIRO-Mk3.0 24.0 12.1 3.7 276.4 3.6 5.1 2.8 312.8
ECHAM5 26.9 10.9 3.1 517.1 3.8 5.8 3.9 418.2
FGOALS-g1.0 10.4 9.8 4.4 287.4 2.7 6.2 2.6 380.4
GFDL-CM2.0 22.2 16.0 5.0 419.9 3.5 6.7 2.6 408.6
GFDL-CM2.1 14.2 13.2 4.9 443.5 3.9 7.3 3.2 441.9
MIROC3.2-hires 16.1 9.2 3.8 192.6 3.0 4.9 2.3 308.0
MIROC3.2-medres 6.9 7.4 4.3 255.3 2.9 4.5 2.0 315.4
MRI-CGCM2.3.2 9.7 9.5 3.5 254.8 2.4 3.7 1.8 280.5
908 R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations
123
discrepancies are observed among models with respect to
the skill in reproducing the spatial pattern of MAD. This is
particularly remarkable for MAD of total precipitation. The
best performance in this case is observed for MIROC3.2-
hires (SC * 0.591). Among all IPCC models investigated
here, MIROC3.2-hires, MIROC3.2-medres and MRI-
CGCM2.3.2 are the most skilful in reproducing the overall
spatial variation of median and MAD characteristics of
SAMS.
The RMSE observed between median and MAD simu-
lations and observations is, in all cases, very high for all
models (Table 2). This is not surprising given that the
RMSE averages the square of the difference of the mag-
nitudes between observations and simulations grid point to
grid point. Therefore, the analysis of RMSE should be
considered along with SC as an indication of the skill of a
model in reproducing SAMS spatial characteristics. Thus,
taking these two parameters into account, MIROC3.2-
hires, MIROC3.2-medres and MRI-CGCM2.3.2 are the
models that best characterize the spatial patterns of the
median and MAD of onset, demise, duration and total
precipitation. They show relatively high SC and low RMSE
for most medians and MAD. Using the same rationale, the
worst performance is observed for ECHAM5, GFDL-
CM2.0 and GFDL-CM2.1.
8 Projections for the A1B scenario
The same methodology applied for the 20CM simulations
was used for the IPCC models in the A1B scenario. Median
SAMS onset, demise, duration and total precipitation
simulated for the A1B scenario were compared with the
respective simulation for the 20CM run. The comparison
was carried out by examining differences between medians
obtained for the 20CM and A1B scenario.
Differences in median onsets are between 0 and 1 pentad
over central Brazil, the core of SAMS (Fig. 10). In addi-
tion, there is no spatial coherence in the regions that passed
the test of difference of medians at 5% significance level in
this region. Over northern Amazon, however, MIROC3.2-
hires (Fig. 10h), MIROC3.2-medres (Fig. 10i) and MRI-
CGCM2.3.2 (Fig. 10j) indicate earlier onsets for the A1B
scenario. Other models such as CGCMT63 (Fig. 10a),
ECHAM5 (Fig. 10d) and FGOALS-g1.0 (Fig. 10e) indi-
cate a delay in the rainy season of 1–2 pentads over the
same region, but with no statistical significance at 5%
level.
Differences in the date of the median demise are near
zero over central Brazil for all models (Fig. 11). MRI-
CGCM2.3.2 is the only model that simulates early onsets in
the A1B scenario with statistical significance at 5% level
over a large fraction of the La Plata Basin in SA (Fig. 11j).
All models indicate shorter duration of the rainy season
over the monsoon region (Fig. 12). However, these dif-
ferences are not statistically significant in any region and
for any model (Fig. 12). Likewise, there is no statistically
significant difference in MAD and inter-quartile range over
most of tropical SA (not shown). What these results indi-
cate is that the characteristics of the distributions of onset,
demise and duration of SAMS for the 20CM and A1B
scenario are very similar.
On the other hand, distinct conclusions can be drawn
with respect to total monsoonal precipitation. Figure 13
shows the difference between the median total precipitation
for the A1B and 20CM scenarios for all IPCC models.
Regions with statistically significant differences of medi-
ans are observed for all models. Nevertheless, there are
remarkable discrepancies among IPCC models. For
instance, CNRM-CM3 (Fig. 13a) shows statistically sig-
nificant differences of about 250 mm over central Brazil,
whereas MIROC3.2-medres (Fig. 13i) shows differences of
the same magnitude but with opposite sign in the same
region for A1B with respect to 20CM simulations. None-
theless, six out of ten models examined here show a
decrease of total monsoonal precipitation over central-east
Brazil for the A1B scenario, with differences ranging from
-300 to -50 mm depending on the model. These models
are CSIRO-Mk3.0 (Fig. 13c), FGOALS-g1.0 (Fig. 13e),
GFDL-CM2.0 (Fig. 13
f), GFDL-CM2.1 (Fig. 13g), MI-
ROC3.2-hires (Fig. 13i) and MIROC2.2-medres (Fig. 13j).
Furthermore, in four models (FGOALS-g1.0, GFDL-
CM2.0, MIROc3.2-medres and MIROC3.2-hires) differ-
ences are statistically significant at 5% level. These results
are important because they indicate consistency in the
simulations of IPCC models for the A1B scenario with
respect to 20CM simulations. More importantly, they
identify a reduction of the monsoonal precipitation in a
future scenario of global change. Since FGOALS-g1.0,
MIROC3.2-medres and particularly MIROC3.2-hires are
models with high skill in representing the total monsoonal
precipitation over central and eastern Brazil, a decrease in
the total precipitation for the A1B in this region is likely a
robust feature.
As discussed in the introduction and according to Meehl
et al. (2005) a warmer planet could result in the enhancement
of the potential for intense precipitation events and more
snowstorms. Consistently with the present study, these
authors also observed a decrease in the mean daily precipi-
tation for central-eastern Brazil. A possible explanation for
this apparent paradox may be found in Tebaldi et al. (2007).
They verified that the IPCC models show a consistent signal
of longer consecutive dry days and an increase in the rainfall
intensity for the same region in a scenario of global change.
Our results indicate that for most models and for central-
eastern Brazil no statistically significant differences in the
R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations 909
123
median onset, demise and duration are observed for the A1B
scenario with respect to the present climate. These combined
projections suggest that although rainfall intensity might
increase in a future scenario of global change, the total
monsoonal precipitation could decrease due to the increase
in the number of dry days.
Fig. 10 Differences between
the median of onset of the rainy
season simulated for the A1B
scenario and simulated for the
20CM scenario for each model:
a CNRM-CM3; b CGCMT63;
c CSIRO-MK3.0; d ECHAM5;
e FGOALS-g1.0; f GFDL-
CM2.0; g GFDL-CM2.1;
h MIROC3.2-hires;
i MIROC3.2-medres and j MRI-
CGCM2.3.2. Contour interval
equal one pentad. Solid lines
indicate positive values and
dashed lines indicate negative
values. Shading shows regions
where the difference is
statistically significant at 5%
level
910 R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations
123
Another remarkable feature indicated by MIROC3.2-
hires is the increase in the monsoonal precipitation over
North Brazil with maximum of more than 400 mm at the
coast of Maranha
˜
o and Para
´
states in Brazil. These features
are observed in a region where MIROC3.2-hires shows a
maximum in total precipitation for the 20CM run that is
Fig. 11 Same as in Fig. 10, but
for median of demise of the
rainy season. Contour interval
equal one pentad. Solid lines
indicate positive values and
dashed lines indicate negative
values. Shading shows regions
where the difference is
statistically significant at 5%
level
R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations 911
123
misplaced eastward of the actual observations (Fig. 4j). The
increase in total precipitation over southern Brazil, Uruguay
and Argentina in most models that show a decrease of total
precipitation over eastern Brazil is consistent with obser-
vations of a seesaw in precipitation on several timescales
reported in many previous studies (e.g. Vera et al. 2006a).
Fig. 12 Same as in Fig. 10, but
for median of duration of the
rainy season. Contour interval
equal one pentad. Solid lines
indicate positive values and
dashed lines indicate negative
values. Shading shows regions
where the difference is
statistically significant at 5%
level
912 R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations
123
The difference between MAD for the A1B and 20CM
scenarios was computed for the total monsoon precipita-
tion. There is no statistically significant difference (at 5%
level) observed for MAD, suggesting that the dispersion
around the median is similar in both scenarios (not shown).
This result does not exclude differences at the very end of
Fig. 13 Same as in Fig. 10, but
for median of total precipitation
during the rainy season. Contour
interval is equal 50 mm. Solid
lines indicate positive values
and dashed lines indicate
negative values. Shading shows
regions where the difference is
statistically significant at 5%
level
R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations 913
123
the tails of the distributions that characterize extreme
events, an issue not investigated in the present study.
9 Conclusions
This study examined simulations of SAMS characteristics
by ten coupled IPCC global climate models with distinct
physics and resolutions. The skill in the simulations was
evaluated with GPCP precipitation during 28 rainy seasons
(1979–2006). The focus of this analysis was on the correct
simulations of the spatial patterns and variability of daily
and total monsoonal precipitation, onset and demise dates,
and duration of the monsoon. In this analysis every model
was examined individually, in order to identify those that
have poor skill against those with good skill in representing
SAMS characteristics and how they can potentially influ-
ence the ensemble in the present climate and future
scenarios of global change.
The SACZ is one of the most important features of
SAMS. With the exception of ECHAM5, all models tend to
represent an elongated band of precipitation emanating
from the Amazon with an orientation NW–SE similar to
the SACZ. MIROC3.2hires is the model that simulates
more realistically precipitation daily average and standard
deviation during the peak of the summer (DJF) in associ-
ation with the SACZ. FGOALS-g1.0 and ECHAM5, on the
other hand, show the poorest representation of the SACZ
daily precipitation characteristics.
The mean annual cycle of precipitation was examined in
some regions with distinct regimes. We show that over
north SA, the annual cycle is poorly represented by most
models. With the exception of MIROC3.2-hires, most
models tend to underestimate precipitation during the peak
of the rainy season. The misrepresentation of the ITCZ and
its seasonal cycle seems to be one of the main reasons for
the unrealistic out-of-phase annual cycles simulated near
the equator by many GCMs. As a consequence, simulations
of the total seasonal precipitation, onset and end of the rainy
season diverge among models and are notoriously unrea-
listic over north and northwest Amazon for most models.
On the other hand, the good perspective in using IPCC
models to understand and predict future climate changes in
SAMS is that the large majority of the IPCC models rea-
listically simulate the median characteristics and dispersion,
and phase of the precipitation annual cycle over central SA
for the 20CM runs. Previous studies have shown that this
region is the core of the monsoon, where precipitation, low
and high level winds, and humidity show the largest seasonal
amplitude. In this region, the median precipitation during the
rainy season is between 1,000 and 1,400 mm. The median
onset is observed between the beginning and the end of
October and the demise occurs between end of March and
mid-April, with duration of the rainy season between 32 and
36 pentads. The best performance in simulating onset,
end, duration and total monsoonal precipitation and its
interannual variability is observed with the following mod-
els: MIROC3.2-hires, MIROC3.2-medres and MRI-
CGCM2.3.2. The worst performance in simulating the same
characteristics for central SA is observed with ECHAM5,
GFDL-CM2.0 and GFDL-CM2.1. CSIRO-Mk3.0 model
shows the best simulation of the evolution of the onset and
end of the rainy season over the Amazon.
For the A1B scenario, the most coherent feature shown
in six out of ten models is the decrease of total monsoonal
precipitation over central and eastern SA, which coincides
with a region where models show high skill in simulating
SAMS characteristics in the 20CM runs. MIROC3.2-hires,
which shows the best performance in simulating the char-
acteristics of the total monsoonal precipitation and daily
precipitation in the peak of the rainy season, indicates a
deficit in precipitation between -100 and -200 mm in the
A1B scenario comparatively to the 20CM, extending
approximately from southern Amazon toward eastern
Brazil. Today, this region covers an area of intensive land-
use change, mostly due to expansion of agriculture and
pasture activities. Moreover, the northwestern boundary of
this region makes a frontier with the Amazon forest, a
region where large rate of deforestation, soil degradation
and human conflicts have been observed in the last decade
(IPCC 2007). Regional analyses are therefore necessary to
understand further impact of the projected decrease in
precipitation in this region.
Acknowledgments We thank Dr. Charles Jones and Dr. Humberto
R. Rocha and the two anonymous reviewers for their valuable com-
ments and suggestions for this manuscript. We also thank the Program
for Climate Model Diagnosis and Intercomparison (PCMDI) and the
WCRP’s Working Group on Coupled Modeling (WGCM) for making
available the WCRP CMIP3 multi-model dataset. GPCP data were
provided by NOAA. The authors greatly acknowledge the financial
support of the following agencies: FAPESP (Proc: 02/09289-9);
R. J. Bombardi FAPESP (06/53769-6); L. M. V. Carvalho CNPq
(Proc: 482447/2007-9 and 474033/2004-0) and NOAA Office of
Global Programs (NOAA NA07OAR4310211).
References
Ashrit RG, Douville H, Kumar KR (2003) Response of the Indian
monsoon and ENSO-monsoon teleconnection to enhanced
greenhouse effect in the CNRM coupled model. J Meteorol
Soc Jpn 81:779–803
Carril AF, Mene
´
ndez CG, Nun
˜
ez MN (1997) Climate change
scenarios over the South American region: an intercomparison
of coupled general atmosphere-ocean circulation models. Int J
Climatol 17:1613–1633
Cariolle D, De
´
que
´
M (1986) Southern hemisphere medium-scale
waves and total ozone disturbances in a spectral general
circulation model. J Geophys Res 91:10825–10846
914 R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations
123
Cariolle D, Lasserre-Bigory A, Royer J-F et al (1990) A general
circulation model simulation of the springtime Antarctic ozone
decrease and its impact on mid-latitudes. J Geophys Res
95:1883–1898
Carvalho LMV, Jones C, Liebmann B (2002a) Extreme precipitation
events in southeastern South America and large-scale convective
patterns in the South Atlantic convergence zone. J Clim
15:2377–2394
Carvalho LMV, Jones C, Silva Dias MAF (2002b) Intraseasonal
large-scale circulations and mesoscale convective activity in
tropical South America during the TRMM-LBA campaign.
J Geophys Res. doi:10.102/2001JD000745
Carvalho LMV, Jones C, Liebmann B (2004) The South Atlantic
convergence zone: persistence, intensity, form, extreme precip-
itation and relationships with intraseasonal activity. J Clim
17:88–108
Coelho CAS, Uvo CB, Ambrizzi T (2002) Exploring the impacts of
the tropical Pacific SST on the precipitation patterns over South
America during ENSO periods. Theor Appl Climatol 71:185–
197
Dai A (2006) Precipitation characteristics in eighteen coupled climate
models. J Clim 19:4605–4630
Delworth TL, Broccoli AJ, Rosati A et al (2006) GFDLs CM2 global
coupled climate models—Part 1: formulation and simulation
characteristics. J Clim 19:643–674
Flato GM, Boer GJ, Lee WG et al (2000) The Canadian centre for
climate modelling and analysis global coupled model and its
climate. Clim Dyn 16:451–467
Gan MA, Kousky VE, Ropelewski CF (2004) The South America
monsoon circulation and its relationship to rainfall over west-
central Brazil. J Clim 17:47–66
Giorgi F, Francisco R (2000) Evaluating uncertainties in the
prediction of regional climate change. Geophys Res Lett
27:1295–1298
Gordon HB, Rotstayn LD, Mcgregor JL et al (2002) The CSIRO Mk3
climate system model. CSIRO Atmospheric Research Tech.
Available via DIALOG. http://www.dar.csiro.au/publications/
gordon_2002a.pdf. Accessed 20 Sep 2006
Grimm AM, Barros VR, Doyle ME (1998) Precipitation anomalies in
southern South America associated with El Nin
˜
o and La Nin
˜
a
events. J Clim 11:2863–2880
Grimm AM, Barros VR, Doyle ME (2000) Climate variability in
southern South America associated with El Nin
˜
o and La Nin
˜
a
events. J Clim 13:35–58
Hasumi H, Emori S (2004) K-1 coupled GCM (MIROC). Available
via DIALOG. http://www.ccsr.u-tokyo.ac.jp/kyosei/hasumi/
MIROC/tech-repo.pdf. Accessed 20 Sep 2006
IPCC (2001): Climate Change (2001) Impacts, adaptation, and
vulnerability. In: McCarthy JJ, Canziani OF, Leary NA, Dokken
DJ, White KS (eds) Contribution of working group II to the third
assessment report of the intergovernmental panel on climate
change. Cambridge University Press, Cambridge
IPCC (2007): Climate Change (2007) The physical science basis. In:
Solomon S, Qin D, Manning M, Chen Z, Marquis M, Averyt KB,
Tignor M, Miller HL (eds) Contribution of working group I to
the fourth assessment. Report of the intergovernmental panel on
climate change. Cambridge University Press, Cambridge
Jones C, Carvalho LMV (2002) Active and break phases in the South
American monsoon system. J Clim 15:905–914
Kayano MT, Kousky VE (1992) On the monitoring of intraseasonal
oscillations. Rev Bras Meteorol 7:593–602 (in Portuguese)
Kharin VV, Zwiers FW, Zhang X et al (2007) Changes in temperature
and precipitation extremes in the IPCC ensemble of global
coupled model simulations. J Clim 20:1419–1444
Kodama Y-M (1992) Large-scale common features of subtropical
precipitation zones (the Baiu Frontal Zone, the SPCZ, and the
SACZ). Part I: characteristics of subtropical frontal zones.
J Meteorol Soc Jpn 70:813–835
Kousky VE (1988) Pentad outgoing longwave radiation climatology
for the South American sector. Rev Bras Meteorol 3:217–231
Labraga LC, Lopez M (1997) A comparison of the climate response
to increased carbon dioxide simulated by general circulation
models with mixed-layer and dynamic ocean representations in
the region of South America. Int J Climatol 17:1635–1650
Lambert SJ, Boer GJ (2001) CMIP1 evaluation and intercomparison
of coupled climate models. Clim Dyn 17:83–106
Lenters JD, Cook KH (1999) Summertime precipitation variability
over South America: role of the large-scale circulation. Mon
Wea Rev 127:409–431
Liebmann B, Marengo JA (2001) Interannual variability of the rainy
season and rainfall in the Brazilian Amazon Basin. J Clim
14:4308–4318
Liebmann B, Jones C, Carvalho LMV (2001) Interannual variability
of extreme precipitation events in the State of Sa
˜
o Paulo, Brazil.
J Clim 14:208–218
Liebmann B, Allured D (2005) Daily precipitation grids for South
America. Bull Am Meteorol Soc 86:1567–1570
Liebmann B, Camargo SJ, Seth A, Marengo JA et al (2007) Onset and
end of the rainy season in South America in observations and the
ECHAM 4.5 atmospheric general circulation model. J Clim
20:2037–2050
Lincoln MA, Marengo JA, Camargo H et al (2005) The onset of the
rainy season over southeast Brazil, Part-1. Observational studies.
Rev Bras Metor 20:385–394 (in Portuguese)
Magan
˜
a V, Ambrizzi T (2005) Dynamics of subtropical vertical
motions over the Americas during El Nin
˜
o Boreal winters.
Atmo
´
sfera 18:211–235
Marengo JA (2003) Ensemble simulation of regional rainfall features
in the CPTEC/COLA atmospheric GCM. Skill and predictability
assessment and applications to climate predictions. Clim Dyn
21:459–475
Marengo JA, Liebmann B, Kousky VE et al (2001) Onset and end of
the rainy season in the Brazilian Amazon basin. J Clim 14:833–
852
Meehl GA, Arblaster JM, Tebaldi C (2005) Understanding future
patterns of increased precipitation intensity in climate model
simulations. Geophys Res Lett. doi:10.1029/2005GL023680
Muza MN, Carvalho LMV (2006) Intraseasonal and interannual
variability of extreme precipitation and drought over southern
Amazon during the Austral summer. Rev Bras Meteorol 21:29–
41 (in Portuguese)
Roeckner E, Ba
¨
uml G, Bonaventura L et al (2003) The atmospheric
general circulation model ECHAM5. Part 1. Available via
DIALOG. http://www.mpimet.mpg.de/fileadmin/publikationen/
Reports/max_scirep_349.pdf. Accessed 8 Jan 2008
Salas-Me
´
lia D, Chauvin F, De
´
que
´
M et al (2005) Description and
validation of the CNRM-CM3 global coupled model. Available
via DIALOG. http://www.cnrm.meteo.fr/scenario2004/paper_
cm3.pdf. Accessed 8 Jan 2008
Seth A, Rauscher SA, Camargo SJ (2007) RegCM3 regional
climatologies for South America using reanalysis and ECHAM
global model driving fields. Clim Dyn 28:461–480
Silva AE, Carvalho LMV (2007) Large-scale index for South America
monsoon (LISAM). Atmos Sci Lett. doi:10.1002/asl.150
Silva MES, Carvalho LMV, Dias MAFS et al (2006) Complexity and
predictability of daily precipitation in a semi-arid region: an
application to Ceara
´
, Brazil. Nonlinear Process Geophys 13:651–
659
Sun Y, Solomon S, Dai A (2005) How often does it rain? J Clim
19:916–934
Tebaldi C, Hayhoe K, Arblaster J et al (2007) Going to the extremes:
an intercomparison of model-simulated historical and future
R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations 915
123
changes in extreme events. Clim Change 79:185–211. doi:
10.1007/s10584-006-9051-4
Vera C, Higgins W, Ambrizzi T et al (2006a) Toward a unified view
of the American monsoon systems. J Clim 19:4977–5000
Vera C, Silvestri G, Liebmann B (2006b) Climate change scenarios
for seasonal precipitation in South America from IPCC-AR4
models. Geophys Res Lett. doi:10.1029/2006GL025759
Wilks DS (2006) Statistical methods in the atmospheric sciences.
Academic Press, New York
Xie P, Janowiak JE, Arkin PA et al (2003) GPCP pentad precipitation
analyses: an experimental dataset based on gauge observations
and satellite estimates. J Clim 16:2197–2214
Yu Y, Zhang X, Guo Y (2004) Global coupled ocean atmosphere
general circulation models in LASG/IAP. Adv Atmos Sci
21:444–455
Yukimoto S, Noda A, Kitoh A et al (2006) Present-day climate and
climate sensitivity in the Meteorological Research Institute
Coupled GCM, version 2.3 (MRI-CGCM2.3). J Meteorol Soc
Jpn 84:333–363
Zhou J, Lau KM (1998) Does a monsoon climate exist over South
America? J Clim 11:1020–1040
916 R. J. Bombardi, L. M. V. Carvalho: IPCC global coupled model simulations
123
