3 Results
a. Mean circulation and
precipitation
Figure 1 shows the zonal mean meridional overturning circulation in the
Control and ClimRad runs. Overall the difference between the two
simulations is very small, indicating that the mean circulation remains
almost unchanged without radiative interactions. Also, we note that
suppressing radiative interactions has little impact on the magnitude of
the global-mean precipitation because the overwriting approach applied
in the ClimRad run does not change the magnitude of the global-mean
radiative cooling of the atmosphere. Overall the atmospheric energy
budget remains nearly unaffected in the ClimRad run.
b. Convective aggregation, cloud and relative
humidity
Although the mean circulation is essentially the same between the
Control and ClimRad runs, suppressing radiative interactions
significantly changes features related to synoptic-scale convection.
Figure 2 shows probability density functions (PDFs) of daily subsidence
fraction in the Control and ClimRad runs computed over the entire
tropics (30°S–30°N). Small (large) values of daily subsidence fraction
in the ClimRad run are more (less) frequent than those in the Control
run, indicating that convection becomes less aggregated when radiative
interactions are suppressed. These results are in line with previous
numerical simulations with an aquaplanet configuration (Coppin & Bony,
2015) and in convection-resolving models (Muller & Bony, 2015; Muller
& Held, 2012; Yang, 2018).
Using satellite observations, Bony et al. (2020) found that the spatial
organization of deep convection can modulate high-level clouds and
relative humidity in the free troposphere, which further impacts the
tropical radiation budget. Based on an ensemble of radiative-convective
equilibrium simulations, Wing et al. (2020) showed similar results in
which the occurrence of convective self-aggregation reduces high cloud
coverage and dries the mid-troposphere. Here, we investigate how clouds
and relative humidity respond to suppressed radiative interactions
(Figure 3). Negative values are found in the upper troposphere over the
tropics for the Control run minus the ClimRad run, indicating that a
more aggregated state is associated with fewer high-level clouds. Also,
we find that the free troposphere is in general drier in the Control run
than that in the ClimRad run (Figure 4). The reduction in high-level
clouds and mid-tropospheric relative humidity in the Control run is
qualitatively consistent with observations (Bony et al., 2020; Holloway
et al., 2017; Stein et al., 2017; Tobin et al., 2013; Tobin et al.,
2012) and other model simulations (Bretherton et al., 2005; Wing et al.,
2017; Wing & Emanuel, 2014; Wing et al., 2020). This shows that, even
when the large-scale circulations are nearly identical, differences in
the spatial organization of convection can alter the mean state of the
free troposphere.
However, the responses in cloud fraction and relative humidity are not
vertically uniform. We find that the Control run has an overall increase
in cloud fraction and relative humidity in the boundary layer, which may
not be directly linked with the degree of aggregation. Previous results
on the relationship between convective aggregation and low cloud
fraction are inconclusive. While an increase in low-level clouds with
aggregation is found in Tobin et al. (2013) and Stein et al. (2017),
Tobin et al. (2012) found the opposite result. Recently, Wing et al.
(2020) found that most radiative-convective equilibrium simulations
agree on an increase in low-level clouds with convective
self-aggregation, although such increase is less robust in magnitude.
They proposed that the difference in horizontal grid spacing, rather
than the occurrence of self-aggregation itself, may contribute to the
increase in low-level clouds. In addition, lower tropospheric stability
may also play a role in modulating low-level clouds (Bony et al., 2020).
When lower-tropospheric stability increases, more moisture is trapped in
the boundary layer, which promotes the formation of low-level clouds
(Wood & Bretherton, 2006). However, the impact of lower tropospheric
stability on low-level clouds is thought to be independent of the degree
of aggregation as noted by Bony et al. (2020). Here we use estimated
inversion strength (EIS ), defined as\(EIS=LTS-\Gamma_{m}^{850}\left(z_{700}-LCL\right)\), to
represent the stability in the boundary layer. LTS stands for
lower tropospheric stability and is computed as\(LTS=\theta_{700}-\theta_{1000}\) where \(\theta_{700}\) and\(\theta_{1000}\) are potential temperatures at 700 hPa and 1000 hPa
respectively (Klein & Hartmann, 1993); \(\Gamma_{m}^{850}\) is
Among different factors contributing to the low cloud fraction,
radiative interactions with boundary layer clouds could make a
difference. In the boundary layer, the coupling between clouds,
radiation, turbulence and entrainment was first documented by Lilly
(1968). Strong longwave radiative cooling at the cloud top promotes
vertical mixing and drives turbulent eddies, which transports moisture
from the sea surface upward and maintains the cloud amount (Bretherton
et al., 2004b; Wood, 2012). Additionally, strong radiative cooling at
the cloud top increases relative humidity in the boundary layer. Higher
relative humidity further promotes the formation of low-level clouds
(Brient & Bony, 2012). In the ClimRad run, the coupling between
radiation and low-level clouds is disabled, which may explain the
reduction in both relative humidity and cloud fraction in the boundary
layer.
To explore the sensitivity of clouds and relative humidity to the
vertical structure of radiative coupling, we conduct two other
simulations: one is referred to as ClimRadFT, in which the overwriting
procedure is only applied to the free troposphere, whereas radiation in
the boundary layer is fully interactive; the other is referred to as
ClimRadBL, in which only radiation in the boundary layer is fixed while
that in the free troposphere is interactive (see Table 1 for more
details). Compared to the Control run, we find that changes in low-level
clouds and lower-tropospheric relative humidity in the ClimRadBL run are
similar to those in the ClimRad runs
(Figure
3 and Figure 4). In contrast, the differences in tropical low-level
clouds and relative humidity in the boundary layer are reduced between
the Control and ClimRadFT runs (Figure 3 and Figure 4). However, we note
that the ClimRadBL and ClimRadFT runs exhibit similar changes in lower
tropospheric stability compared to the Control run (Figure S1). This
suggests that it is the direct effects of radiative coupling in the
boundary layer that is primarily responsible for the changes in
low-level clouds and relative humidity in the ClimRad run, rather than
the influence of radiative processes in the free troposphere on lower
tropospheric stability.
In addition, the changes in low-level clouds and relative humidity are
found to be, at least from a qualitative perspective, independent of
variations in the degree of aggregation. A comparison of the PDFs of
daily subsidence fraction from the Control, ClimRad, ClimRadFT and
ClimRadBL runs is shown in Figure 5. Compared to the Control run, the
other three simulations exhibit an overall reduction in the degree of
aggregation although the magnitude of such reduction varies among them.
The qualitatively consistent change in the degree of aggregation cannot
explain the differing responses in cloud fraction and relative humidity,
indicating that variations in the degree of aggregation may not be a
leading factor in modulating the distribution of cloud and humidity.
Here, the coupling between radiation, cloud and humidity plays a more
important role in maintaining the model’s mean state.
c. Response in extreme
precipitation
Previous idealized modeling studies showed that extreme daily
precipitation becomes weaker when convective aggregation is inhibited
(Bao & Sherwood, 2019). To examine the response in extreme
precipitation to suppressed radiative interactions, we compute the
annual maximum daily precipitation (\(P_{e}\)) at each grid point. While
the difference in \(P_{e}\) between the Control and ClimRad runs is
small at middle-to-high latitudes, a significant reduction in \(P_{e}\)is found across the tropics in the ClimRad run (Figure 6, left), which
indicates that suppressing radiative interactions reduces the strength
of extreme daily precipitation. At each grid point, \(P_{e}\) can be
estimated by a physical scaling diagnostic (O’Gorman & Schneider, 2009;
Pfahl et al., 2017; Sugiyama et al., 2010):
\begin{equation}
\begin{matrix}P_{e}\sim-\left\{\omega_{e}\left.\ \frac{dq_{s}}{\text{dp}}\right|_{\theta^{*}}\right\}\#\left(1\right)\\
\end{matrix}\nonumber \\
\end{equation}where \(\omega_{e}\) is the annual maximum daily vertical pressure
velocity, \(q_{s}\) is the saturation specific humidity, \(p\) is the
pressure and \(\theta^{*}\) is the saturation equivalent potential
temperature. Here \(\left\{\bullet\right\}\) means a mass-weighted
vertical integral over the troposphere. We show that the scaling
approach reproduces the spatial patterns of \(P_{e}\) in both
simulations, leading to a consistent reduction in the scaling when
radiative interactions are suppressed (Figure 6, right).
Eq. 1 can be used to decompose changes in extreme precipitation into
thermodynamic and dynamic contributions. A thermodynamic scaling is
implemented by replacing \(\omega_{e}\) in Eq. 1 with long-term averaged
vertical velocity at each grid point, whereas a dynamic scaling is the
difference between the full scaling and the thermodynamic scaling (Pfahl
et al., 2017). There is little difference in the thermodynamic
contribution between the Control and ClimRad runs (Figure S2, left)
because both runs are forced by the same SSTs and CO2concentrations. However, the spatial patterns of difference in dynamic
contribution (Figure S2, right) largely resemble the spatial patterns of
difference in \(P_{e}\) and the scaling, indicating that suppressing
radiative interactions primarily reduces the dynamical contribution to
extreme precipitation.
To verify the robustness of our results, probability distributions of
daily precipitation and updrafts across the tropics (30°S–30°N) are
compared between these two simulations. Figure S3 shows the base-10
logarithm of the probability that daily precipitation and
mid-tropospheric updrafts (\(\omega_{500}<0\)) exceed a particular
value in the Control and ClimRad runs. We find that both variables
exhibit a reduction in the probability of exceedance toward its extreme
values in the ClimRad run, indicating that suppressing radiative
interactions reduces the frequency of extreme convective events. We note
that suppressing radiative interactions also reduces the temporal
variance of daily precipitation (Figure S4).
Having demonstrated the impact of suppressing radiative interactions on
convective organization, we next explore the physical mechanisms which
underlie these changes. To do that we first divide the tropics into\(10\times 10\) regional blocks (Figure 7, top). Within each block, the
grid point with the local maximum precipitation is identified, which
later becomes the new center of that block. The recentered blocks are
then composited based on their domain mean precipitation. Here we show
composites of precipitation in the Control run for blocks with domain
mean precipitation <5, 5–10, 10–15 and >15 mm
day-1 (Figure 7, bottom). Note that composites of
precipitation in the ClimRad run show similar results (not shown).
However, the number of blocks per year (referred to as \(N_{b}\)) in
each bin is different between the Control and ClimRad runs. Boxplots of\(N_{b}\) normalized by the median value in the Control run are shown in
Figure 8 (top). In the >15 mm day-1 bin,\(N_{b}\) is reduced in the ClimRad run, which means that blocks with
heavy precipitation happen less frequently when radiative interactions
are suppressed. Through this block-by-block analysis, we can also
compare the degree of aggregation over blocks with similar amplitude of
domain mean precipitation. A comparison of PDFs of daily subsidence
fraction between the Control and ClimRad runs are shown in Figure 8
(bottom). Higher probabilities of large subsidence fraction are found in
the Control run, indicating that suppressing radiative interactions
leads to an overall reduction in aggregation across convective systems
of different intensities, which is consistent with the results shown in
Figure 2.
In idealized models, it is found that the upgradient transport of moist
static energy (Neelin & Held, 1987) plays an important role in
convective aggregation (Bretherton et al., 2005; Muller & Bony, 2015;
Muller & Held, 2012). Here, radiative cooling and circulation are
composited over bins as shown in Figure 8. Following Bretherton et al.
(2005), we use column relative humidity (CRH), defined as the ratio of
water vapor path to the saturation water vapor path of the atmospheric
column (Bretherton et al., 2004a; Raymond, 2000), to represent the
degree of dryness at each grid point within a block. Next, all grid
points in a block are sorted from lowest to highest CRH and the
circulation is represented by an effective streamfunction \(\Psi\),
which is computed as a horizontal integral over vertical velocity
starting with the driest grid point. The streamfunction \(\Psi\) at a
certain grid point can be interpreted as an accumulation of vertical
mass flux over grid points that are drier than the target grid point.
Primarily, the streamfunction is thought to capture the exchange of
moist static energy between dry and moist regions (Bretherton et al.,
2005).
Figure 9 shows the streamfunction \(\Psi\) and radiative cooling rates
in the Control and ClimRad runs. In the Control run, when the domain
mean precipitation is small, the circulation is weak and there is little
contrast in radiative cooling between dry and moist regions, especially
in the lower troposphere. As the domain mean precipitation increases,
the circulation gets stronger, with its low-level component below
~850 hPa moving air from dry to moist regions. Although
the magnitude of radiative cooling in dry regions does not change much,
the radiative cooling reduces significantly in moist regions as domain
mean precipitation increases, which is equivalent to adding anomalous
radiative heating there. As a result, the horizontal gradient of
radiative cooling is enhanced, which promotes the low-level circulation
and thus the upgradient transport of energy. In comparison, the enhanced
horizontal gradient of radiative cooling shown in the Control run is
missing in the ClimRad run, indicating that suppressing radiative
interactions inhibits the horizontal gradient of radiative cooling from
increasing, which explains why the degree of aggregation and extreme
precipitation events are reduced in the ClimRad run.
d. Meridional width of the tropical rain
belt
Recent studies measure the width of tropical ascending regions by the
fraction of vertical pressure velocity at 500 hPa less than zero in the
tropics (Su et al., 2020; Su et al., 2019). Given the same domain,
greater ascending fraction corresponds to smaller subsidence fraction.
While in section 3b we show that daily subsidence fraction in the
tropics is reduced without radiative interactions. On longer time
scales, the mean vertical pressure velocity at 500 hPa exhibits little
difference between the Control and ClimRad runs (not shown). However,
this definition may not be an appropriate measure of the meridional
width of zonal mean Hadley circulation or the width of the intertropical
convergence zone (ITCZ), as noted by Su et al. (2020). Therefore, other
metrics are required to quantify the width of the tropical rain belt.
Based on observations, Popp and Bony (2019) reported a strong link
between zonal convective clustering (CC) and the tropical rain belt:
when convection becomes more clustered in the zonal direction, the
meridional width of tropical rain belt increases and exhibits a
double-peak structure. However, it remains unclear how CC is related to
the width of ITCZ in climate models (Popp et al., 2020b). In section 3b,
we show that suppressing radiative interactions reduces the degree of
aggregation across the tropics. Thus, convection should become less
clustered in the zonal direction as well without radiative interactions.
Here, two metrics are used to characterize zonal CC: i) the
precipitation-inferred CC index, which is defined as monthly mean of the
meridionally averaged daily zonal standard deviation of precipitation
from 6°S to 6°N normalized by the daily mean precipitation over the same
region (Popp & Bony, 2019); and ii) the dynamically inferred CC index,
which is defined as the monthly average of the daily zonal fraction of
positive values of the meridional-mean vertical pressure velocity at 500
hPa between 6°S to 6°N (Popp et al., 2020a). Also, we only consider
months during which the tropical precipitation distribution is symmetric
about the equator with the tropical precipitation asymmetry index (Hwang
& Frierson, 2013; Popp & Bony, 2019) less than 0.4. Another two
metrics are used to quantify the ITCZ width: i) the
precipitation-inferred ITCZ width, which is defined as the area mean of
precipitation from 15°S to 15°N divided by the area mean of
precipitation from 6°S to 6°N (Popp & Bony, 2019); and ii) the
dynamically inferred ITCZ width, which is defined by the contiguous
width in degrees latitude of zonal mean ascent region at 500 hPa around
the absolute maximum of zonal mean precipitation (Byrne & Schneider,
2016; Popp & Bony, 2019).
Scatter plots of zonal CC and the ITCZ width in the Control and ClimRad
runs are shown in Figure 10. Positive temporal correlations are found
between zonal CC and the ITCZ width using either precipitation or
dynamically inferred metrics in both simulations, which is consistent
with observations (Popp & Bony, 2019). We note that the mean ITCZ width
exhibits little difference between the Control and ClimRad runs, which
is supported by Figure S5 and Figure S6 based on precipitation minus
evaporation. These results indicate that suppressing radiative
interactions has little impact on the mean ITCZ width. In comparison,
the mean value of zonal CC is reduced in the ClimRad run, which comes as
no surprise since the degree of aggregation is also reduced without
radiative interactions as illustrated in section 3b. Based on model
simulations participating in CMIP5 (Taylor et al., 2012), Popp et al.
(2020b) showed that biases in CC cannot explain biases in the ITCZ width
and no dominant mechanism could explain the link between the temporal
variability of CC and that of the ITCZ width. However, they found a
tendency for models with higher spatial resolution to exhibit stronger
links between zonal CC and the dynamically inferred ITCZ width. In this
study, suppressing radiative interactions has a robust impact on zonal
CC but little impact on the mean ITCZ width. One possibility is that
while the degree of aggregation/clustering is more sensitive to
synoptic-scale radiation-circulation coupling (i.e. the spatial contrast
in radiative cooling), the ITCZ width is more dependent on the long-term
averaged large-scale circulation in this GCM.