Sie sind auf Seite 1von 31

Q. J. R. Meteorol. SOC. (1996). 122, pp.

23-53
An improved linear model of tropical surface wind variability
By N. H. SAJI and B. N. GOSWAMI'
Indian Institute of Science, Bangalore, India
(Received 28 January 1994; revised 3 April 1995)
SUMMARY
For coupled modelling studies, the atmospheric component is required to simulate the large-scale part of the
observed surface winds accurately, as only this part is responsible for driving low-frequency interannual variability
in the oceans. A simple linear model of the tropical atmosphere is developed in which surface winds are viewed
as the response of the atmospheric boundary layer to pressure perturbations produced by deep convection and
sea surface temperature (SST) gradients. An empirical parametrization of total convection as a nonlinear function
of total SST is adopted. The large-scale variability of the simulated surface winds is examined for the period
1974-1991. The model not only captures the large-scale low-frequency part of the surface winds given by the first
two empirical orthogonal functions (EOFs), it also successfully simulates the equatorial surface wind anomalies
and their evolution during the entire period 1974-1991.
Certain interesting differences between precipitation forcing and SST induced forcing of surface winds is
brought out in this study. The precipitation forcing was found to be dominant over the central and western Pacific,
while the SST forcing was found to be dominant over the eastern Pacific. The resultant wind response was also
found to reflect this behaviour. The most interesting result is the change in balance of the precipitation-related
and SST-gradient-related terms in the forcing of anomalous and climatological winds. Our results indicate that
convective heating predominates over SST-gradient-induced effects in forcing anomalous winds, but the balance
reverses in the case of the climatological winds. Wealso find that SST-gradient effects are critical in the simulation
of the climatological wind field.
KEYWORDS: Pacific surface winds Tropical linear model Deep convection Boundary-layer pressure
gradients
1. INTRODUCTION
The surface winds play a crucial role in tropical air-sea interactions. They drive'the
ocean currents that determine the evolution of the sea surface temperature (SST). The SST
modulates the atmospheric heating which in turn determines the surface winds. Such air-
sea interactions are at the heart of the observed interannual variability in the tropics such as
the El Niiio southern oscillation (ENSO) phenomenon. Coupled ocean-atmosphere models
of different complexity are being employed to simulate the observed interannual variability
(e.g. Neelin et al. 1992; Zebiak and Cane 1987; Philander et al. 1992; Lau et al. 1992; Latif
et al. 1993; Nagai et al. 1992). However, most coupled models suffer from certain climate
drifts. These climate drifts may be due either to imperfection in the atmospheric or oceanic
components or to imperfection in the parametrization of the coupling processes (Neelin
et al. 1992). Therefore, the ability of the model's atmospheric component to simulate the
surface winds realistically when forced by observed SST is a prerequisite for using it in a
coupled model. The atmospheric component used in various coupled models range from a
simple steady Gill model of Zebiak and Cane (1987) or the hybrid model of Neelin (1990),
through low-resolution atmospheric general circulation models (AGCM) of Philander et al.
(1992), Lau etal. (1992) and Latif etal. (1993), to the somewhat higher-resolution AGCM
of Nagai et al. (1992). Most atmospheric models including low-resolution AGCMs have
certain systematic errors in simulating the interannual variability of the tropical surface
wind when forced by observed SST. The simple Gill type models with atmospheric heating
proportional to SST anomalies (SSTA) are known to produce far too strong easterlies in
the eastern Pacific during warm phases (Zebiak 1986; Goswami and Shukla 1991). The
low-resolution AGCMs also fail to simulate correctly the location and asymmetry about
the equator of the westerly anomalies associated with the warm phase (Graham et al.
1989; Latif et al. 1990; Kitoh 1991). The objective of this study is to develop a simple
* Corresponding author: Centre for Atmospheric Sciences, Indian Institute of Science, Banglore 560 012, India.
23
24 N. H. SNl and B. N. GOSWAMI
atmospheric model for better simulation of the tropical surface winds. Development of a
model of this kind is guided by the two factors set out below.
First, we note that the primary objective of the coupled model is to simulate the annual
and interannual variations of the tropical climate. If one is interested only in these low-
frequency variations, is it necessary for the atmospheric model to simulate the observed
surface wind in all its details? The observed surface wind field is known to have significant
small-scale high-frequency variability. Is this small-scale high-frequency component of the
surface winds important in forcing the interannual variability in the oceanic component?
Recent studies (Latif et al. 1990 and Goswami and Shukla 1991) show that only the large-
scale part of the observed surface winds represented by the first two or three empirical
orthogonal functions (EOFs) are necessary to force the observed interannual variability
in the ocean. These studies show that the forcing associated with the small-scale high-
frequency part of the surface winds is not important in forcing the interannual variability
in the ocean; they simply act as noise on the forcing associated with the large-scale part
of the atmospheric winds. Thus, an atmospheric model that is capable of simulating the
large-scale part of the surface winds accurately, but which does not necessarily simulate
the small-scale part of the surface winds well, may still be a good choice for using in a
coupled model. The recognition of this fact also renders hope that even a relatively simple
atmospheric model which is successful in simulating the large-scale part of the surface
winds accurately would provide a good atmospheric component for a coupled model.
These arguments motivated us in our objective of developing a simple atmospheric model.
Secondly, we deal with the subject of the important dynamical factors that govern
the maintenance of the observed surface winds in the tropics. Murphree and van den
Dool (1988) studied the momentum balance for the monthly mean surface winds and
concluded that nonlinearity played only a minor role in the momentum balance. Zebiak
(1990) studied the vorticity budget of monthly surface wind anomalies in the tropical
Pacific and also concluded that the nonlinear advection terms are of second order. He then
proceeded to infer the forcing field, that would be required to drive the observed surface
winds in a linear model, by solving the inverse problem. The interesting result that emerged
from this study was that the inferred forcing field was very different from the sea surface
temperature anomaly (SSTA) field but bore close resemblance to highly reflective cloud
(HRC) anomalies that may be closely related to the atmospheric heating. Thus, the problem
with many simple linear models used in earlier studies to simulate surface winds was not
that they neglected nonlinearities but that they parametrized atmospheric heating as being
directly proportional to SSTA. Such a parametrization fails to simulate both the location and
horizontal scale of the atmospheric heating. For example, during a mature warm phase of
ENSO, the SSTA maximum is located in the eastern part of the Pacific but the atmospheric
heating inferred from outgoing longwave radiation anomalies (OLR) or HRC is centred
around the date line. Similarly, Neelin (1988) showed that if a linear model is forced with
a GCM precipitation field it can simulate tropical winds that closely resemble the surface
winds simulated by the nonlinear GCM. Therefore, a linear model may be sufficient for
simulation of the surface winds if the model can parametrize the atmospheric heating field
correctly. However, the organized convective heating in the tropics is governed by nonlinear
dynamical and thermodynamical processes. To parametrize it within the framework of a
linear model is a daunting task. An attempt to do this by including a simple moisture
balance equation has not been successful (Seager 1991); therefore, we shall not attempt to
calculate the atmospheric heating following some simple dynamical framework. Instead,
we estimate it from SST on the basis of an empirical relation between SST and observed
precipitation. (Observed precipitation here refers to the precipitation derived from OLR
as described in the next section.) In section 3, we describe the basic model. Although it is
TROPICAL LINEAR MODEL 25
a linear model, it includes both processes that drive the surface wind, namely deep heating
of the atmosphere and surface pressure gradients associated with horizontal SST gradients.
The background for developing this empirical parametrization of the large-scale part of the
atmospheric heating, and the method itself, is discussed in section 3. Section 4 describes
results from our simple model and brings out the contribution from the two processes.
Results are summarized and some concluding remarks are made in section 5.
2. DATAUSED
To verify the simulation of the surface wind variability by our model, we used zonal
and meridional components of the monthly mean surface winds from the comprehensive
ocean-atmosphere data-set, (COADS, Slutz et al. 1985) for the period 1974-1987, and
from the European Centre for Medium-Range Weather Forecasts (ECMWF) analyses at
1000 mb for the period 1988-1991. As we are primarily interested in the simulation of
the low-frequency variability, a nine-month running mean was used in both model and
observed winds. SST data was taken from COADS for the period 1974-1987 and from
the blended SST product of Reynolds (Reynolds 1988; Reynolds and Marisco 1993) for
the rest of the period. Both winds and SST data are available for 2" x 2" boxes. The pre-
cipitation anomalies are derived from outgoing longwave radiation using the empirical
relation given by Vernekar (Yo0 and Carton 1988). The monthly mean OLR data is based
on observations from the NOAA polar orbiting operational satellites (Gruber and Winston
1978). The data cover the period between J une 1974 and May 1993 with a gap between
March and December 1978. The OLR data are available for 2.5" x 2.5" boxes. From esti-
mates of precipitation for individual months at each gridpoint, climatological precipitation
estimates have been calculated. By subtracting the climatological means from individual
observations we derived the precipitation anomalies for each month.
3. THEMODEL
The surface winds in the tropics may be viewed as the response of the atmospheric
boundary layer to two types of forcing: deep convection and SST-gradient-induced effects.
Gill (1980) studied the response of the tropical atmosphere to the forcing associated with
latent heat release in deep cumulus convection using a simple linear model. The Gill
model can be viewed as a model of the boundary-layer flow alone, which is forced by
precipitation (see Neelin 1988). Lindzen and Nigam (1987) put forth the idea that surface
winds in the tropics are also forced by pressure gradients which develop hydrostatically in
the turbulently mixed boundary layer because of the underlying SST gradients. However,
until recently the attitude has been to consider only one of these two processes, completely
excluding the other, in modelling tropical surface winds. Lately, the importance of both the
processes has been recognized and some of the recent work reflects this (Gutzler and Wood
1990; Wang and Li 1993; Eltahir and Bras 1993). Werecognize that the aforementioned
processes should be viewed as complementary. One or the other of these processes may
be the dominant one in a given geographical location. Thus, in the western Pacific warm
pool region with high SST and weak SST gradients, conditions are more favourable for
deep convection, whereas the strong SST gradients in the eastern Pacific cold tongue could
make boundary-layer pressure gradients the dominant mechanism. In the present work,
we have incorporated both mechanisms in a simple linear model for the tropical Pacific
and we have investigated the relative importance of these mechanisms in forcing surface
winds.
26 N. H. SAJI and B. N. GOSWAMI
The boundary-layer response to the mid-tropospheric latent heat forcing associated
with the large-scale part of organized deep convection is modelled using a Gill-type model,
Vi z.
u; - f u + 4; + EXU = 0
u; + fu + 4; + Ey U = 0
4; + c(u; + u;> + &/ = -a P
where u and u are perturbations in the horizontal component of mass flux in the boundary
layer, 4 is the mass-weighted integral of the perturbation in geopotential height in the
boundary layer, f is the Coriolis parameter, E, is the coefficient of zonal Rayleigh friction,
cY that for meridional Rayleigh friction, E+ is the coefficient of Newtonian cooling, P is the
precipitation anomaly, and a = L RAp/(2pcP) where L is the latent heat of condensation,
R is the gas constant for dry air, cp is the specific heat capacity of dry air at constant
pressure, Ap is half the depth of the troposphere (about 500mb) corresponding to the
region heated by condensation of water vapour, and p is a mid-level tropospheric pressure
(about 500 mb); c is the wave speed defined as c2 = SRAp2/2p, where S is static stability
(taken to be 4.2 x lo- m K s2/kg).
A note about the friction used before we proceed further. Li and Wang (1994) have
demonstrated that the use of latitude-dependent friction coefficients is critical for the
successful simulation of the surface wind field in a three-force balance boundary-layer
model. From a momentum budget study of the tropical surface winds they found that
the zonal friction coefficient is smallest at the equator and increases with latitude. The
meridional friction coefficient was found to be smallest between the equator and latitude
5 S, and largest around latitudes 20N and 20s. Following them we prescribe a value for
the zonal friction coefficient, t x, such that it is 0.8 x lo- s-l at latitude 2N and increases
to 2.8 x ~ O - S - ~ at latitudes 30N and 30s. The meridional friction coefficient in our
model has the value 2.1 x lO-s-l at latitude 4s increasing to about 2.9 x lo- s-l at
latitudes 20N and 20S, thereafter decreasing to about 2.5 x lO-s-l at latitudes 30N
and 30s. The Newtonian cooling coefficient, E ~ , is constant (1 x lo- s-l).
For the part of the boundary-layer flow due to pressure gradients arising from the
SST gradients, we use the Lindzen-Nigam (Lindzen and Nigam 1987, hereafter referred
to as LN) formulation with the back-pressure effect included. As Neelin (1989) has
demonstrated, the Lindzen-Nigam model can be transformed into the following form,
which is similar to the Gill model,
u; - fu + 4; + E,U = 0
u; + f u r + 4; + Ey U = 0
4: + C(U; + u;) + E& = -bTsr
here T, is the SST anomaly, b = E 6p H0/2To, where 6p is the depth of the boundary layer
in units of pressure (about 300 mb), HO is the depth of the boundary layer in units of length
(about 3 x lo3 m) and To is a constant reference temperature (To = 288 K), (Lindzen and
Nigam 1987); 4 = g(h - (H0/2T0)T,), where h is the perturbation to the height of the
boundary layer.
We exploit the similarity and linearity of Eqs. (1) and (2) to combine them in the
following set of equations:
TROPICAL LINEAR MODEL 27
where 4 = i(4 + 4 ).
In this case P represents the precipitation anomalies associated with deep convec-
tion. Gill (1980) assumed very simple idealized forms for P with which he could obtain
analytic solutions. In reality P assumes complicated forms and the leading problem was
how to parametrize this quantity in terms of the underlying SST with which it was known
to be significantly related. Thus Zebiak (1982), in a simple model to explain the surface
wind anomalies associated with El Niiio, adopted Bjerkness (1969) hypothesis that pre-
cipitation anomalies were linked to SST anomalies through the concomitant anomalies in
evaporation. To quantify the evaporative anomalies associated with the SSTAs he used the
linearized Clausius-Clapeyron relation. In a later version, Zebiak (1986) incorporated a
convergence feedback mechanism which was designed to mimic conditional instability of
the second kind (CISK). Neelin and Held (1987) used the concept of moist static energy
in parametrizing deep convection. Seagers (1991) hypothesis was that buoyancy fluxes
were more important than the convergence field in determining deep convection. However
none of these schemes have been successful in representing the large-scale organized part
of deep convection. Werealize that modelling the nonlinear moist processes that produce
deep convection and its organization into large-scale systems is a formidable task within
the framework of a simple linear model. Thus we were led to approach this problem in a dif-
ferent way4.e. by looking at the observed relationship between SST and deep convection
and formulating an empirical parametrization based on this.
The variability of cumulus convection in relation to the variability of SST has been
the subject of a fairly large amount of study. In many of these studies OLR has been used
as a proxy for cumulus convection. The most interesting finding that has evolved from
these studies is that deep convection is supported only in regions where the SST is above
a critical value, T,, which is about 27.5 C (Gadgil et al. 1984; Graham and Barnett 1987;
Fu et al. 1990). However, one of the puzzling observations was the large scatter in the
OLR-SST relation beyond T, leading to the inference that SST and deep convection are
poorly correlated beyond T,. Recent studies by Fu et al. (1990), Zhang (1993) and Waliser
et al. (1993) have to a great extent cleared up this issue. These studies, based on larger
data-sets of OLR as well as of HRC and ISCCP (International Satellite Cloud Climatology
Project) stage C2 data-set containing cloud types, their frequency of occurrence and cloud-
top temperatures, have shown that below 26 C deep organized convection rarely occurs.
Both frequency and intensity of deep convection increase dramatically with SST between
26.5 C and 29.5 C, and decay with further increase in SST. Moreover, in the regions
with SST greater than 27 C, situations with both no deep convection and vigorous deep
convection are encountered, leading to a higher variability of deep convection for a given
temperature. Decrease of convective activity for SST greater than 29.5 C shows that the
maximum convective activity does not occur over the warmest water (SST > 29.5 C) but
rather that the warmest water occurs under clear (less convective) skies. These studies
have clearly established that although there is larger variability of convection with SST in
regions of high SST, the mean atmospheric heating increases dramatically in these regions.
This increase in atmospheric heating is the result of the increase in frequency of occurrence
and intensity of deep convection. Another interesting observation made by Fu et al. (1990)
is that in the warm pool region with SST greater than 28 C, strong surface convergence
does not enhance deep convection. However, if the warmest SST in a region is between
28 N. H. SAJI and B. N. GOSWAMI
*. .. *
. .
25 26 27 28 28
SST
Figure 1. Scatter plot of observed SST vs. precipitation (estimated from observed OLR using Eq. (4)) and the
modelled curve of precipitation as a function of SST (thick line). The thin line represents the mean value of the
observed precipitation over each 0.5 " C SST bin.
26 "C and 28 "C, the deep convection is significantly enhanced by strong surface wind
convergence.
With this background, and keeping in mind that an increase in organized convection
with SST from 26.5 "C to 29.5 "C is quite nonlinear, we proceed to parametrize the large-
scale part of the convective heating in the following manner. First we estimate precipitation
from an empirical formula similar to the one proposed by Vernekar (Yo0 and Carton 1988),
viz.
P(mm d-') = 48.38 - 0.186 OLR(W m-')
(4)
Figure 1 is a scatter plot of concomitant values of observed SST and OLR-derived
precipitation (as estimated from Eq. (4)) over the tropical Pacific domain (defined as
between 1WE and 60"W and between 30"s and 30'"). The thin curve in the figure is
the mean value of the precipitation calculated over each 0.5 "C SST bin. As discussed
earlier, although variability of precipitation for a given SST increases with SST, the mean
precipitation can be found to be increasing with SST in a nonlinear manner. To model this
large-scale nonlinear increase of precipitation with SST over warm waters, we constructed a
least-square polynomial fit between precipitation obtained with Eq. (4) and the concomitant
values of the observed SST in the tropical Pacific domain. We found that the following
third-degree polynomial fit (represented by the bold line in Fig. 1) captures many of the
features of the observed precipitation field quite well, e.g.
P(mm d-') = 0.0186 x ( SST) 3 - 1.220 x ( SST) 2 + 26.093 x ( S S T) - 179.785.
( 5)
TROPICAL LINEAR MODEL 29
With this parametrization, we estimated the total convective heating for each month
at each gridpoint using the observed SST. Then the convective heating anomalies were
calculated by subtracting from the climatological convective heating based on the entire
period (1974-1991).
As shown in Eq. (3c), the forcing for the tropical surface winds has two components,
one coming from convective precipitation (Gill forcing) and the other from the SST gradi-
ents (LN forcing). The LN forcing anomalies for a given month is the observed SST minus
the climatological SST for that month. Finally, adding the Gill and the LN components we
get the total forcing for the surface wind anomalies.
The model equations 3(a) to (c) were written in a finite difference form with lateral
boundary conditions that require the gradients of dependent variables to vanish in the
direction normal to the boundaries. The model was integrated in the domain 100"E to
60"W and 30"s to 30"N, with 5" resolution in longitude and 2" resolution in latitude. The
model variables were deployed in the Arakawa-C grid, and time marching was performed
with the leap-frog scheme. The model domain contains little land surface and hence it
was not considered necessary to include land surface processes. Steady solutions were
considered to be obtained when the model solutions converged to within a reasonable
accuracy.
4. RESULTS
As emphasized in the Introduction, we are primarily interested in the model's ability
to simulate the low-frequency large-scale part of the surface wind, and in studying the
contribution to the tropical surface winds by the Gill and LN forcings. Therefore, unless
otherwise stated, all results presented here will be for winds (both observed and simulated)
filtered with a 9-month running mean filter. Weshall present the results in two subsections.
In the first we compare the model results with that of the observations. Since the focus
of the paper is in simulating the large-scale part of the variability we present comparisons
between the first two EOFs and the corresponding principal components (PC) of the
simulated and observed wind anomalies. Presented also are time series of equatorial winds
in three regions of the equatorial Pacific which are representative of the western, the central
and the eastern Pacific. The Hovmoller maps of the observed and modelled zonal wind
anomalies are presented next. Last in this subsection is a comparison of the modelled and
observed anomaly wind variance and the correlation map.
In the second part we examine the contribution of the Gill and LN components to the
model variability. Wefocus again on the large-scale features of the two forcings and their
response. Case-studies of a warm event and a cold event are also presented.
(a ) Model versus observations
(i) Comparison between estimated and observed forcing. The first two EOFs and the cor-
responding principal components (PC) of the observed OLR-derived precipitation anomaly
field (Eq. (4)), explaining 23.87% and 14.42% of the total variance, are shown in Fig. 2.
The first EOF reveals an elongated pattern of positive precipitation anomalies (to facilitate
comparison, the patterns associated with the EOFs will be interpreted with respect to a
warm episode) straddling the equator but displaced a little to the south and with a maxi-
mum centred just below the equator extending about 20" east from the date line and with
maximum loading around the date line. In the second EOF, the positive pattern is located
to the east relative to the similar pattern in EOF1. This fact together with the phase lags
of the PC2 relative to PC1 during a warm episode (e.g. 1982-83) reflects the eastward
movement of positive precipitation anomalies during the warm episode. The suppression
30 N. H. SAJI and B. N. GOSWAMI
OBSERVED PREC EOFl 23.87% Q
30N
15N
EO
15s
30s
OBSERVED PREC EOF2 14.42% b
1 i OE 180 1 i o w 1 ow
C
1976 1978 1980 1982 1984 1986 1988 1990
Figure 2. The first two EOFs and the corresponding PCs of the filtered precipitation anomalies estimated from
observed OLR (Eq. (4)). The solid line is PC1, the dashed line is PC2.
of convection to the west of this pattern is also seen in EOF2. Although the method em-
ployed here to estimate precipitation from OLR is probably not universally applicable, we
compared the first two EOFs of precipitation anomaly derived this way with those of the
HRC anomaly shown in Fig. 3. The agreement between them is good, showing that the
empirical formulation captures the dominant patterns of the precipitation anomalies quite
well. The corresponding EOF patterns of the model precipitation field (Eq. (5)) are shown
in Fig. 4. EOFl and EOF2 show close similarity with those of the observations. Particu-
larly well simulated is the location of the centre of the precipitation anomaly maximum
TROPICAL LINEAR MODEL 31
HRC EOFl 18% a
HRC EOF2 12% b
140E 180 140W 1 ow
C
-20 I
1976 1978 1980 1982 1984 1986 1988 1990
Same as for Fig. 2, but for the filtered HRC anomalies. Figure 3.
in EOF1, and the model EOF2 is able to capture the eastward migration during the warm
episodes. Wenote, however, that the meridional scale of the EOFl pattern of the model
precipitation is somewhat larger than that for the corresponding pattern of the observed
precipitation anomalies. Taken together EOFl and EOF2 of the model explain 72% of the
total variance compared to 38% explained by EOFl and EOF2 of the observations. Thus
it is seen that the first two EOFs of both the observations and the model are responsible
for most of the variability. These also are found to be large-scale patterns while higher EOFs
display smaller and smaller spatial structures. That the first two model EOFs are
32
20 -
N. H. SAJI and B. N. GOSWAMI
GILL QEOFl 56.99% a
140E 180 14OW 1OOW
Figure 4. Same as for Fig. 2, but for the filtered model precipitation (Eq. (5)) anomalies.
remarkably similar to those derived from observations demonstrate that the model pre-
cipitation field is able to capture the large-scale part of the observed precipitation field
very well.
(ii) Simulated surface winds with full forcing. The model forcing field as mentioned in
the last section is a linear combination of the Gill forcing and the LN forcing. Therefore
the large-scale features of the full forcing field, as seen from Fig. 7, exhibit features of both
the Gill and the LN forcing fields. Using the full anomalous forcing, we have calculated
TROPICAL LINEAR MODEL 33
OBSERVED WINDS EOF 1 2 9 . 9 1 % a
EQ
15s
. . & , , , , , I l l , . . . . . . . . . : . . I
. . . . . . . . . . . . . . . . . . . . . . . . .
180 14OW 1 ow
072
OBSERVED WINDS EOF 2 18. 11% b
30N
15N
EO
15s
30s
C
I
1976 1978 1980 1982 1984 1986 1988 1990
-30'
Figure 5. The first two EOFs and the corresponding PCs of the filtered vector wind anomalies from observations.
The solid line is PC1, the dashed line is PC2.
the steady responses corresponding to each month between J anuary 1974 and December
1991 with the aid of our model (Eq. (3)). Now we present the results of an EOF analysis
on the surface winds simulated by the model and compare them with the EOFs of the
observed surface winds. The first two EOFs of the observed winds explain about 48%
(EOF1,29.91%; EOF2, 18.11%) of the total variance (Fig. 5). The principal components
reveal that these patterns are associated with the low-frequency part of the surface wind
variability. The most prominent feature of the observed wind EOFl is the narrow belt of
34 N. H. SAJI and B. N. GOSWAMI
15-
-30 I
1976 1978 1980 1982 1984 1986 1988 1990
Figure 6. Same as for Fig. 5, but for the filtered vector wind anomalies simulated by the Gill +LN model.
strong westerly anomalies extending from about 160E to 120W, and centred a little to
the south of the equator. The second EOF reveals the eastward migration of this pattern,
with strong easterlies to the west. There is a divergence zone at around 160W and a
convergence zone in the far south-east Pacific at around 100W. The model EOFl and
EOF2 (Fig. 6) are found to display features similar to the observations. The patch of
westerly anomalies and its eastward migration is well captured by EOFl and EOF2 of
the model winds. However, they display easterlies over the south-east Pacific which are
TROPICAL LINEAR MODEL
G I LL+LN QEOFl 57. 42% a
GI LL+LN QEOF2 14. 55% b
30N -
-
140E 180 140W 1 oow
( O B C
35
-20 J
1976 1978 1980 1982 1984 1986 1988 1990
Same as in Fig. 2, but for the filtered Gill +LN forcing anomalies. Figure 7.
somewhat stronger than observed. It should, however, be noted that the large spurious
easterlies simulated by earlier linear models (e.g. Zebiak 1986) during a warm phase
of the ENS0 are replaced by weaker easterlies much closer to the observations in our
model simulations. This significant improvement may be attributed to our more realistic
parametrization of the convective heating as well as to inclusion of the forcing driven by
surface temperature gradients.
The first two EOFs represent a large part of total low-frequency variability of the
observed winds; and the model's ability to simulate them is noteworthy. In this respect,
36 N. H. SAJI and B. N. GOSWAMI
ZONAL WIND ANOMALIES EQ3 a
1976 1978 1980 1982 1984 1986 1988 1990
ZONAL WIND ANOMALIES EQ 1 C
2
1
0
- 1
-2
Figure 8. Simulated (solid line) and observed (dashed line) zonal wind anomalies averaged over (a) EQ3 (SON-
5 5 ; 15OoE-170"W), @) EQ2 (5"N-S"S; 13OoW-17O0W), (c) EQ1 (S"N-5"S; 90W-1300W). Fig. 8(d) to ( f ) is
the same as for Figs. 8(a) to (c), but for the meridional wind anomalies.
our simple model performs as well as some low resolution GCMs (e.g. Graham et al. 1989;
Latif et al. 1990) used in many coupled models. Although our model does not simulate
the small-scale part of the observed surface winds well, this part not only contains a small
fraction of the total variance but also is irrelevant for low-frequency air-sea interactions
in the tropics (Latif et al. 1990; Goswami and Shukla 1991). The model's success in
simulating the dominant large-scale part of the surface winds makes it a good low-cost
candidate for a coupled model.
TROPICAL LINEAR MODEL
- 1 -
37
. . . . . ;. . .;. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
, I , .
: : '
I ,
' I ,
. * , .
8 ) .
, I
MERIDIONAL WIND ANOMALIES EQ3 d
11
2 -
1 -
- 1 -
0 -
: I :
. . . . . . : . . .; .: . . . . . . . . . . . . . . . . . . I . . . . . . . . . . . . .
- 1 - , I , , .
I , , I .
q
1976 1978 1980 1982 1904 1986 1988 1990
: i'.
. I
. ,--
' :
: . . . . . . . . . . . . .:. . . . . . . .:. . . . . . 8. :. .
.J.
:
I .
, .
. ,.., . #I 1 .
. , """...:'
; , . . / L.. .
;<.,- :-*\
. ,
,At ~ :'., ;. I .- ' ',
,.2 t, : . * .-.* ; -.& . I
. . .
h
c '
: :.-: ; -
o--* & $ ! # ' b1-.?- , ,
A
I ,
,-, ..,
. , . . . . . . . . . . . . . . . . . . . . . > . . . . . . . . . . .
Figure 8. Continued.
(iii) Equatorial winds. Since it is widely recognized that equatorial winds are of special
importance in tropical air-sea interactions, we are documenting here the model simulations
in this crucial region. To do this we first define three regions in the equatorial Pacific,
EQ1, EQ2 and EQ3, which are representative of the eastern, the central and the western
Pacific, respectively. The area-averaged zonal and meridional wind anomalies over EQ1
(5"!%5"N, 90"W-l30"W), EQ2 (5"!%5"N, 130"W-l70"W) and EQ3 (5"!%5"N, 170"W-
150"E) are shown in Figs. 8(a) to (f ). It is found that the observed zonal wind anomalies
exhibit eastward propagation during major events (e.g. 198243,1987). For example during
the 1982-83 event zonal wind anomalies peaked first over the western Pacific around
38 N. H. SAJI and B. N. GOSWAMI
MODEL PREC ANOM b
COADS+ECMWF C MODEL d
Figure 9. Time-longitude sections of (a) observed precipitation anomalies, (b) model precipitation anomalies,
(c) observed zonal wind anomalies and (d) model zonal wind anomalies, all averaged between 6"!+6"N.
TROPICAL LINEAR MODEL 39
December 1982, the strongest anomalies then propagating to the central and then to the
eastern Pacific at later times. The model simulates the magnitude and the propagation of
these events quite well. Also well captured is the magnitude and eastward propagation
characteristics of the 1987 event. It is noted that in general the model equatorial zonal, and
also meridional anomalies, agree well with observations, especially during major events.
However the following discrepancies are noted. As the model is designed to simulate only
the large-scale low-frequency component, it fails to simulate some of the higher-frequency
components. Wealso note that the largest zonal wind errors occur in the eastern Pacific,
and the largest meridional wind errors occur in the western Pacific.
(iv) Time-longitude sections. To gain further insight into our model's ability to parame-
trize the precipitation anomalies and wind anomalies at all times, we show, in Fig. 9, the
time evolution of observed precipitation anomalies and model precipitation anomalies,
observed zonal wind anomalies and simulated zonal wind anomalies averaged between
6"s and 6"N from J anuary 1974 to December 1991. From Fig. 9(a) and (b), we note the
ability of the parametrization scheme to simulate most of the major features of evolution
of precipitation anomalies. The negative precipitation anomalies during 1975, the evolu-
tion of the positive anomalies from 1980 to 1983 followed by the negative anomaly phase
and the positive anomalies during the 1987 warm episode, again followed by the negative
anomaly phase, are all simulated well by our parametrization. This is also reflected in
the good simulation of the equatorial wind anomalies (Fig. 9(c) and (d)). As we men-
tioned earlier, the centres and evolution of the western anomalies during 1982-83, easterly
anomalies during 1984-85 and again westerly anomalies during 1985-87, followed by
easterly anomalies, are all well simulated. The model produced strong easterlies in the
eastern Pacific during the early part of 1982 and stronger than observed westerlies in the
eastern Pacific during 1974-75. The model also simulated the observed easterlies in the
central and eastern Pacific during 1977-1980. On the whole, simulation of the winds by
the simple linear model is quite good.
Although our parametrization of precipitation (Eq. (5)) simulates most of the spatial
and temporal variations of the observed field (Eq. (4)) quite well, we note that the peak
simulated precipitation is often weaker than observed. This is due to the fact that observed
precipitation has a large scatter at high SST (Fig. 1) and our parametrization (Eq. (5)) tries
only to model the mean.
(v) Variance and correlation maps. To provide a more objective measure of the per-
formance of our simple model, the structure of the zonal wind variance of the observed
and model winds and the correlation map of the model versus the observed zonal wind
anomalies are shown in Fig. lO(a) to (c). The observed zonal wind variance (Fig. lO(a))
shows the following features. The variance is strongly confined to the equatorial western
and central Pacific i.e., between 10"N and 10"S, 140"E and 140"W. It is also noted that the
eastern Pacific shows weak variance. The model zonal wind variance (Fig. 10(b)) shows
similar features except that the region of maximum variance is shifted by about 20" to the
west of the observed maximum.
Figure 1O(c) shows the correlation between the observed and zonal winds at each
gridpoint. Good correlation between model and observations is found over the equatorial
Pacific with the best correlation over the western Pacific. It is encouraging to note that
the correlation is good (0.6 to 0.8) over the region where the observed winds show strong
variability. The negative correlation coefficients around 140"W and 15"s are somewhat
disconcerting. This is due to the linear-model response to heating anomalies which produce
spurious easterlies in that region.
40 N. H. SAJI and B. N. GOSWAMI
30N
15N
EQ
15s
30s
Z0NP.L W I ND ANOMAL I ES
Var i Once Observed a
J \I
140E 180 140w 1 oow
MERl Dl ONAL WIND ANOMALIES
Var i ance Observed d
30N
15N
EQ
15s
30s
14OE 180 140w 1 oow
Figure 10. Maps of (a) observed zonal wind variance, @) model zonal wind variance and (c) correlation between
observed and model zonal wind anomalies; (d) to ( f ) are the same as for (a) to (c), but for the meridional wind
anomaly field.
In Fig. 10(d) to (f ), the variance of the observed and simulated meridional winds
and the correlation between them are shown. The model clearly fails to simulate the large
variance of the meridional winds in the eastern Pacific ITCZ region owing to the north-
south migration of the ITCZ; but it simulates the variance in the SPCZ region quite well.
Although the amplitude of the meridional wind anomalies in the ITCZ region is weak, the
correlation between the observed and the simulated anomalies is good even in the ITCZ
region. This means that the model simulates the nature of the meridional wind variability
well but with a reduced amplitude.
Thus, even though the model simulates the large-scale component quite well (as
seen in the previous subsections), certain systematic errors are evident when compared
on a point-by-point basis. Our parametrization for convection is designed to capture only
the large-scale part of the heating. As convection has a large variability in the warm
SST regions, the model parametrization is unable to capture it, and we expect the model's
simulated winds to have large errors in the regions climatologically covered by the warmest
waters. Closer examination of the systematic errors indeed shows that the largest errors
occur in those regions of the western and eastern Pacific normally covered by the warmest
TROPICAL LINEAR MODEL
LN QEOFl 55.35% a
LN QEOF2 13.30% b
30N
15N
EQ
15s
30s
1 i OE 180 l4OW 1 ow
C
41
1i 76 1976 1960 1962 1964 1966 1986 1990
-20 J
Figure 11. The first two EOFs and the corresponding PCs of the filtered LN forcing anomalies. PCl is the solid
line, PC2 is the dashed line.
waters (e.g. > 28 "C). Thus most of the systematic errors of the model are likely to be due
to the inability of the model to simulate the small-scale fluctuations in the precipitation
forcing.
( b ) Gill versus LN contributions
(i) Gill versus LN forcings: large-scale features. Here we present large-scale features
of the variability exhibited by the Gill and LN forcing functions. The region of maximum
variability for the Gill forcing EOFl (see Fig. 4) is over the central Pacific around 165"W.
42 N. H. SAJI and B. N. GOSWAMI
072
C
1
30 -
15 -
-151
I
1976 1978 1980 1982 1984 1986 1988 1990
-30'
Figure 12. The first two EOFs and the corresponding PCs of the filtered Gill vector wind anomalies. PC1 is the
solid line, PC2 is the dashed line.
The second EOF displays a dipole structure, and the first two PCs are phase-shifted with
respect to each other. Thus when the first two EOFs are considered in tandem with the
corresponding PCs, the zonally propagating nature of the Gill forcing is clearly brought
out.
The large-scale features of the LN forcing variability, on the other hand, are differ-
ent. First of all, the maximum variability for EOFl is over the equatorial eastern Pacific
(Fig. 11). This is to be expected since the SST anomalies also exhibit maximum
variability in this region. The LN EOF2 also displays a dipole structure, but with the
TROPICAL LINEAR MODEL
43
.............
30sJ i i o E ' 180 14OW 1 oow
072
LN EOF 2 15.84% b
30N
15N
EQ
15s
. . . . . ! ...................
30SJ '.
140E 180 140W 1 oow
072
C
30 -
15-
A /
----.- ---._--
-.-_
-- ---- v.*/
-15-
-30 -
Figure 13.
1976 1978 1980 1982 1984 1986 1988 1990
Same as Fig. 12, but for the filtered LN vector wind anomalies.
negative anomalies spread out more in the meridional direction than in the case of the
Gill EOF2. It is also seen that the largest negative anomaly in the case of LN EOF2 is
around 140"W and 20"N, whereas for the Gill EOF2 it is around 160"E and just north of
the equator.
Thus we find that the major difference between the Gill and the LN forcing anomalies
lies in their spatial pattern of variability. The Gill forcing anomalies are found to dominate
the variability in the western and central Pacific whereas the LN forcing is dominant over
the eastern Pacific.
44 N. H. SAJI and B. N. GOSWAMI
(ii) Gill versus LN winds: large-scale features. Figure 12 shows the first two EOFs and
the corresponding PCs of the vector wind anomalies simulated in response to the Gill
forcing anomalies, and Fig. 13 is for the LN vector wind anomalies. It is seen that EOFl
of both the Gill and LN winds explains most of the variability. The second EOFs explain,
typically, about one third of what was explained by the first EOFs. The magnitudes of both
the LN and Gill PCls are almost the same, but the LN EOFl loadings are about half that
of the Gill EOFl loadings. This is reflected in the Gill + LN EOF1, which is found to
resemble the Gill EOFl closely, indicating the predominance of the large-scale convective
heating in driving the large-scale surface winds. However we find that the inclusion of the
LN forcing improves the simulation of the surface winds in the eastern half of the model
domain. It can be seen that, in the equatorial region east of 160W, easterlies (westerlies) are
found in the Gill (LN) EOFl vector winds. This results in the Gill + LN EOFl exhibiting
weaker easterlies in this region than would have been the case if the Gill component alone
had been used in the simulation.
It is also noted that the centres of the divergence and convergence zones, as well as
the region of maximum variability of the Gill wind EOFs, are located to the west of those
of the LN wind EOFs. For example, the region of maximum variability and also the region
of anomalous convergence in the Gill EOFl pattern are found about 20" to the west of
these same features in the LN EOFl pattern. The same is true for the second EOFs. An
inspection of the first and second PCs of the winds forced by the Gill and LN components
reveals that the patterns in both cases seem to evolve in a similar manner. The eastward
migration of the westerly anomalies during a warm phase (e.g. during 1982-83) is seen in
both cases. However, we note that the westerlies, in response to LN forcing, peak about
three to four months later than in the case with the Gill forcing.
Thus we find that convective heating has the predominant role, compared with SST-
gradient-induced effects, in forcing anomalous winds in the tropical Pacific. Nevertheless,
the inclusion of the LN forcing was found to improve the simulation over the eastern
Pacific in a subtle manner. Differences in the spatial pattern and in the temporal evolution
were also noted between the Gill and the LN winds.
(iii) Simulation of specific events: role of Gill versus LN forcings. So far we have shown
only the simulation of the large-scale low-frequency patterns of the surface winds by our
model. While the model simulates the large-scale part of the surface wind variability quite
well, we have noted that it is unable to simulate the small-scale part of the surface wind
variability, which explains a smaller (but not negligible) fraction of the total surface wind
variability. Therefore, it is expected that on a month to month basis the simulation of the
surface winds by the model may sometimes differ from the observations. Nevertheless,
it is instructive to examine how the model performs when simulating the surface winds
during some special events such as the extreme warm or cold phases of the ENSO. Here
we show two such examples. Till now we have presented results with filtered (9-month
running-mean filter) data; now the results in this part will be based on unfiltered data.
Figure 14 shows the simulation of the surface winds by our model for J anuary 1983
(during a peak warm episode). Here, we also compare the individual contributions from the
Gill and LN forcings with the final simulation. The correspondence between the COADS
vector wind anomalies for this month and our simulated winds is remarkable. The maxi-
mum westerly anomalies are just south of the equator around 160"W as in the observations.
The convergence zone is also slightly south of the equator, around 130W, as in the ob-
servations. The equatorial easterlies in the western Pacific around 150"E are also well
simulated. Wenote that most of the contribution to the simulated winds comes from the
Gill forcing. The LN winds are found to be about half as strong as the Gill winds. It is
TROPICAL LINEAR MODEL 45
30N
15N
EQ
15s
30s
OBSERVED WINDS Jan 83 OBSERVED PREC. Jan 83
14OE 180 1 bow 1 oow
..........................
180 140W 1 OOW
30s
' 1 i OE '
-
10
GI LL WINDS Jan 83
,,,,,.. ,
...... ....,,,,,,,,,,, ...
140W 1 OOW
GI LL + LN FORCING Jan 83
30N
15N
EQ
15s
30s
MODEL PREC. Jan 83
I
140E 180 140W 1 OOW
-
10
LN WINDS Jan 83 SST Jan 83
30N
15N
EQ
15s
30s
140E 180 140W 1 OOW 140E 180 140W 1 OOW
--*
10
Figure 14. Simulation of the vector winds of J anuary 1983 by the Gill +LN model, the Gill model and the
LN model, along with their respective forcing fields. For comparison the observed precipitation and vector wind
anomalies are shown in the top panel. Unfiltered data is used. Precipitation is in mmd-', SST in deg C and winds
in m s-'.
46 N. H. SAJI and B. N. GOSWAMI
also noted that the LN winds are, in certain subtle ways, different from the Gill winds.
A notable difference is that the patch of anomalous westerlies extends further to the east
than for the Gill model. The easterly winds found to the east of the westerly wind patch
in the Gill winds is absent or weaker in the LN case. Apart from this the pattern of wind
anomalies is similar for both.
Similarly, Fig. 15 shows the simulation of the surface winds by our model for Decem-
ber 1988 (during a peak cold episode within the period of our study). The model does quite
well in the western half of the Pacific, i.e. west of 170"W. However, the observed equatorial
divergence region located around 130"W is simulated by the model to be around 160"W.
This results in stronger simulated equatorial westerlies in the eastern Pacific. Again there
are some subtle differences in the simulation by the two components. The anomalous di-
vergence zone in the LN model occurs around 130"W, whereas for the Gill model it occurs
around 170"W. The effect of this in the full model is to place the divergence zone midway
between the LN and Gill divergence zones.
Thus in general it is found that both the Gill and LN simulations are similar, with
the LN winds about half as strong as the Gill winds. It should also be noted that the
Gill forcing is quite different from the SST anomaly field which forces the LN winds.
In both of the cases presented above, the model precipitation fields have the following
two features: they are displaced to the west of the SST anomaly field; it is also found
that their structure resembles the observed precipitation field closely. It is interesting to
compare the two cases presented above with similar results presented by Li and Wang
(1994) (see their Figs. 12 and 13) who used a more realistic treatment of precipitation in a
simple linear model with an explicit boundary layer. Our model's results for J anuary 1983
compare better with the observations than do those by their model, while both results are
comparable for December 1988. At least, the comparison bears testimony to the soundness
of the empirical relationship for precipitation that we have employed.
In the previous two case studies of the anomalous wind field, it was found that the LN
model winds were weaker than the Gill winds. To see if this is true also for the climatological
winds, we undertook another run in which the model was forced by climatological SSTs and
precipitation. The results for J anuary and J uly are shown in Figs. 16 and 17, respectively.
It is seen that the model performs quite well in general. However serious discrepancies
are noted in J anuary over the south Pacific between 150" W and 110"W. Also during J uly
the model simulates 'spurious' westerlies in the north-east Pacific. In both the J anuary
and July simulations we note that the simulated eastern Pacific ITCZ is too diffuse. This
can be related to the fact that the simulated precipitation has a much larger meridional
scale compared to that of the observed precipitation field. In spite of the above mentioned
deficiencies, the model's results bring out the remarkable difference in the simulation
by the Gill and LN models. Unlike the case of the anomalous winds, where both models
exhibited more or less the same features, the mean field simulations are strikingly different.
It is seen that the LN model simulates most of the features of the observations with the
Gill model, supplementing itin places. Thus it is seen that the inclusion of LN forcing is
essential for the simulation of the mean winds.
5. CONCLUDING REMARKS
With the objective of providing an improved atmospheric model for coupled exper-
iments, we have presented here a new improved linear model for the simulation of the
tropical surface winds. This model falls in the large class of linear models used in earlier
studies (Gill 1980; Zebiak 1986; Neelin 1989; Davey and Gill 1987; Lau and Shen 1988:
TROPICAL LINEAR MODEL
15N
-.
47
. ( . . . . . . . . . . . . . . . . . . .
;... . . . I. . I , , . * # ' 1 ' 1 ' . , ..........
- - ~ - , , l , l l f , , , , , . . . , , . ~ . ,
. , . . . . . . . ; . . . :.<,.;, .. .!
- - - , , \ \ \ L l I I I , , #, .............
OBSERVED WINDS Dec 8 8
140E 180 140W 1 OOW
30N
15N
EQ
15s
30s
-
6
GI L L + LN WINDS Dec 88
14OE 180 1 dow 1 oow
30N
15N
EQ
15s
LN WINDS Dec 88
. ,
.... \...\\......
. .
. . . . . . . - ......,,..,,.
. . . . . . . . . . . . . . . . . . (. . . . . .
30s' ' '.
140E 180 140W 1 OOW
OBSERVED PREC. Dec 88
SON
15N
EQ
15s
30s
GI L L + L N FORCING Dec 88
1 dOE 180 1 dow 1 dow
MODEL PREC. Dec 88
140E 180 140W 1 OOW
SST Dec 88
140E 180 140W l OOW
Figure 15. Same as in Fig. 14, but for December 1988.
48
1 5 ~ .
N. H. SAJI and B. N. GOSWAMI
............. . . {. . . . . . . . .
J 3 ; ; . : . ; . ; ; l.;.:.:. 1 J ; .:.:.:. 1 ; .: .:.:.:.
, , , , . C C I C C ~ C r r - - . . . . . . . .
I,,,,,,,.., C C C C ~ " . . . . . . . . . . .
30N
15N
EQ
15s
30s
OBSERVED WINDS J an
14OE 1 a0 1 bow 1 ow
---t
10
GI L L + LN WINDS J an
d
10
LN WINDS J an
EQ
15s
30s
I ; y. . < ...................
........................
140E %iow row
OBSERVED PREC. J an
30N
15N
EQ
15s
30s
GI L L + LN FORCING Jon
- / zp=$=- q . . . . . . . . . . . . . . . . . . . .
, . . . . . . . . . . . . .y.. . . . . .; . . .T
' .
. .: ,436:. ... . . . . . . . . . . :+
.-<5
. . . . .
/ - 5 0 ~ J /1
-
140E 180 140W 1 OOW
MODEL PREC. J an
SST J an
30N
15N
EQ
15s
30s
Figure 16. Same as in Fig. 14, but for the January mean conditions.
TROPICAL LINEAR MODEL
49
OBSERVED WINDS J u I
""" l40E 180 140w 1 oow
10
d
10
GI L L WINDS
...............
I , \ . . A ........
J u I
I , , , I l l ,
i '\ %. i i i \ '\.
........
........
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .
1
180 140W 1 OOW
30s ' l d
-
10
LN WINDS J u I
d
10
OBSERVED PREC. J u I
30N
15N
0
15s
30s
GI L L + L N FORCING J u I
MODEL PREC. J u I
30N
15N
EQ
15s
I
I
180 140W 1 OOW
SST J u I
Figure 17. Same as in Fig. 14, but for the July mean conditions.
50 N. H. SAJI and B. N. GOSWAMI
Seager 1991; Lindzen and Nigam 1987; Kleeman 1991) but with two important and sig-
nificant improvements. Wepropose that the surface winds are driven both by low-level
convergence associated with deep convection (Gill forcing) and by surface pressure gra-
dients associated with SST gradients (LN forcing). In all earlier studies, either one or the
other of these physical mechanisms was included. We emphasize that, depending on the
mean conditions, whereas one forcing may be important in one geographical region, the
other forcing may be important in another location. For example, in the western Pacific,
where the mean SST is high but the SST gradients are weak, the Gill forcing may be
dominant, but in the eastern Pacific, where the mean SST is low but gradients are sharp,
the LN forcing may be strong (Gutzler and Wood 1990). The LN model and the Gill model
being equivalent (Neelin 1989), we have simply combined the two effects in one model,
as was done by Eltahir and Bras (1993) in their model for the atmospheric response to
Amazon deforestation.
The second, and more important, improvement we have made is in the parametrization
of the organized part of the convective heating. It was recognized that linear models could
simulate the surface winds well if the atmospheric heating could be parametrized correctly
(Zebiak 1990). Earlier linear models such as the one used by Zebiak (1986) had serious
problems including that of spurious easterlies in the eastern Pacific. This was due to
the fact that organized convective heating is a strong nonlinear function of the mean SST.
Kleeman (1991) also recognized the importance of the mean SST in the calculation of the
heating anomalies. However, he had to introduce a threshold in the moist static energy
(MSE) for deep convection in an ad hoc manner. Owing to the intrinsic nonlinearity in
the formation of deep convection and its organization into large scales, the inclusion of a
simple moisture equation is insufficient for simulating the heating anomalies correctly (e.g.
Seager 1991). Here we have adopted an empirical parametrization of the total convection
as a nonlinear function of the total SST. This has been possible owing to some recent
studies (Fu et al. 1990; Waliser et al. 1993; Zhang 1993) establishing a clear, unambiguous
relationship between SST and intensity and frequency of occurrence of organized deep
convection.
With these two improvements added to the earlier similar models, we give below the
main conclusions of the study.
(i) The parametrization is successful in simulating the large-scale part of the precipitation
anomalies (e.g. the first two EOFs) well. This results in good simulation of the large-
scale part of the surface winds (the first two EOFs). The location of the centres and
the intensity of these patterns are simulated so well that they are comparable to those
simulated by many low-horizontal-resolution GCMs. Most low-resolution GCMs do
not simulate the location of the equatorial westerly maximum associated with EOFl
correctly. This is related to the GCMs inability to simulate the precipitation patterns
correctly in their location and intensity. In that respect, our parametrization does quite
well.
(ii) The eastward migration of precipitation and equatorial westerly anomalies during the
evolution of a warm episode is simulated quite well.
(iii) The Gill and LN forcing fields, also the winds, exhibit different behaviour. We find
that the Gill forcing (response) is dominant over the western and central Pacific and
that the LN forcing (response) is dominant over the eastern Pacific.
(iv) Simulation of equatorial anomalies are close to observations even on a monthly scale
(Fig. 14), especially during warm episodes. During cold episodes, whereas the large
part of the wind anomalies are again simulated well, the model is unable to simulate
some smaller-scale aspects. Noteworthy, in particular, is the reduction of the spurious
TROPICAL LINEAR MODEL 51
easterlies in the eastern Pacific in our simulations and the good comparison with the
results of Li and Wang (1994).
(v) It is found that, with regard to the anomalous wind simulation, the Gill forcing pre-
dominates compared with the LN forcing. The LN winds are in general weaker than
the Gill winds by about a factor of two. On the other hand, it is found that for the
mean wind simulation, the LN forcing has a more important role. It is found that the
inclusion of LN forcing is critical for simulating the mean winds in the tropics.
What we have done in our parametrization of atmospheric heating is to model the observed
curves showing increase of precipitation with SST. We note that these curves are not
identical in all geographical locations. Therefore, the empirical formula that we have used
here may not be universally applicable in all geographical locations. Some generalization
of the formula in this regard will be desirable.
There is another more difficult problem associated with modelling the increase in
convection with SST. Although it is now established that there is a dramatic increase in
both intensity and frequency of occurrence of deep convection with higher SST (> 27 "C),
the variability of deep convection also increases with SST in this region (e.g. Fig, 16 of
Waliser et al. 1993). This means that for a given high SST (say 28 "C) frequent intense
deep convection may also alternate with clear sky conditions. Our parametrization, though
quite successful in simulating both the intensity and distribution of the mean convective
heating, does not contain any information on the increase in the variability of convection
with increase in SST. Since the large-scale low-frequency part of the surface winds is
forced by the sustained mean convective heating, our model successfully simulates this
part of the surface wind field. However, on a month to month basis the model does not do
as well in simulating individual monthly mean anomalies. This is because the model does
not simulate the high-frequency part of the convective heating and this high-frequency part
influences the monthly means to some extent. This is probably the main reason why most
GCMs also have significant errors in simulating monthly mean precipitation.
The simplicity of our parametrization and its remarkable success in simulating the
low-frequency large-scale part of the surface winds makes it an ideal candidate as an
atmospheric component in a coupled model.
ACKNOWLEDGEMENTS
The authors are grateful to the Department of Science and Technology, Government
of India for partial support for this work. The authors are also grateful to the two anony-
mous reviewers whose critical comments on a previous version of the manuscript led to
significant improvements. Weare also grateful to Dr D. Sengupta and Dr H. Annamalai
for some very constructive comments and to Mrs R. Rama for preparing the manuscript.
REFERENCES
Bjerknes, J. 1969 Atmospheric teleconnections from the equatorial Pacific. Mon.
Davey, M. K. and Gill, A. E. Experiments on tropical circulation with a simple moist model. Q.
Eltahir, E. A. B. and Bras, R. L. On the response of tropical atmosphere to large-scale deforesta-
Fu, R., DelGenio, A. D. and Behaviour of deep convective clouds in the tropical Pacific deduced
Gadgil, S., Joseph, P. V. and Ocean-atmosphere coupling over monsoon regions. Nature, 312,
Weather Rev., 97,163-172
J. R. Meteorol. SOC., 113,1237-1269
tion. Q. J. R. Meteorol. Soc., 119,779-793
from ISCCP radiances.J. Clim., 3,1129-1152
1987
1993
1990
1984
Rossow, W. B.
Joshi. N. V. 141-143
52 N. H. SAJI and B. N. GOSWAMI
Gill, A. E.
Goswami, B. N. and Shukla, J .
Graham, N. E. and Barnett, T. P.
Graham, N. E., Barnett, T. P.,
Chevin, R. M.,
Schlesinger, M. E. and
Schese, U.
Gruber, A. and Winston, J . S.
Gutzler, D. S. and Wood, T. M.
Hirst, A. C.
Kitoh, A.
Kleeman, R.
Latif, M., Biercamp, J.,
Storch, H. V., McPhaden, M.
and Kirk, E.
Latif, M., Maier-Reimer, E. and
J unge, M. M.
Lau, K. M. and Shen, S.
Lau, N. C., Philander, S. G. H. and
Nath, M. J .
Li, T. and Wang, B.
Lindzen, R. S. and Nigam, S.
Murphree, T. and van den Dool, H.
Nagai, T., Takioka, T., Endoh, M.
Neelin, J . D.
and Kitamura, Y.
Neelin, J. D. and Held, I. M.
Neelin, J. D., Latif, M.,
Allbaart, M. A. F.,
Cane, M. A., Cubasch, U.,
Gates, W. L., Gent, P. R.,
Ghil, M., Gordon, C.,
Lau, N. C., Mechoso, C. R.,
Meehl, G. A.,
Oberhuber, J . M.,
Philander, S. G. H.,
Schopf, P. S., Sperber, K. R.,
Sterl, A., Takioka, T.,
Tribbia, J. and Zebiak, S. E.
Philander, S. G. H.,
Pacanowski, R. C., Lau, N. C.
and Nath, M. J .
1980
1991
1987
1989
1978
1990
1986
1991
1991
1990
1993
1988
1992
1994
1987
1988
1992
1988
1989
1990
1987
1992
1992
Some simple solutions for heat-induced tropical circulations. Q. J.
R. Meteorol. Soc., 106,447462
Predictability of a coupled ocean-atmosphere model. J. Clim., 4,
3-22
Sea surface temperature, surface wind divergence and convecrion
over tropical oceans. Science, 238,657-659
Comparison of GCM and observed surface winds over the tropical
Indian and Pacific Oceans. J. Armos. Sci., 46,760-788
Earth-atmosphere radiative heating based on NOAA scanning ra-
diometer measurements. Bull. Am. Meteorol. Soc., 59,1570-
1573
Structure of large-scale convective anomalies over tropical oceans.
J. Clim., 3,483-496
Unstable and damped equatorial modes in simple coupled ocean-
atmosphere models. J. Atmos. Sci., 43,606-630
Interannual variations in an atmospheric GCM forced by the 1970-
1989 SST. Part I: Response of the tropical atmosphere. J.
Meteorol. SOC. Jpn., 69,251-269
A simple model of atmospheric response to ENSO sea surface
temperature anomalies. J. Atmos. Sci., 48,3-18
Simulation of ENSO related surface wind anomalies with an
atmospheric GCM forced by observed SST. J. Clim., 3,509-
521
Climate variability in a coupled GCM 1: The tropical Pacific. J.
Clim., 6,5-21
On the dynamics of intraseasonal oscillations and ENS0.J . Armos.
Sci., 45,1781-1797
Simulation of ENSO-like phenomenon with a low-resolution cou-
pled GCM of the global ocean and atmosphere. J. Clim., 5,
284-307
A thermodynamic equilibrium climate model for monthly mean
surface winds and precipitation over the tropical Pacific. J.
Atmos. Sci., 51,1372-1385
On the role of sea surface temperature gradients in forcing low-
level winds in the tropics. J. Atmos. Sci., 44,2440-2458
Calculating winds from time mean sea level pressure fields. J.
Atmos. Sci., 45,3269-3281
El Niiio-southern oscillation simulated in an MRI atmosphere-
ocean coupled general circulation model. J. Clim., 5, 1202-
1233
A simple model for surface stress and low level flow in the tropical
atmosphere driven by prescribed heating. Q. J. R. Meteorol.
On the interprtation of the Gill model. J . Atmos. Sci., 46, 2466-
A hybrid coupled general circulation model for El Nirio studies. J.
Modelling tropical convergence based on the moist static energy
Tropical air-sea interactions in general circulation models. Clim.
SOC., 114,747-770
2468
Atmos. Sci., 41,614-693
budget. Mon. Weather Rev., 115,3-12
Dyn., 7,73-104
Simulation of ENSO with a global atmospheric GCM coupled to
a high-resolution tropical Pacific Ocean GCM. J. Clim., 5,
308-329
TROPICAL LINEAR MODEL 53
Reynolds, R. W. 1988
Reynolds, R. W. and Marisco, D. C. 1993
Seager, R. 1991
Slutz, R. J., Lubkar, S. J ., 1985
Hiscox, J. D., Wood~ff, S. D.,
J enne, R. L., J oseph, D. H.,
Steurer, P. M. and Elms, J . D.
Wang, B. and Li, T.
Walker, D., Graham, N. E. and
1993
1993
Gautier, C.
Yoo, J . M. and Carton, J . M. 1988
Zebiak, S. E. 1982
1986
1990
Zebiak, S. E. and Cane, M. A. 1987
Zhang Chidong 1993
A real-time global sea surface temperature analysis. J. Cfim., 1,
75-86
An improved real-time global sea surface temperature analysis. J.
Clim., 6,114119
A simple model of climatology and variability of the low-level
wind field in the tropics. J. Cfim., 4,164-179
Comprehensive ocean-atmosphere data set: Release 1 . Cfimate
Research Program, NOAA Environmental Research Labora-
tories, Boulder CO
A simple tropical atmosphere model of relevance to short-term
climate variations. J. Atmos. Sci., 50,260-284
Comparison of highly reflective cloud and outgoing longwave radi-
ation data sets for use in estimating tropical deep convection.
J. Clim., 6,331-353
Spatial dependence of the relationship between rainfall and out-
going longwave radiation in the tropical Atlantic. J. Cfim., 1,
A simple atmospheric model of relevance to El Nhio. J. Atmos.
Atmospheric convergence feedback in a simple model for El Niio.
Diagnostic study of Pacific surface winds. J. Clim., 3,1016-1031
A model El Nhio-southern oscillation. Mon. Weather Rev., 115,
Large-scale variability of atmospheric deep convection in relation
to sea surface temperature in the tropics. J. Cfim., 6, 1898-
1913
1047-1054
Sci,, 39,2017-2027
Mon. WeatherRev., 114,1263-1271
2262-2278

Das könnte Ihnen auch gefallen