834
JOURNAL OF CLIMATE
VOLUME 16
The Maritime Continent and Its Role in the Global Climate: A GCM Study
RICHARD NEALE*
AND
JULIA SLINGO
Centre for Global Atmospheric Modelling, Department of Meteorology, University of Reading, Reading, United Kingdom
(Manuscript received 16 July 2001, in final form 5 September 2002)
ABSTRACT
The Maritime Continent, with its complex system of islands and shallow seas, presents a major challenge to
models, which tend to systematically underestimate the precipitation in this region. Experiments with a climate
version of the Met Office model (HadAM3) show that even with a threefold increase in horizontal resolution
there is no improvement in the dry bias. It is argued that the diurnal cycle over the islands and the complex
circulation patterns generated by land–sea contrasts are crucial for the energy and hydrological cycles of the
Maritime Continent and for determining the mean climate. It is shown that the model has substantial errors in
its simulation of the diurnal cycle over the islands, which can rectify onto the seasonal mean climate.
It is further argued that deficient rainfall over the Maritime Continent could be a driver for other systematic
errors, such as the excess precipitation over the western Indian Ocean. To demonstrate the sensitivity of global
systematic model errors to the heating in this region, two experiments have been performed, one with the existing
distribution of islands and a second where the island grid points are replaced by ocean grid points. In the absence
of the islands of the Maritime Continent, the local precipitation increases by 15%, reducing the existing dry
bias and bringing the model closer to observations. In response to this improved heating distribution, precipitation
decreases over the west Indian Ocean and South Pacific convergence zone, reducing the systematic wet bias in
these regions. This supports the hypothesis that tropical systematic errors are often related through vertical
(Walker) circulations.
The extratropical response to changes in the Maritime Continent heat source is also well demonstrated by
these experiments. The enhanced heating and, hence, divergent outflow generates Rossby waves, which have a
significant impact on the winter circulation and surface temperatures across much of North America and the
northeast Eurasian region. These changes are such as to substantially reduce model systematic error in these
regions. These results reinforce the critical role played by the Maritime Continent in the global circulation. It
emphasizes the need for better representation of convective organization over regions of complex land–sea
terrains and the importance of considering the global context of model systematic errors in which biases in the
Tropics may be a key factor.
1. Introduction
The tropical Maritime Continent has a unique environment where convective activity responds to forcing
on many timescales and space scales, the net result of
which is able to influence climate on the global scale.
This region, so named because of the complex distribution of several large islands with elevated orography
occupying the domain 108S–208N and 908–1508E, also
encompasses some of the warmest ocean temperatures
of the world and is known as the ‘‘boiler box’’ of the
Tropics (Ramage 1968). The Maritime Continent receives much of its rainfall from convective activity as-
* Current affiliation: NOAA–CIRES Climate Diagnostics Center,
Boulder, Colorado.
Corresponding author address: Dr. Richard Neale, NOAA–CIRES
Climate Diagnostics Center, R/CDC1, 325 Broadway, Boulder, CO
80305-3328.
E-mail: [email protected]
q 2003 American Meteorological Society
sociated with localized thunderstorms. The Island Thunderstorm Experiment (ITEX; Keenan et al. 1989) and
the Maritime Continent Thunderstorm Experiment
(MCTEX; Keenan et al. 2000) were established in order
to diagnose and model the importance of the thunderstorm-scale convection for the climate of the region.
The islands play an important part in the meteorology
of the Maritime Continent. Larger-scale organization of
thunderstorm activity is strongly influenced by the orography of the region as well as the sea-breeze circulations, which show strong diurnal variations. Satellite
images show this diurnal variation over the islands to
be a striking feature of this region (Holland and Keenan
1980). The nature of these sea-breeze circulations has
been studied extensively. The convection related to seabreeze convergence is able to aggregate into mesoscale
convective complexes (MCCs) later in the day, which
move offland to give the greatest precipitation during
the local morning time over the oceans (Williams and
Houze 1987).
The timing and magnitude of this convective activity
1 MARCH 2003
NEALE AND SLINGO
on diurnal timescales is known to vary significantly between land and ocean regions. Using a climatology of
window brightness derived from high temporally sampled satellite data, Yang and Slingo (2001) show that
the surface inhomogeneity of the Maritime Continent
makes for complex diurnal forcing of convection. During Northern Hemisphere winter, in particular, the diurnal amplitude of convective rainfall over the islands
can be three times as great as that over the adjacent
ocean. This is in response to the smaller thermal heat
capacity of the land surface leading to large diurnal
variations in low-level instability. However, there are
further complications to the response, since the strong
variability over the islands is able to propagate out over
the oceans as gravity waves leading to coherent variations in the phase of the convective peak. Much of this
variability may be related to the standard lifetime of
convective systems that start over land, build, then move
away over the oceans (Saito et al. 2001). Such an evolution suggests that the atmosphere has some memory
of a particular convective disturbance at some previous
time.
The Maritime Continent also exhibits an influence on
larger-scale intraseasonal activity. The Madden–Julian
oscillation (MJO) propagates over the region in its mature phase and is modulated by the presence of the underlying surface properties and the diurnal cycle before
moving out over the West Pacific. This makes for a
complex response whereby the diurnal cycle of convective activity becomes suppressed during the active
phase of the MJO and enhanced during the break phase
(Sui and Lau 1992). The presence of the Maritime Continent also modulates the strength and phase speed of
the MJO. In particular, it is able to both weaken and
split the active phase of the oscillation before it reintensifies in the South Pacific convergence zone (SPCZ;
Zhu and Wang 1993).
The southern part of the Maritime Continent region
is strongly influenced by the winter monsoon circulation
involving significant transport of warm moist air from
north of the equator that meets the trade wind flow south
of the equator to generate the austral monsoon trough.
The first strong MJO of the winter season usually signifies the onset of the monsoon regime (Sui and Lau
1992). The winter monsoon associated with convection
over the Maritime Continent has also been shown to
influence the midlatitude circulation through short-term
teleconnections (Lau et al. 1983). For example, during
an active phase of the monsoon the tropical and extratropical circulations vary in a coherent way, intensifying
the local Hadley and Walker circulations and strengthening the east Asia subtropical jet (Chang and Lau
1982). The action of cold surges, which strengthen the
Asian winter monsoon northeasterlies, also results in
periods of enhanced convective activity throughout the
season, particularly north of Borneo (Houze et al. 1981).
These cold surges can typically last from 5 to 14 days
(Zhang et al. 1997) and so may account for a large
835
proportion of climate variability over the Maritime Continent during winter.
Variability on seasonal timescales modulates the rainfall over the Maritime Continent particularly during the
warm phase of ENSO (Philander 1985). As the warmest
SSTs move out into the central Pacific, the strongest
convection follows and generates an anomalous longitudinal circulation, leading to suppression over the Maritime Continent. This is a particularly large effect since
ENSO events typically reach a maximum in Northern
Hemisphere winter when the precipitation totals usually
reach a maximum over the Maritime Continent.
On the planetary scale, convection over the Maritime
Continent represents a dominant heat source for the atmospheric circulation. Convection is most intense and
the tropopause at its highest over this region. Uppertropospheric divergent outflow from this convective region has been shown, in an idealized model, to be a
major source of wave activity due to the generation of
global rotational flow (Sardeshmukh and Hoskins 1988).
Such a response establishes the stationary wave patterns
clearly observed in time-mean fields. So it is clear that
the Maritime Continent has an important role to play
both in the variability of the tropical climate and for the
global circulation as a whole.
The aim of this paper is to gain an insight into the
climate of the Maritime Continent region and to highlight its role in the tropical and global mean climate.
The following section assesses the skill of the Met Office Unified Model (UM) at reproducing the mean climate over the Maritime Continent region and identifies
some possible shortcomings of the model. Specific errors relating to the presence of the Maritime Continent
islands are identified and results from idealized sensitivity experiments, which attempt to remove these errors, are described. Further shortcomings of the UM’s
representation of the diurnal cycle are then highlighted.
Investigation of a possible sensitivity to the details of
the UM’s radiation scheme is performed with a further
short 1-yr integration of the model. Significant largescale influences of changes in the Maritime Continent
heat source region are revealed in a section on the global
impacts; the implications of this study and conclusions
are summarized in the final section.
2. Description of the model and its systematic
errors
A suite of experiments using the Met Office Unified
Model (UM), performed as part of the second phase of
the Atmospheric Model Intercomparison Project (AMIP
II; Gates 1992), has been analyzed and the performance
of the UM assessed over our region of interest, the Maritime Continent. The experimental period extends over
17 years (1979–95) and the model is forced by monthly
mean SSTs and sea ice. The climate version of the UM
(HadAM3; Pope et al. 2000) is used at a resolution of
2.58 in latitude, 3.758 in longitude, and 19 levels in the
836
JOURNAL OF CLIMATE
VOLUME 16
FIG. 1. Seasonal mean errors from a six-member standard AMIP II ensemble of the UM when compared with the Xie–Arkin climatology
(precipitation) and ERA (winds): (a) DJF precipitation (mm day 21 ); (b) as in (a) but for JJA; (c) DJF 850-hPa wind vectors (m s 21 ); (d) as
in (c) but for JJA. Precipitation values less the 22 mm day 21 are shaded. The zero contour is omitted.
vertical. This model version has significant changes
compared to earlier versions, namely, the inclusion of
momentum transports by subgrid-scale convective processes (Kershaw and Gregory 1997), an updated radiation scheme (Edwards and Slingo 1996), and a new
land surface and vegetation scheme [Met Office Surface
Exchange Scheme (MOSES); Cox et al. 1999]. For comparison, observational data are provided by the European Centre for Medium-Range Weather Forecasts
(ECMWF) Re-Analysis (ERA; Gibson et al. 1997) and
the Climate Prediction Center Merged Analysis of Precipitation (CMAP) rainfall climatology (Xie and Arkin
1996) covering the same analysis period as the AMIP
II experiments.
In the UM, as with many climate models, the Maritime Continent exhibits a deficit of precipitation in all
seasons (see the AMIP II Web site online at http://wwwpcmdi.llnl.gov/amip). Figure 1 shows the precipitation
and lower-tropospheric wind errors in the model during
Northern Hemisphere winter [December–January–February (DJF)] and summer [June–July–August (JJA)].
During DJF precipitation deficits in excess of 2 mm
day 21 cover most of the Maritime Continent north of
the equator and the western part south of the equator.
Only to the east and west of Guinea and just north of
the Australian coast is there a significant excess of precipitation in the model.
In JJA the errors in precipitation appear less coherent,
but in general the same pattern as in DJF occurs with a
dry bias centered mostly off the coasts of the large islands. The pattern of errors over the Asian monsoon re-
gion show that the rainfall is overestimated across northern and central India as well as into the far southeast of
Asia. In addition, the equatorial maximum over the Indian
Ocean, which characterizes the break phase of the monsoon, is positioned too far to the west. Due to the presence
of large-scale tropical vertical circulations it is possible
that shortcomings over the Maritime Continent may have
some role to play in the model’s simulation of the monsoon during JJA also. Certain very local errors persist
throughout the whole year. For instance, the negative
errors over the Philippines and off the southwest coast
of Sumatra suggest that some of the problems may lie
in the model’s response to the presence of these islands.
Associated with the errors in the seasonal mean precipitation are errors in the lower-tropospheric flow. In
DJF (Fig. 1c), the zonal flow has a strong easterly bias
from the West Pacific through to the Indian Ocean north
of the equator. It is conceivable that the large-scale nature of this error is related to the lack of a strong organized MJO, which would have periods of significant
westerly wind activity projecting onto the mean basic
state. Indeed the AMIP I study by Slingo et al. (1996)
showed that a poor model MJO and errors in the Maritime Continent basic state could be closely linked. Embedded within the large-scale flow bias are errors on the
scale of the Maritime Continent islands. In general, the
easterly bias tends to be increased on the west side of
the islands and points toward problems in simulating
the local island-scale circulation associated with seabreeze effects and the diurnal cycle. The problems with
simulating the coupling between the diurnal cycle and
1 MARCH 2003
NEALE AND SLINGO
837
FIG. 2. Annual mean precipitation error (mm day 21 ) from four UM AMIP II experiments with different horizontal resolution: (a) climate
resolution (2.58 3 3.758); (b) 1.5 3 climate resolution (1.678 3 2.58); (c) 2 3 climate resolution (1.258 3 1.8758); (d) 3 3 climate resolution
(0.838 3 1.258). Values less than 22 mm day 21 are shaded. The zero contour is omitted.
sea-breeze circulation and its implications for the mean
climate of the model are addressed in a later section.
Much of the important geographic detail of the Maritime Continent is not resolved at the coarse climate
resolution of the model, and at first sight, it could be
concluded that the errors described above might be corrected with higher horizontal resolution. A number of
AMIP II experiments have been performed to address
the sensitivity of model errors to horizontal resolution
(Stratton 1999). Many improvements are seen in these
experiments, such as a stronger midlatitude circulation
due to better resolved storm tracks. However, even with
a threefold increase in horizontal resolution there is no
systematic improvement of the dry bias in the model
(Fig. 2). The pattern of the errors in the tropical precipitation persists and, if anything, is enhanced with
increasing resolution.
The results in Fig. 2 demonstrate that computationally
achievable increases in resolution do not necessarily
lead to improvements in model error, and indicate that
it is probably deficiencies in the representation of the
physical system that are the primary source of these
errors. The fact that many models show similar problems over the Maritime Continent suggests that they all
lack some key ingredient. In section 4 it is hypothesized
that the diurnal cycle and the generation of land–sea
breezes around this complex system of islands may be
possible factors.
The precipitation errors shown in Fig. 1 represent a
substantial fraction of the mean precipitation, which severely compromises the tropical heating pattern in the
model (not shown). The global scale of this problem is
demonstrated in section 5. In the following sections, a
sensitivity experiment is described, which aims to improve the heating distribution over the Maritime Continent and thereby show the importance of an accurate
simulation of the climate of the Maritime Continent for
the global circulation.
3. Role of the islands in the climate of the
Maritime Continent
As shown in the AMIP II experiments, the UM is
unable to simulate well the mean climate over the Maritime Continent in terms of both the precipitation distribution and the circulation. It is clear, therefore, that
the UM has great trouble in representing the observed
impact of the islands of the region on the resolvable
portion of the flow. Closer investigation over the individual islands reveals errors in their surface properties
that may be having a detrimental impact on the modeled
convection, possibly leading to the mean negative bias.
The correct mean surface air temperature is important
to the boundary layer stability, which is the closure
method for the convection scheme in the UM (Gregory
and Rowntree 1990). Figure 3a shows the mean surface
air temperature error, when compared with the climatology of Legates and Willmott (1990), from the standard AMIP II ensemble experiments at climate resolution. Over the island grid points a cold bias predominates (see Fig. 4a for exact location of land grid points).
This is in excess of 28C over Borneo, the Philippines,
838
JOURNAL OF CLIMATE
VOLUME 16
FIG. 3. Annual mean surface air temperature error (8C) for the UM AMIP II experiments over the Maritime Continent region at
(a) climate resolution and (b) 3 3 climate resolution.
and New Guinea. The problem is not simply a resolution
issue, since even at 3 times climate resolution there is
still a mean cold bias in surface air temperature over
the islands (Fig. 3b).
A further hindrance to convection may be the presence of layer cloud in the lowest layer of the model,
essentially a fog layer. Mean values can reach in excess
of 30% over the islands (not shown). This also masks
the fact that there exists a strong diurnal signal where
cloud amounts in the lowest model layer can build to
almost 100% cover during the night. Therefore after
sunrise, insolation has to burn off this excessive low
cloud before it can begin to heat the surface in order to
generate low-level instability. This could conceivably
be a hindrance to the correct evolution of convection,
the consequences of which are discussed in the next
section.
In an attempt to investigate the influence of the cold
bias over the islands, and to artificially improve the
climate of the region, two further AMIP II–type experiments have been carried out. One of the aims of these
experiments is to identify how much of the error in the
global Tropics seen in Fig. 1 can be attributed to a
remote response to errors over the Maritime Continent.
The first experiment is equivalent to the standard AMIP
II integration for the full 17 years of the AMIP II period
but with additional diagnostics. This will be referred to
as the ‘‘control’’ experiment. The second is an identical
FIG. 4. Elevation of land grid points (m) and annual mean SST (8C) used in (a) control and (b) no islands experiments.
1 MARCH 2003
NEALE AND SLINGO
839
FIG. 5. Mean differences between the no islands and control experiments: (a) DJF precipitation (mm day 21 ); (b) as in (a) but for JJA; (c)
DJF 850-hPa wind vectors (m s 21 ); (d) as in (c) but for JJA. Precipitation changes less than 21 mm day 21 are shaded and the zero contour
is omitted.
experiment except that the land grid points composing
the Maritime Continent islands have been removed and
replaced by ocean grid points with SST bilinearly interpolated from the surrounding existing ocean grid
points (see Fig. 4). This will be referred to as the ‘‘no
islands’’ experiment. In fact, two realizations of the no
islands experiment were performed to ensure that the
results were significant.
The rationale behind the removal of the land grid
points is to determine how detrimental the incorrect representation of the Maritime Continent islands may be
to the generation of strong convection throughout the
year. By replacing these land grid points with warmer
SSTs, the cold bias noted in Fig. 3 is eliminated, which
with the enhanced moisture availability from the sea
surface should lead to greater convective activity. Therefore, it is hoped that the mean bias may be corrected
by providing boundary forcing, which is more conducive to strong convection than in the control experiment.
While this experiment is a useful excerise in determining
the impacts of improving model error, it would not represent a viable solution given that any improvements
would essentially be for the wrong reason.
The mean response to the removal of the islands in
the Maritime Continent is a net increase in precipitation
over and around the region. Figure 5 shows the changes
in precipitation and lower-tropospheric flow associated
with the removal of the Maritime Continent islands. This
figure should be compared with Fig. 1, showing the
errors from observations of the original AMIP II experiments. In DJF there is an increase in precipitation
off the coast of the western islands of the Maritime
Continent and a decrease in the SPCZ region and north
of the Australian continent, reducing the strength of the
winter monsoon. These changes largely correct for the
complex pattern of precipitation error seen in the standard version of the UM. Nonlocal changes in precipitation are also evident in the western Indian Ocean,
which again partially correct for errors there.
In JJA, changes in precipitation are more dramatic
and extend to a greater part of the adjacent regions north
and east of the equator. Over the Maritime Continent
region there is a coherent increase in precipitation correcting for the existing bias. In response to these increases there is a general reduction in precipitation on
the periphery of the Maritime Continent. Over the equatorial Indian Ocean this leads to an improved, more
zonally oriented precipitation distribution. To the north
and northeast of the region, the changes lead to a more
realistic strength for the Asian summer monsoon and
the removal of the spurious extension of the precipitation pattern into the western Pacific, east of the Philippines. As in DJF the SPCZ wet bias is reduced.
Although Fig. 5 shows a consistent improvement in
the precipitation field, the lower-tropospheric wind
changes tend to reinforce existing errors in DJF, with
enhanced easterlies in the region, particularly north of
the equator. In JJA the wind changes also add to the
existing errors in the west Pacific, but in the Indian
Ocean and over southern Asia the changes are somewhat
more favorable, consistent with the precipitation changes. The low-level flow into the Indian and Asian mon-
840
JOURNAL OF CLIMATE
FIG. 6. Mean monthly variation of precipitation (mm day 21 ) averaged over the region 208S–108N and 908–1508E for the 17 years
of the standard AMIP II integrations for the control and the no islands
experiments.
soon region is reduced and the flow is diverted more
toward the Maritime Continent after crossing the equator in the western Indian Ocean. Such a change in the
flow is consistent with a stationary Rossby wave response to the enhanced heating over the Maritime Continent. However, considerable errors remain, particularly
in the low-level flow over the Indian subcontinent. This
suggests that these errors may be due to more local
problems associated with the interaction of the model
physics in the monsoon region (Martin 1999).
The changes seen in the no islands experiment are
significant when compared to both the standard AMIP
II ensemble and the control integration. Figure 6 shows
this clearly, with the mean annual variation in precipitation averaged over the Maritime Continent area lying
outside the ensemble spread of the standard AMIP II
integrations and being significantly different from the
control experiment, particularly during northern summer. More importantly the mean annual evolution is in
much closer agreement with observations than any of
the standard AMIP II or control experiments.
The overall increase in precipitation over the Maritime Continent region must be accompanied by an increase in the supply of moisture. Since the island grid
points are being replaced by warmer, saturated ocean
grid points, one possibility is that the low-level buoyancy is increased, due to temperature and humidity effects, with the moisture supply for the increased convection being provided locally by the enhanced evaporation from the sea surface. However, this is not entirely the case. Table 1 summarizes the surface energy
VOLUME 16
and moisture budgets for the Maritime Continent region.
There is a net increase in precipitation in the no islands
experiment, which brings the total into closer agreement
with estimates from the Xie–Arkin climatology. However, local evaporation within the Maritime Continent
region accounts for only a quarter of the precipitation
change between the two experiments. Therefore, the remaining 75% must be provided by nonlocal moisture
convergence. This also raises the issue of whether the
increase in surface temperature, due to the reduction of
orographic elevation to sea level, or the increase in surface moisture availability, is important for the increase
in precipitation in the no islands experiment.
To address this issue a second much shorter sensitivity
experiment was performed where the land grid points
were retained but the orography was reduced to zero.
Briefly, the results show a local and remote response
that is quantitatively similar to the no islands experiment, but of much reduced magnitude. Therefore, we
can conclude that the absence of an interactive land
surface and its replacement by a fixed SST boundary
forcing provides a much greater contribution toward the
enhanced mean precipitation in the no islands experiment than the removal of the orographic forcing.
As well as a change in the hydrological cycle, Table
1 also shows that the surface shortwave radiation increases despite the enhanced convective activity. This
is because the diurnal cycle of cloud in the lowest model
layer is eliminated over the island grid points. Table 1
suggests that although the desired effects of increased
precipitation and some improved aspects of the climatology have been achieved, this has been at the expense
of a deterioration in the surface energy budget with the
caveat that the ERA fluxes are themselves not always
reliable.
The no islands experiment has demonstrated that inadequate treatment of the islands of the Maritime Continent in global climate models may be responsible for
the deficiencies in precipitation in this region and hence
to errors in the tropical heating distribution throughout
the warm pool region of the west Pacific and Indian
Oceans. Despite the unrealistic nature of the no islands
experiment, the improved rainfall climatology over the
Maritime Continent in this integration has proved key
to answering questions relating to the global effects of
errors in the atmospheric heat source of the Maritime
Continent, as discussed in section 5. It is clear that the
meteorology of the islands is a key factor in determining
TABLE 1. Surface energy components (W m 22) [latent heating (LH); sensible heating (SH); shortwave heating (SW); Longwave heating
(LW)] and surface moisture components of precipitation and evaporation (mm day 21 ) averaged over the Maritime Continent region (158S–
158N; 908E–1508E) during the AMIP II experiment period (1979–95) for the control and no islands experiments, and the ERA/Xie–Arkin
climatologies. Positive energy budget values indicate fluxes out of the surface.
Source
LH
SH
SW
LW
Precipitation
Evaporation
Control
No islands
ERA/Xie–Arkin
130.9
136.9
118.6
12.28
10.25
9.09
2226.4
2230.3
2188.8
56.3
56.3
48.3
5.83
6.64
6.43
4.53
4.72
4.10
1 MARCH 2003
NEALE AND SLINGO
841
FIG. 7. Local time of the maximum in the diurnal cycle of precipitation over the Maritime Continent: (a) mean of two winters derived
from the global window brightness temperature of the Cloud Archive User Service (CLAUS; see Yang and Slingo 2001); (b) a single winter
from the control experiment.
the heat and moisture budgets. In the next section a
particular aspect of that meteorology, the diurnal cycle,
will be examined.
4. Role of the diurnal cycle in the Maritime
Continent
On shorter timescales the characteristics of the diurnal
cycle, its phase and amplitude, are important for the
organization of precipitation in the Tropics. The study
of Yang and Slingo (2001) reveals that the UM has low
skill in reproducing the observed diurnal variation of
rainfall over the Maritime Continent, and indeed over
all tropical land areas. Throughout the convectively active Tropics, the model systematically maximizes precipitation too early during the day. Over land the peak
occurs predominantly near local noon, too soon after
the solar maximum, while over the ocean the peak is
around local midnight.
More specifically, over the Maritime Continent region
the model demonstrates particularly poor performance
(cf. Figs. 7a,b). In the satellite observations, convection
is seen to maximize in the late evening over the islands,
whereas the model shows a maximum near local noon.
This implies that the life cycle of convection in the
model, from initial buoyancy excess of air parcels,
through shallow cumulus and cumulus congestus, to organized thunderstorms, is much too short. Strong convection develops too soon during the day, which then
cuts off the solar radiation to the surface, leading to
time-average surface temperatures that are consistently
lower than observations over all the Maritime Continent
islands (Fig. 3). Such shortcomings of the model may
not be surprising, since the observed diurnal cycle over
the Maritime Continent islands is known to involve sub-
tle interactions with small horizontal scale (tens of kilometers) sea-breeze circulations (Keenan et al. 2000),
as well as a smooth transition through a multistage convective life cycle (Saito et al. 2001).
As well as the problems noted above, other shortcomings in the simulated diurnal cycle have been highlighted, most acutely over land and just off the coasts
where the diurnal cycle accounts for a large proportion
of the precipitation variability. As an example, Fig. 8
demonstrates the contrasting characteristics in the diurnal cycle averaged separately over land and ocean
areas within the Maritime Continent. Here the diurnal
cycle is calculated using the first three diurnal harmonics
derived from instantaneous data at three hourly intervals.
The convective heating in Figs. 8a,b shows that, averaged over the land grid points, there is a strong maximum just before noon in the midtroposphere. What is
also evident is the dominance of deep convection in the
model, with no apparent buildup of convection through
the morning. Over ocean grid points, the weaker amplitude shows that the diurnal cycle is a less important
component of precipitation variability. The convective
moistening in Figs. 8c,d reveals that the action of the
convection is predominantly to dry the atmosphere, implying that the convection rapidly develops to precipitating convection in the model without the moistening,
preconditioning phase associated with cumulus congestus. Idealized experiments with an aquaplanet version
of the UM have confirmed that the drying of the atmosphere by precipitating convection is an overdominant process in the model (Inness et al. 2001), in that
case with implications for the intraseasonal organization
of convection associated with the MJO.
842
JOURNAL OF CLIMATE
VOLUME 16
FIG. 8. Mean diurnal cycle derived from one year of a 3-hourly sampled integration of the control experiment: (a) convective heating (8C day 21 ) over the land grid points of the Maritime Continent; (b) convective
heating (8C day 21 ) over the ocean grid points adjacent to the Maritime Continent; (c) as in (a) [and (d) as
in (b)] but for convective moistening (g kg 21 day 21 ); (e) as in (a) [and (f ) as in (b)] but for total cloud
cover fraction.
The mean diurnal evolution of the total cloud fraction
is shown in Figs. 8e,f. First, the diurnal cycle of cloud
fraction in the lowest model layer is very marked over
land, ranging from 0% at the time of the convective
maximum to an islands-wide maximum of 70% a couple
of hours prior to sunrise. This undoubtedly has an impact on the evolution of convection since clouds will
delay the time at which insolation can start to heat the
land surface. As already noted, it also affects the overall
energy budget of the island grid points.
It is clear from the above results that the model has
serious difficulties in simulating the diurnal cycle over
the islands of the Maritime Continent and it is important
to investigate whether this failing has implications for
the simulated mean climate of the region. In common
with many climate models (e.g., ECMWF; Morcrette
1 MARCH 2003
NEALE AND SLINGO
2002), the UM does not perform a full radiation calculation at every time step. For reasons of computational
cost, the standard practice is to perform a full radiation
calculation every 3 h or six model time steps. At intermediate time steps, for the purpose of the radiation
calculation, the cloud amounts and surface and atmospheric temperatures remain unchanged, although the
solar fluxes and heating rates are scaled by the correct
zenith angle. This means, first, that the longwave fluxes
cannot respond to changes in solar forcing, and second,
that there is a time lag in the cloud forcing, particularly
for the surface fluxes.
It is possible that this approximation may be important for the evolution of the diurnal cycle. For example,
a strong surface energy imbalance may occur at the start
and end of the solar day, when the insolation is changing
rapidly but the atmospheric and surface properties are
only updated every 3 h. The combination of the shortwave and longwave approximations could contribute to
a too-rapid increase in surface temperature and the generation of strong lower-tropospheric buoyancy excess
too early in the day. This could lead to a maximum in
convection over the tropical islands also too early in the
day, which would in turn cut off the solar radiation and
limit the net energy input to the land surface (Yang and
Slingo 2001).
To investigate the effect of such approximations in
the model’s radiation calculation, a sensitivity experiment was performed with a full radiation calculation
carried out every time step, hereafter referred to as the
‘‘full radiation’’ experiment. Due to the high computational cost, the integration was run for only 15 months,
with the final 12 months retained from March 1979–
February 1980 to provide data for each season. Figure
9a shows the mean diurnal evolution of surface air temperature in the control and full radiation experiments.
Although the results from the full radiation experiment
show a less rapid rise in surface temperature early in
the day, changes in the diurnal cycle of convective rainfall (Fig. 9b) are slight. There is a systematic increase
in precipitation of between 1 and 2 mm day 21 between
0600 and 1800 local time but no significant shift in the
phase of the maximum precipitation. There does, however, appear to be a shift in the phase of the maximum
outgoing longwave radiation (OLR) by about 2 h from
1100 to 1300 local time (Fig. 9c). This can be related
to an increase in the terminal detrainment from the enhanced convection (Fig. 9d), evident in the increase in
upper-tropospheric cloud of almost 10% over the land
grid points (Fig. 9f). Over the ocean grid points of the
Maritime Continent, the characteristics of the diurnal
cycle are mainly unchanged in the full radiation experiment.
Although an accurate treatment of the diurnal cycle
in the radiation calculation has not had a major impact
on the phase of the diurnal cycle, nevertheless the
changes in strength of the diurnal cycle appear to affect
the mean climate of the model. Figure 10a shows the
843
precipitation difference between the full radiation and
control experiments. In general, over the islands there
is a net increase in precipitation of around 1 mm day 21 ,
consistent with the 1–2 mm day 21 increase seen during
the daytime in the diurnal cycle. In response to these
localized changes in heating there are larger-scale
changes in the circulation. These lead to a general enhancement of precipitation on and south of the equator,
coupled with a decrease north of the equator. They go
some way to correcting for the geographical distribution
of the systematic precipitation errors shown in Fig. 1,
although, unlike the no islands experiment, the changes
are not significant in the context of the ensemble of
AMIP II integrations described earlier (see Fig. 6).
The full radiation experiment has enabled two important conclusions to be drawn. The first is that errors
in the phase of the diurnal cycle over land are related
to more fundamental errors in the physical processes of
the model, such as the evolution of the convection field,
as discussed by Yang and Slingo (2001). Second, it has
demonstrated that even quite small but systematic
changes to the diurnal cycle can rectify onto the mean
climate, suggesting that significant improvements to the
diurnal cycle over the Maritime Continent could have
the potential to make a major impact on the model’s
large-scale systematic errors.
5. Global effects of systematic errors over the
Maritime Continent
In both the no islands and full radiation experiments,
there is an enhancement of the heat source over the
Maritime Continent region. On seasonal to interannual
timescales this will have an impact away from this region through the generation of Rossby waves. Therefore, it is appropriate to assess the global effects of this
enhanced heating, and to determine the possible impacts
on model error if improvements in the mean climate of
the Maritime Continent were achieved. The opposite
side of this argument is that we can also assess where
the model may be in error remotely, because of the errors
associated with the Maritime Continent heat source.
As identified in the analysis of the standard AMIP II
experiments, the model underestimates the tropical convective heat source over the Maritime Continent, compared to observations, while creating local maxima over
the west Indian Ocean and west Pacific. The increased
deep convection in the no islands experiment enhances
the divergent outflow and, depending on the ambient
conditions, has the potential to influence remote areas
of the globe through planetary wave propagation, as
proposed by Sardeshmukh and Hoskins (1988). Figure
11 shows that this is certainly the case. The change in
the tropical heat source in the no islands experiment has
significant remote effects, which give rise to large-scale
surface temperature increases over western Russia and
Scandinavia, as well as to other significant changes over
North America. The extratropical surface temperature
844
JOURNAL OF CLIMATE
VOLUME 16
FIG. 9. Comparison of the mean diurnal cycle over the Maritime Continent land grid points from 12month integrations of the control and full radiation experiments: (a) surface air temperature (8C) using 3hourly radiation time steps (solid line) and 0.5-hourly radiation time steps (dashed line); (b) as in (a) but
for precipitation (mm day 21 ); (c) as in (a) but for OLR (W m 22 ); (d) change in the mean diurnal cycle of
convective heating (8C day 21 ), 0.5-hourly radiation time steps minus 3-hourly radiation time steps; (e) as
in (d) but for convective moistening (g kg 21 day 21 ); (f ) as in (d) but for total cloud cover fraction.
changes in Figs. 11a,b are a consequence of changes in
the global circulation, which can be described by the
500-hPa geopotential height anomalies in Figs. 11c,d,
respectively. Only the winter hemisphere is shown in
each case since it is in this season that the ambient
conditions are most conducive to tropical–extratropical
interaction.
The changes in surface temperature and 500-hPa geopotential height can be compared with the UM’s systematic errors in these fields, as shown in Fig. 12. This
reveals the importance of an improved representation of
the Maritime Continent heat source. In DJF the Northern
Hemisphere is mainly dominated by a cold bias in surface temperature, although over North America the
model has a warm bias over Canada and a cold bias
over the United States and Mexico. This general largescale pattern of systematic errors is substantially corrected in the no islands experiment by changes of 0.58–
18C (Fig. 11a). Similarly over Scandinavia and northeast
Russia, the systematic cold bias of over 58C is partially
1 MARCH 2003
NEALE AND SLINGO
845
FIG. 10. Change in the mean climate over the Maritime Continent from the full radiation experiment compared with the control
integration: (a) precipitation (mm day 21 ); (b) OLR (W m 22 ).
corrected by the changes in the no islands experiment
with an increase in surface temperature of over 28C.
Errors in Southern Hemisphere surface temperature during JJA are complicated by the problems with model
sea ice on the periphery of the Antarctic continent (Fig.
12b).
The 500-hPa height changes during both DJF and JJA
(Figs. 11c,d) in the no islands experiment lead to more
obvious improvements in the model’s systematic errors
(Figs. 12c,d). Again the effects are wide ranging, with
substantial changes in the stationary waves over the
Euro–Atlantic sector. The potential for heating anomalies over the warm pool to influence the Euro–Atlantic
sector is consistent with the response of the global circulation to La Niña. The pattern of 500-hPa geopotential
height anomalies over the Euro-Atlantic sector in Fig.
11c is reminiscent of that associated with the 1998/99
La Niña. This was also characterized by enhanced convection over the Maritime Continent and west Pacific
regions (Dong et al. 2000), similar to the changes in the
FIG. 11. Seasonal mean differences averaged for 1979–95 between the no islands and the control experiments: (a) DJF Northern Hemisphere
surface air temperature (8C); (b) JJA Southern Hemisphere surface air temperature (8C); (c) DJF Northern Hemisphere 500-hPa geopotential
height (m); (d) JJA Southern Hemisphere 500-hPa geopotential height (m).
846
JOURNAL OF CLIMATE
VOLUME 16
FIG. 12. Seasonal mean errors in the standard AMIP II ensemble experiments: (a) DJF Northern Hemisphere surface air temperature (8C);
(b) JJA Southern Hemisphere surface air temperature (8C); (c) DJF Northern Hemisphere 500-hPa geopotential height (m); (d) JJA Southern
Hemisphere 500-hPa geopotential height (m).
heating pattern between the control and no islands experiments.
In JJA, the 500-hPa geopotential height anomalies
over the southern oceans are clearly linked to an equivalent barotropic planetary wave response emanating
from the enhanced convection over the Maritime Continent region. However, in DJF the wave propagation
patterns are less coherent and, if anything, the surface
temperature anomalies over western Russia appear to
be the result of upstream development from the Maritime Continent. During DJF, although there is a net increase in precipitation over the Maritime Continent, the
change in the heating pattern is less coherent spatially.
This leads to a smaller-scale increase in upper-level divergence over the region, which does not excite the
coherent wave activity seen in JJA, where there is a
larger and more expansive region of enhanced upperlevel divergence.
Although the mechanism for the remote, extratropical
response in DJF is unclear, it is certainly the case that
these are significant changes. The same patterns were
reproduced in both no islands integrations, and the
anomalies were statistically different from the intraensemble variability of the climate in the standard AMIP
II experiments. Such changes to the global circulation
raise two points. First, they show that it is vital that the
hydrological cycle over the Maritime Continent is simulated accurately since the effects of correcting the local
heat source are indeed global. Also, if tropical changes
in heat sources can have such dramatic effects on the
surface temperatures of the continents of the mid-latitudes and high latitudes, then they serve as a warning
when trying to correct locally for biases in the model.
For example, over western Eurasia, the significant cold
bias of up to 88C has been attributed to problems with
freezing of soil moisture. However, it has been shown
here that even a modest improvement in the heat source
over the Maritime Continent can account for up to 28C
of this bias.
6. Discussion and conclusions
This paper has investigated the errors in the simulation of the climate of the Maritime Continent in the
Met Office Unified Model. The lack of precipitation
over the region is an ubiquitous feature of the model
results during all seasons, as shown in standard AMIP
II experiments. These errors persist even in AMIP II
integrations at much higher resolution, implying that,
at these resolutions, deficiencies in the representation of
the physical system are primarily responsible.
The importance of the Maritime Continent climate
has been demonstrated using AMIP II sensitivity experiments. By removing the island grid points of the
region and replacing them with oceanic grid points,
some aspects of the mean climate are improved. Precipitation increases to be closer to observed amounts
and the flow patterns over the Indian Ocean during the
Asian monsoon season are improved, such that the excessive circulation strength is reduced. However, sig-
1 MARCH 2003
847
NEALE AND SLINGO
nificant errors remain. In particular the excessive lowertropospheric easterlies in the west Pacific have been
made worse by the removal of the island grid points.
Following the results of Slingo et al. (1996), which
suggested that the MJO was sensitive to the mean climate in the warm pool region, it was hypothesized that
the no islands experiment might lead to an improved
MJO. Diagnosis of the MJO shows no significant change
in this experiment suggesting that, like the diurnal cycle,
it is more fundamentally dependent on aspects of the
model’s formulation, such as the vertical resolution
(e.g., Inness et al. 2001), interaction with the ocean
surface (Inness and Slingo 2003), and the physical parametrizations.
The influence of the Maritime Continent has been
shown to be global. A robust teleconnection pattern is
seen, particularly in the Southern Hemisphere winter
season, which is consistent with an equivalent barotropic Rossby wave response to an enhanced tropospheric heat source located over the Maritime Continent.
Such a response raises important questions concerning
the methodology for tackling systematic errors in climate models. It is possible that model errors in a particular region, for example, Eurasia, may be attributed
to local problems such as soil moisture freezing (e.g.,
Viterbo et al. 1999). However, given that improvements
over the Maritime Continent have been shown to lead
to significant changes in remote locations, then care has
to be taken when addressing systematic errors based
entirely on local processes.
Although no final solution to the systematic errors in
the model’s tropical heating distribution has been presented here, this paper has highlighted the importance
of the diurnal cycle for the climate of the Maritime
Continent. Recent analysis of high-resolution satellite
data by Yang and Slingo (2001) has demonstrated the
coherent propagation of convection away from the islands, indicative of gravity waves. In particular, coastal
regions with a large dry bias are strongly collocated
with regions where these propagating diurnal signals are
absent in the model. It is hypothesized that capturing
or better representing the effects of these diurnally
forced gravity waves and land–sea breezes may lead to
improvements in the mean precipitation field in two
ways. First, they may increase the wind variability near
the surface leading to enhanced surface fluxes, particularly of moisture. Second, they may act as triggers for
convection.
It is clear that models with coarse-grid domains are
incapable of capturing the smaller-scale land–sea breeze
circulations and diurnal variability, which are likely to
be key processes in the regions of largest dry bias
around the Maritime Continent in the UM. At the same
time, experiments that have addressed island-scale convection at smaller scales (e.g., Saito et al. 2001) do not
attempt to reproduce the large-scale organization in the
off-coastal regions by the diurnal forcing initiated over
land, essentially the interaction of convection with the
larger scales. Therefore, to understand relevant processes an intermediate regional modeling approach is
needed. This requires the domain of the Maritime Continent to be at fine enough resolution to capture the
diurnal variability and sea breezes initiated over the island regions, but with a domain large enough to simulate
the large-scale forcing and organization. Such a modeling activity is currently being undertaken with the goal
of developing a parameterization of the subgrid mesoscale organization in the context of a fractional tiling
of coastal grid cells, which takes into account the different surface fluxes from land and sea.
Much work is still needed to improve the subgridscale organization aspect of convective parameterization
in course-grid-scale general circulation models, both
through collaborative programs aimed at better understanding the physics of convection [e.g., European Project on Cloud Systems in Climate Models (EUROCS),
online at http://www.cnrm.meteo.fr/gcss/EUROCS/EUROCS.html] and the use of novel techniques such as
grid-box cloud resolving model sampling (e.g., Grabowski 2001). In the intermediate term the use of a
parameterization that is able to represent the known
dominant organization processes related to sea breezes
is considered to be the best option for providing improvements to the simulation of the Maritime Continent
climate.
Acknowledgments. The authors acknowledge many
useful contributions from the CGAM Tropical Group
and the much appreciated input from three anonymous
reviewers. Richard Neale is supported by the Met Office
through grant Met 1b/2601. J. Slingo acknowledges support through the Natural Environment Research Council
(NERC)–funded UK Universities’ Global Atmospheric
Modeling Programme.
REFERENCES
Chang, C. P., and K. M. Lau, 1982: Short term planetary-scale interactions over the tropics and mid-latitudes during northern winter. Part I: Contrasts between active and inactive periods. Mon.
Wea. Rev., 110, 933–946.
Cox, P. M., R. A. Betts, C. B. Bunton, R. L. H. Essery, P. R. Rowntree,
and J. Smith, 1999: The impact of new land surface physics on
the GCM simulation of climate sensitivity. Climate Dyn., 15,
183–203.
Dong, B. W., R. T. Sutton, S. P. Jewson, A. O’Neill, and J. M. Slingo,
2000: Predictable winter climate in the North Atlantic sector
during the 1997–1999 ENSO cycle. Geophys. Res. Lett., 27, 985–
988.
Edwards, J. M., and A. Slingo, 1996: Studies with a flexible new
radiation code. I. Choosing a configuration for a large-scale model. Quart. J. Roy. Meteor. Soc., 122, 689–719.
Gates, W. L., 1992: AMIP: The atmospheric model intercomparison
project. Bull. Amer. Meteor. Soc., 73, 1962–1970.
Gibson, J. K., P. Kallberg, S. Uppsala, A. Hernandez, A. Nomura,
and E. Serrano, 1997: ERA Description. ECMWF Re-analysis,
Project Report Series 1, Tech. Rep., ECMWF, Reading, United
Kingdom, 72 pp.
Grabowski, W. W., 2001: Coupling cloud processes with the large-
848
JOURNAL OF CLIMATE
scale dynamics using the Cloud-Resolving Convection Parameterization (CRCP). J. Atmos. Sci., 58, 978–997.
Gregory, D., and P. R. Rowntree, 1990: A mass flux convection
scheme with representation of cloud ensemble characteristics and
stability dependent closure. Mon. Wea. Rev., 118, 1483–1506.
Holland, G. J., and T. D. Keenan, 1980: Diurnal variations of convection over the ‘‘maritime continent.’’ Mon. Wea. Rev., 108,
223–225.
Houze, R. A., S. G. Geotis, F. D. Marks, and A. K. West, 1981:
Winter monsoon convection in the vicinity of North Borneo.
Part I: Structure and time variation of the clouds and precipitation. Mon. Wea. Rev., 109, 1595–1614.
Inness, P. M., and J. M. Slingo, 2003: Simulation of the Madden–
Julian oscillation in a coupled general circulation model. Part I:
Comparison with observations and an atmospheric-only GCM.
J. Climate, 16, 345–364.
——, ——, S. J. Woolnough, R. B. Neale, and V. D. Pope, 2001:
Organization of tropical convection in a GCM with varying horizontal resolution: Implications for the simulation of the Madden
Julian Oscillation. Climate Dyn., 17, 777–793.
Keenan, T. D., B. R. Morton, M. J. Manton, and G. J. Holland, 1989:
The island thunderstorm experiment (ITEX)—a study of tropical
thunderstorms in the maritime continent. Bull. Amer. Meteor.
Soc., 70, 152–159.
——, and Coauthors, 2000: The Maritime Continent Thunderstorm
Experiment (MCTEX): Overview and some results. Bull. Amer.
Meteor. Soc., 81, 2433–2455.
Kershaw, R., and D. Gregory, 1997: Parameterization of momentum
transport by convection. I: Theory and cloud modelling results.
Quart. J. Roy. Meteor. Soc., 123, 1133–1151.
Lau, K. M., C. P. Chang, and P. H. Chan, 1983: Short term planetaryscale interactions over the Tropics and mid-latitudes during
northern winter. Part II: Winter-MONEX period. Mon. Wea. Rev.,
111, 1372–1388.
Legates, D. R., and C. J. Willmott, 1990: Mean seasonal and spatial
variability in global surface air temperature. Theor. Appl. Climatol., 41, 11–21.
Martin, G. M., 1999: The simulation of the Asian summer monsoon,
and its sensitivity to horizontal resolution, in the UK Meteorological Office Unified Model. Quart. J. Roy. Meteor. Soc., 125,
1499–1525.
Morcrette, J. J., 2002: Assessment of the ECMWF model cloudiness
VOLUME 16
and surface radiation fields at the ARM SGP site. Mon. Wea.
Rev., 130, 257–277.
Philander, S. G. H., 1985: El Niño and La Niña. J. Atmos. Sci., 42,
652–662.
Pope, V. D., M. L. Gallani, P. R. Rowntree, and R. A. Stratton, 2000:
The impact of new physical parameterizations in the Hadley
Centre climate model: HadAM3. Climate Dyn., 16, 123–146.
Ramage, C. S., 1968: Role of a tropical ‘‘maritime continent’’ in the
atmospheric circulation. Mon. Wea. Rev., 96, 365–370.
Saito, K., T. Keenan, G. Holland, and K. Puri, 2001: Numerical simulation of the diurnal evolution of tropical convection over the
Maritime Continent. Mon. Wea. Rev., 129, 378–400.
Sardeshmukh, P. D., and B. J. Hoskins, 1988: The generation of global
rotational flow by steady idealized tropical divergence. J. Atmos.
Sci., 45, 1228–1251.
Slingo, J. M., and Coauthors, 1996: Intraseasonal oscillations in 15
atmospheric general circulation models: Results from an AMIP
diagnostic subproject. Climate Dyn., 12, 325–357.
Stratton, R. A., 1999: The impact of increasing horizontal resolution
on the HADAM3 climate simulation. Hadley Centre Tech. Note
13, Hadley Centre, Met Office, United Kingdom.
Sui, C. H., and K. M. Lau, 1992: Multiscale phenomena in the tropical
atmosphere over the western Pacific. Mon. Wea. Rev., 120, 407–
430.
Viterbo, P., A. Beljaars, J. F. Mahfouf, and J. Teixeira, 1999: The
representation of soil moisture freezing and its impact on the
stable boundary layer. Quart. J. Roy. Meteor. Soc., 125, 2401–
2426.
Williams, M., and R. A. Houze, 1987: Satellite-observed characteristics of winter monsoon cloud clusters. Mon. Wea. Rev., 115,
505–519.
Xie, P., and P. A. Arkin, 1996: Analyses of global monthly precipitation using gauge observations, satellite estimates, and numerical model predictions. J. Climate, 9, 840–858.
Yang, G. Y., and J. M. Slingo, 2001: The diurnal cycle in the Tropics.
Mon. Wea. Rev., 129, 784–801.
Zhang, Y., K. R. Sperber, and J. S. Boyle, 1997: Climatology and
interannual variability of the East Asian winter monsoon: Results
from the 1979–95 NCEP/NCAR reanalysis. Mon. Wea. Rev.,
125, 2605–2619.
Zhu, B., and B. Wang, 1993: The 30–60 day convection seesaw
between the tropical Indian and western Pacific Oceans. J. Atmos.
Sci., 50, 184–199.