Stratospheric water vapour (SWV) is a climatically important atmospheric constituent due to its impacts on the radiation budget and atmospheric chemical composition. Despite the important role of SWV in the climate system, the processes controlling the distribution and variation in water vapour in the upper troposphere and lower stratosphere (UTLS) are not well understood. In order to better understand the mechanism of transport of water vapour through the tropopause, this study uses the high-resolution Global Environmental Multiscale model of the Environment and Climate Change Canada to simulate a lower stratosphere moistening event over North America. Satellite remote sensing and aircraft in situ observations are used to evaluate the quality of model simulation. The main focus of this study is to evaluate the processes that influence the lower stratosphere water vapour budget, particularly the direct water vapour transport and the moistening due to the ice sublimation. In the high-resolution simulations with horizontal grid spacing of less than 2.5 km, it is found that the main contribution to lower stratospheric moistening is the upward transport caused by the breaking of gravity waves. In contrast, for the lower-resolution simulation with horizontal grid spacing of 10 km, the lower stratospheric moistening is dominated by the sublimation of ice. In comparison with the aircraft in situ observations, the high-resolution simulations predict the water vapour content in the UTLS well, while the lower-resolution simulation overestimates the water vapour content. This overestimation is associated with the overly abundant ice in the UTLS along with a sublimation rate that is too high in the lower stratosphere. The results of this study affirm the strong influence of overshooting convection on the lower stratospheric water vapour and highlight the importance of both dynamics and microphysics in simulating the water vapour distribution in the UTLS region.
Stratospheric water vapour (SWV) strongly influences the Earth radiation
budget (IPCC, 2013) and stratospheric chemistry (e.g., Anderson et al.,
2012). Global climate models (GCMs) generally project an increase in SWV
during global warming, which may lead to cooling of the stratosphere and
further warming of the troposphere and surface (Forster and Shine, 1999,
2002; Solomon et al., 2010; Riese et al., 2012) and thus constitutes a
potentially important climate feedback mechanism (Dessler et al., 2013; Huang
2013; Huang et al., 2016; Banerjee et al., 2019). Dessler et al. (2013)
estimated the SWV feedback to be
Despite its importance, the processes that control the distribution and variation in water vapour in the upper troposphere and lower stratosphere (UTLS) are not well understood. Large discrepancies are found between the “A-Train” satellite observations and the GCMs of Phase 5 of the Coupled Model Inter-comparison Project (CMIP5) (Jiang, 2012). This study shows that the ratio of water vapour content in the GCMs to that from satellite observations can be as large as two to five in the mid-latitude UTLS region. Such discrepancies cast significant uncertainty in the SWV radiative feedback simulated by the GCMs (Huang et al., 2016). Global reanalyses also suffer from SWV biases, including the Modern-Era Retrospective Analysis for Research and Applications (MERRA), its newer release MERRA2 and the Interim Reanalysis of the European Centre for Medium-Range Weather Forecasts (ECMWF) (Jiang et al., 2015). One of the motivations of this study is therefore to investigate the possible causes of such overestimation of water vapour in the UTLS in GCMs and numerical weather prediction (NWP) models.
The mechanisms controlling the transport of water vapour into the stratosphere are different for tropical and mid-latitude regions. In the tropical region, water vapour enters the stratosphere primarily through the slow ascent associated with the Brewer–Dobson circulation (BDC) (Brewer, 1949). The cold temperature of the tropical tropopause layer (TTL) regulates the humidity of the air and therefore is responsible for the moistening of the stratosphere. However, it remains uncertain how factors such as the temperature in the TTL, strength of the BDC, and the vertical and horizontal mixing are weighted to determine SWV distribution and variation (Fueglistaler et al., 2014). In the extratropical region, there are several mechanisms that can influence the distribution and variation in the water vapour in the lower stratosphere (Weinstock et al., 2007). Water can be transported to the lowermost extratropical stratosphere by poleward transport from the TTL, by isentropic transport due to planetary wave activity from the tropical troposphere and by deep convection in the extratropics. Among these mechanisms, the vertical transport by mid-latitude convection, although demonstrated to be impactful by studies using in situ and remote sensing measurements, remains poorly understood (Poulida et al., 1996; Hegglin et al., 2004; Dessler and Sherwood, 2004; Ray et al., 2004; Hanisco et al., 2007; Weinstock et al., 2007; Homeyer et al., 2014, 2017; Sun and Huang, 2015; Smith et al., 2017).
A few studies have attempted to simulate the injection of water into the lower stratosphere using high-resolution NWP models. These studies found that the transport of water vapour into the stratosphere occurs through gravity wave breaking near overshooting tops (e.g., Wang, 2003; Wang et al., 2009, 2011; Homeyer et al., 2017; Dauhut et al., 2018; Lee et al., 2019). The overshooting tops form as strong updrafts within convective cells penetrate the stable stratification at the tropopause. They act as obstacles to the lower stratospheric flow and generate gravity waves. In favourable conditions (Baines, 1995; Sachsperger et al., 2015), the gravity waves break near the overshooting cloud tops, dissipate wave energy through strong turbulence and cause sudden “jumps” of air flow up to more than 2 km height. This upward wind with strong turbulence transports a substantial amount of water vapour and ice into higher altitudes in the stratosphere. The mechanism of gravity wave breaking is well demonstrated, e.g., by Fig. 7 in Wang (2003). An associated phenomenon is the so-called “jumping cirrus” (Fujita 1982), which provides evidence that ice particles are brought into and potentially hydrate the lower stratosphere. The mechanism of cross-tropopause transport of humidity associated with gravity wave breaking is generally well simulated, using idealized forcing for a short duration over a limited domain (Wang, 2003; Wang et al., 2009, 2011; Homeyer et al., 2017; Dauhut et al., 2018). In order to evaluate model results against satellite and aircraft measurements it is necessary to develop an experimental framework in which high-resolution simulations can be performed over an extended period in which observations are available.
In this study, we use a high-resolution NWP model, Global Environmental Multiscale (GEM), to simulate an observed lower stratospheric moistening event over North America from 26 to 27 August 2013 (Smith et al., 2017). The first objective is to evaluate the model capability to successfully simulate the vertical transport of water vapour through mid-latitude tropopause and reproduce the observed increase in lower stratospheric humidity during and after the deep convection event. The second objective is to evaluate, using all available satellite and aircraft measurements, the simulated water vapour fields at different horizontal resolutions, ranging from low resolution with parameterized deep convection to high resolutions with explicitly simulated convection. The third objective is to compare processes, such as direct water vapour transport vs. ice sublimation, that influence the lower stratosphere water vapour budget. In the global NWP and GCM models, the deep convection is parameterized using a mass flux approach. The complex phenomena near the tropopause during the convection are parameterized in a simplified manner, e.g., overshooting convection, or not parameterized, e.g., the falling of overshooting cloud tops (not sedimentation), gravity wave breaking and formation of jumping cirrus. In light of the aforementioned lower stratospheric humidity bias in coarse-resolution models, we are especially interested to identify possible causes of such biases.
This paper is structured as follows. The next section provides a brief description of the GEM model and the configuration of the simulation experiment, as well as the observation data for comparison and a trajectory model used to link the simulated and observed samples. This is followed by the analysis of the GEM simulation results, with a focus on the lower stratospheric water vapour budget. We then conclude with a summary of the findings and perspectives for further studies.
The NWP model used in this study is the GEM model of Environment and Climate
Change Canada (ECCC, Côté et al., 1998; Girard et al.,
2014). The dynamics of GEM are formulated in terms of the non-hydrostatic
primitive equations with a terrain-following hybrid vertical grid. It can be
run as a global model or a limited-area model and is capable of one-way
self-nesting. Milbrandt et al. (2016) described the self-nesting
configuration with horizontal grid spacing
GEM cascade domains. Thick solid lines in black, blue, red and green represent simulation domains at 10, 2.5, 1 and 0.25 km horizontal grid spacing, respectively. The thin cyan line represents the ER-2 aircraft flight path on 27 August 2013. The dotted magenta line represents evaluation Domain A, which covers the major convective events of this study. The dashed magenta line represents Domain B for comparison between aircraft observations and model simulations. The four cyan stars show the locations and times (UTC) of the lowest location of the descending–ascending trajectories of the aircraft.
For the three high-resolution simulations with 2.5, 1 and 0.25 km horizontal grid spacing, the double-moment version of the bulk cloud microphysics scheme of Milbrandt and Yau (2005a, b; hereafter referred to as MY2) is used. This scheme predicts mass and number mixing ratio for each of six hydrometeors including non-precipitating liquid droplets, ice crystals, rain, snow, graupel and hail. Condensation (ice nucleation) is formed only upon reaching grid scale supersaturation with respect to liquid (ice). For the simulation with 10 km horizontal grid spacing, the Kain–Fritsch deep convection scheme (Kain and Fritsch, 1990, 1993; hereafter referred to as KFC) is incurred. The liquid and solid cloud water content from the KFC scheme are later passed to the MY2 scheme as hydrometeors of non-precipitating liquid droplet and ice crystal category, respectively.
In addition to the MY2 and KFC schemes, the planetary boundary-layer scheme can also produce implicit clouds, particularly cumulus and stratocumulus (Bélair et al., 2005). It predicts mean liquid and ice water contents as well as cloud fraction. The shallow convection scheme (Bélair et al., 2005) is the third means by which GEM can produce clouds. It predicts mean liquid and ice water contents and cloud fraction for cells that contain shallow cumulus clouds.
The simulation at 10 km grid spacing is initialized with conditions from the ECCC global atmospheric analysis at 00:00 UTC, 24 August 2013. It runs for 96 h until 00:00 UTC, 28 August 2013. The second nested simulation at 2.5 km grid spacing runs for the same period of time. The simulations at 1 and 0.25 km grid spacing are initialized at 12:00 UTC, 25 August 2013, and run for 24 h, during which the convective event that we focus on in this study developed. Model outputs are saved every 1 min for the 2.5, 1 and 0.25 km simulations and every 5 min for the 10 km simulation.
We use the water vapour measurements from a NASA field campaign, the Studies
of Emissions and Atmospheric Composition, Clouds and Climate Coupling by
Regional Surveys (SEAC
ER-2 aircraft observation of
In order to include all of the different convective events potentially
responsible for the moistening of the lower stratosphere captured by the
aircraft measurements on 27 August 2013, we start the GEM simulations with the
largest low-resolution domain (see Fig. 1) and several days earlier.
Several mesoscale convective events developed on different days near the
Great Lakes within this domain. To identify the source of water vapour for
the aircraft-measured samples, the back trajectories of the air parcels are
simulated using the trajectory model, LAGRANTO (Sprenger and Wernli, 2015),
and GEM-generated wind fields. Using this technique, we find that the large
water vapour anomalies observed by the aircraft in Domain B on 27 August 2013
originated from two deep convection events. The first one began at the end
of 25 August and ended at the beginning of 26 August in Domain A (46 to 50
We first examine how well GEM simulates the general features of the deep
convection events of central interest (within Domain A). Figure 3 shows the
brightness temperature for the middle-infrared atmospheric window, which
indicates the cloud top height, synthesized from the GEM simulations at
different horizontal resolutions and observed by GOES-13 geostationary
satellite (the 11.2
GEM-simulated deep convective clouds compared to satellite
observation. Brightness temperatures are simulated using the RRTMG radiative
transfer model, from GEM simulations at three resolutions with 10 km (02:30 UTC, 26 August), 2.5 km (23:00 UTC, 25 August) and 1 km (23:00 UTC, 25 August)
grid spacing and compared to the brightness temperature of 11.2
From Fig. 3, we can identify the location and extent of the convective system from the white-coloured areas that signify low cloud top temperatures (high cloud tops). GEM succeeds to predict a strong convective system in the area near the Great Lakes. The locations of the convection are slightly different from one simulation to another. The 10 km simulation places the convective system west of Lake Superior. The two higher-resolution simulations put the same convective system slightly northwest of Lake Superior. The satellite image shows the storm system over Lake Superior. Another difference is the horizontal extent of the anvil clouds. The two higher-resolution simulations generate anvil clouds of very similar forms to the observation. The 10 km simulation, however, generates clouds that extend in the northeast–southwest direction and covers a noticeably larger area than what is observed by the satellite. We notice that the 10 km simulation has a larger area, with the brightness temperature lower than 210 K (magenta-highlighted areas) than those in the high-resolution simulations or those in the GOES-13 images. These highlighted zones with cold cloud tops represent the intensive convective areas. For all three simulations with different horizontal grid spacing, the convective areas are all located within Domain A during the 5 h period after the initiation of convection.
In order to inter-compare the simulations at different resolutions, we perform the evaluation for Domain A, which encompasses the convective event of interest in the three simulations at 10, 2.5 and 1 km grid spacing. The time window for the evaluation is the initial 5 h of the convection development. During this time, the convection system of interest initiated, developed multiple overshooting tops and moved from the west to the east end of Domain A. At the end of the evaluation time window, the heights of overshooting tops are observed to generally decrease (not shown), which shows that the chosen window captures the primary cross-tropopause transport of water by the convective system. Due to its limited domain area, the results of the 0.25 km simulation are not included for inter-comparisons based on Domain A. Instead, the first hour of the convection event in this simulation (before the system begin to move out of the simulation domain illustrated by the green box in Fig. 1) is analyzed for comparing some aspects of the convection (see below).
We examine the overshooting tops in the GEM simulation and especially the
gravity wave breaking process that was found to primarily account for the
water transport into the lower stratosphere in overshooting events. In our
simulations, we find both the 1 and 0.25 km simulations generate similar
structure of jumping cirrus to the previous studies (e.g., Wang et al., 2009,
2011). To illustrate the results, we show in Fig. 4 vertical cross section
in Domain A at 19:46 UTC on 25 August from the 1 km simulation. In addition,
two movies made from this simulation are included at
GEM-simulated overshooting convection. The results illustrated
here are taken from the 1 km simulation at 19:46 UTC, 25 August 2013:
Figure 4 shows a few key variables that highlight the impacts of
overshooting tops and induced gravity wave breaking. Two overshooting tops
are well identified between the longitudes of 95.94 and
95.57
The cloud ice properties are different in the overshooting tops and in the
thin ice plumes. At the altitude of
We find in our simulations that the breaking of gravity waves occurs in many
ways similar to the breaking of lee waves, which are formed when air flows
through a mountain range. On the leeward side of the mountain, when the wave
amplitude reaches a critical level, a convectively unstable region develops
and consequently leads to wave breaking (Wurtele et al., 1993; Dörnbrack,
1998; Strauss et al., 2015). In the regime of gravity wave breaking, we can
identify a sudden jump of the stratiform flow (Houghton and Kasahara, 1968).
In its vicinity, wave energy is dissipated through turbulence which causes a
strong mixing. It was found that such wave breaking occurs when the
horizontal wind speed perturbation opposes the mean flow and causes
stagnation, meeting a prognostic condition (Baines, 1995; Sachsperger et
al., 2015):
Analogies can be drawn to the gravity wave breaking near the overshooting
tops. The overshooting tops carry air mass of different horizontal velocity
into the lower stratosphere and act to block the pre-existing horizontal
flow there (westerlies with speeds ranging from 5 to 25 m s
Cross section as shown in Fig. 4 but for 19:42 UTC, 25 August 2013:
Figure 5 shows several key variables for the same cross section shown in Fig. 4 but 4 min earlier (19:42 UTC), when the condition of gravity wave
breaking (Eq. 1) is satisfied. At this moment before the formation of
jumping cirrus, the overshooting cloud top near
We emphasize the “irreversibility” of the vertical upward transport during the gravity wave breaking event. In case of a non-breaking gravity wave, the ascending air will later descend after reaching the wave ridge. In this case, the upward transport is “reversible”. In addition, there is weaker turbulent mixing to bring up the moister air below because the wave energy is less transferred to turbulence but is propagated away. In the case of gravity wave breaking, a sudden jump of air flow occurs. The wave energy is dissipated through turbulence in the vicinity of the jump which enhances the mixing and the transport of water vapour and ice from the upper troposphere to the lower stratosphere.
The occurrence of gravity wave breaking depends on the intensity of the overshooting strength. As shown in Eq. (2), the magnitude of the horizontal speed perturbation is linked to the vertical wind speed perturbation, which in this case is related to the overshooting strength. This is in agreement with the finding of Dauhut et al. (2018) that stronger overshooting tops favour wave breaking and thus facilitate more vertical water transport.
We find in our simulations that the overshooting tops and wave breaking are frequently observed in the 0.25 and 1 km simulations, with the breaking waves and jumping cirrus of typical horizontal sizes of 2 to 3 km. These phenomena are visible in the 2.5 km simulation, although with less frequency and intensity, and are not found in the 10 km simulation because the grid size cannot resolve the process.
We further examine the water vapour fields simulated by GEM at different
horizontal grid spacing. Figure 6a and b show the mean vertical profiles of
water vapour volume mixing ratio and temperature within the above-defined
Domain A and 5 h time window. All the simulations show irregular moisture
profiles near 16 km, where the vertical trend of the humidity profiles bends
and produces “bumps” (elevated water vapour contents) above the tropopause
(indicated by the circles in Fig. 6; hereafter the tropopause is defined
by the altitude where the mean lapse rate
It is challenging to observe the humidity at the levels near tropopause.
Nadir-view satellite remote sensing instruments, such as the Atmospheric
Infrared Sounder (AIRS) on board the NASA Aqua satellite, usually cannot
accurately measure the low water vapour concentrations in the lower
stratosphere (e.g., Divakarla et al., 2006), although attempts have been
made to improve the retrieval under special circumstance (e.g., Feng and
Huang, 2018). Limb-view sounders such as the Microwave Limb Sounder (MLS) on
board the Aura satellite have higher sensitivity and provide measurements of
water vapour content in the UTLS, although biases have also been noted in
these datasets. These biases might be caused, among many others, by the
averaging kernels of limb sounders which smear out the strong vertical
gradient in water vapour at the tropopause (Hegglin et al., 2013). For
instance, an underestimation of water vapour with mean bias up to
Figure 6c and d show the comparisons between the GEM simulations after
applying averaging kernels of MLS and MLS retrievals (v4.2). Because of the
scarcity of the co-located satellite data and the aforementioned
mismatch in time and location of the simulated convective system, we conduct
the comparison with respect to area averages rather than individual samples.
The MLS measurements used here include five MLS footprints located between
38–45
Back trajectories of air parcels. All the trajectories, in black lines, are initialized in Domain B at 15.5 km altitude and 19:40 UTC, 27 August. The circles indicate the initial locations of each back trajectory. The grey line illustrates the ER-2 aircraft flight path (in clockwise direction). The background image shows the water vapour content in ppmv at this level from the 2.5 km simulation. The red line highlights the back trajectory of one air parcel from its initial location in Domain B to its location in Domain A at around 23:00 UTC, 25 August 2013, when the overshooting convection occurred. The evolution of the properties of this air parcel are shown in Fig. 8. The red diamonds indicate the centre of five MLS footprints on 26 August.
High-accuracy hygrometers on-board high-altitude aircrafts provide benchmark
water vapour measurements, although the temporal and spatial coverage of the
aircraft data are limited. On 27 August 2013, the ER-2 aircraft deployed in the
SEAC
The changes of the properties of a tracked air parcel (red circle highlighted in Fig. 7) along its back and forward trajectory. The beginning time of the back and forward tracing is 19:40 UTC, 27 August. Highlighted in the rectangular shaded area is the encountering of the air parcel with the overshooting convections in Domain A at around 22:40 UTC, 25 August, which is shown by fast-rising water vapour concentration, decrease in temperature, rise in altitude and sudden changes in vertical wind speeds.
One particular note for Fig. 8 is the formation of ice shortly after 19:40 UTC, 27 August, when the humid air parcel slowly ascends with decreasing
temperature and increasing relative humidity with regard to ice. At these
altitudes, the ice water content is relatively low (
For the atmospheric layer under the tropopause between the altitude of 13.5
and 14.5 km, the horizontal wind speed increases significantly. The
back-tracking results show that the humidity and ice field in the northern
part of Domain B are linked to the convection initiated at the beginning of
27 August in Domain A. The locations of ice water content in Domain B from the
simulation partly agree with what are observed by the aircraft in Fig. 2b.
Based on the back tracing results, we noticed that the ice in Domain B is
not originally formed during the convection, but later during the slow
ascent of the humid air parcel. This is similar to the dehydration process
discussed above but at a lower altitude below the tropopause. The ice formed
at this lower altitude is more abundant (on the order of
Given the variability of water vapour in Domain B, as shown in Fig. 7, and
possible errors in the location and time of GEM-simulated water vapour
features, it would not be meaningful to compare the aircraft measurements
with GEM simulations at exactly matched locations and times. Instead, we
compare the mean vertical profiles averaged for Domain B between the
aircraft observations and GEM simulations in Fig. 6e and f. We note that only
the 10 and 2.5 km simulations cover Domain B where aircraft data are
available (see Fig. 1). We observe a slight temperature bias from both model
simulations of
We diagnose the water vapour transported across the tropopause into the lower
stratosphere in the GEM simulations. The water budget is calculated for the
rectangular box surrounded by a given lower boundary (e.g., tropopause) and
model top (
First, we use the Reynolds decomposition (Eqs. 3–5) to diagnose the direct
vertical transport (vertical advection) of water vapour for the two high-resolution simulations (1 and 2.5 km grid spacing). For the 10 km
grid spacing simulation, the vertical advection is composed of two parts:
the grid-scale advection, which is solved explicitly, and the parameterized
sub-grid-scale transport (tendency on water vapour due to KFC), which makes
this case not suitable for Reynolds decomposition.
Applying Eq. (5) at the tropopause level in Domain A (mean
Secondly, we calculate the total transport of water vapour and the contributions from direct transport and ice sublimation for each simulation. The comparison between the high-resolution simulations and the 10 km simulation is less straightforward because their tropopause heights are different (Fig. 6a and b). We therefore calculated the water vapour change due to the vertical advection and ice sublimation and vapour deposition with different altitude levels as the lower boundary from 14 to 16 km. These results are shown in Fig. 9.
Change of water vapour in Domain A during the 5 h evaluation
period with different altitude levels as the lower boundary. The circles
represent the height of tropopause (mean
The vertical advection simulated by the high-resolution models is relatively constant from 14.5 to 16 km altitude with positive values (upward transport, Fig. 9a). This upward advection is linked to the gravity wave breaking (see discussion in Sect. 3.2), which makes the stratospheric air flow jump by about 2 km upward and transports humidity into the stratosphere. The vertical advection in the 1 km simulation is generally stronger than that of 2.5 km simulation due to the more important role of wave breaking. This is in agreement with the results from Reynolds decomposition.
It is not a surprise that the higher-resolution NWP models tend to produce stronger direct vertical transports across the tropopause because, as shown in Sect. 3.1, the transport is closely related to the strength of overshooting and the breaking of the gravity waves. Similar to what was found by Weisman and Klemp (1982), we find in our GEM simulations that the simulated maximal vertical wind speed is inversely proportional to the horizontal grid spacing of the NWP model. The stronger vertical wind speed in the convection updraft leads to higher overshooting cloud top. In our cases with high-resolution simulation, the maximum cloud top altitude is 16.64 and 16.96 km for 2.5 and 1.0 km simulation, respectively. We find that the stronger overshooting wind speed in the higher-resolution simulations leads to favourable conditions for gravity wave breaking (see the discussions in Sect. 3.1) and thus more direct vertical transport. This agrees with Dauhut et al. (2018). In total, the direct vertical transport of water vapour contributes to 40 % of the total transport at the tropopause level for the 2.5 km simulation and makes up to 89 % for the 1.0 km simulation.
The total vertical advection from 10 km simulation is different from those of the two high-resolution simulations. It can be decomposed into two parts: the grid-scale explicit advection (dashed blue line in Fig. 9a) and the sub-grid-scale advection by KFC (dotted blue line in Fig. 9a). The grid-scale vertical advection of 10 km simulation is positive below the altitude of 14.3 km. It turns to negative values from 14.4 km. One of the reasons for these negative values may be the lack of representation of gravity wave breaking. Another reason is possibly the large-scale circulation induced by the convection. In the convective area, the air transported to the level above 14 km is relatively dry and cold, whereas the descending areas surrounding the convective zone are moister due to the sublimation of the ice. The sub-grid-scale advection is strongly negative at lower altitude. In KFC, this downward transport comes from the effect of compensating subsidence outside of the convective updrafts. This strong downward transport gradually reduces to zero near the tropopause.
We find large discrepancies in the contribution of ice sublimation throughout the UTLS region between the high-resolution simulations (1 and 2.5 km) and the 10 km simulations. Ice sublimation (hydration) and vapour deposition on ice (dehydration) are two opposing microphysical processes competing for a dynamical balance. Hereafter, we use sublimation to denote the combined effect of these two processes. The positive value signifies that the ice sublimation is faster than the vapour deposition, and the negative value signifies the opposite. In this study, the value of sublimation includes all the ice-phase categories. For the two high-resolution simulations, the contribution of ice sublimation reaches its maximal positive value near the altitude of 14.5 km. Toward higher altitudes, the contribution of ice sublimation decreases gradually to a small value. To focus on the tropopause level, ice sublimation has a non-negligible contribution to the total transport of water vapour for the 1 km simulation (11 %) and a larger contribution for the 2.5 km simulation (60 %). For the 10 km simulation, the mean ice sublimation rate is large and always positive. Ice sublimation is therefore the primary source of moistening of the UTLS region above the altitude of 14 km. Vertical advection does not contribute to the moistening of the UTLS region but transports a significant part of water vapour back to lower altitudes. Overall, the contribution from both processes generate a strong moistening of the UTLS region for the 10 km simulation (Fig. 9c).
An additional comparison including the 0.25 km simulation for a smaller domain (see Fig. 1) and shorter period (1 h) corroborates the above finding that the simulation tends to have a larger contribution from advection and less contribution from sublimation as the resolution increases (see Fig. SI.1 in Qu, 2019).
It is, however, interesting that compared to the high-resolution
simulations, the sublimation-induced lower stratospheric moistening is
stronger in the lower-resolution simulations (10 km grid spacing), as shown
by Fig. 9b. We find that this higher sublimation rate may be attributed to
the following factors: first, more abundant ice particles in the lower
stratosphere in the 10 km simulation, as shown by Fig. 10. The cause of
this higher mean ice water content may be due to the lack of the
parameterization of downward transport to bring the ice within the
overshooting cloud tops back into the upper troposphere in the KFC deep convection
scheme. In this scheme, the ice transported into the lower stratosphere by
the parameterized updrafts will be distributed uniformly into the 10 km
Mean profiles between 14 km and 16.5 km within Domain A and the
evaluation time window (5 h):
A separate test run with 10 km model grid spacing without the KFC scheme has been performed and shows that the ice water content at the UTLS region above the altitude of 14 km is very close to the 2.5 km simulation (Fig. 10a, purple line). These results suggest that the overestimation of ice is partly due to the use of the deep convection parameterization. With the reduction of ice in the test run, we find that the mean ice sublimation tendency is largely reduced (blue line and purple line in Fig. 10b). However, with similar amount of ice in the UTLS, the 10 km simulation without KFC still shows a higher ice sublimation rate above the altitude of 14.5 km compared with the two high-resolution simulations.
This second reason leading to stronger moistening of the lower stratosphere
in the coarse-resolution (10 km) simulation is due to the sublimation
efficiency of ice. We find that the ice transported into lower stratosphere
is largely sublimated into water vapour in the 10 km simulation and is much lower
for the higher-resolution simulations. The amount of ice transported across
the tropopause in the 1 and 2.5 km simulations is similar, i.e., 2.3 and
Distribution of a few variables in Domain A during overshooting.
The results are taken from one instance, i.e., 23:25 UTC, 25 August 2013, and one
vertical level of
What leads to the drastically different ice sublimation processes in the
coarse-resolution simulation? We find that one important factor influencing
ice sublimation efficiency in the lower stratosphere is the spatial
distribution of ice. Figure 11 shows the ice sublimation rates, IWC
distributions and a few related fields from the GEM simulations at
different resolutions at the level of
Distribution of ice with respect to
In this study we use the GEM model of ECCC to reproduce a mid-latitude lower
stratospheric moistening event over North America near the Great Lakes
during 25–26 August 2013. Simulations are conducted with a set of nested
domains at increasing resolutions from 10 to 0.25 km grid spacing.
Satellite remote sensing data from MLS as well as aircraft in situ observations
from the SEAC
By performing an intercomparison of simulations using different horizontal resolutions, we
find that the high-resolution simulations (with grid spacing
A lower stratospheric water budget has been performed to quantify the contributions of different processes. It shows that vertical advection of water vapour is one main contributor to the lower stratospheric moistening in the overshooting events. In the high-resolution simulations (0.25 and 1 km) where the gravity wave breaking process is well simulated, eddies resulting from wave breaking are found to mainly account for the direct vertical transport of water vapour into the lower stratosphere. This transport mechanism is largely dependent on the strength of overshooting (updraft speed), with higher-resolution simulations generating stronger updrafts, which enhance overall the transport of water vapour into the stratosphere.
Another important source of water vapour in the lower stratosphere is ice
sublimation. The comparisons conducted in this study show that the 10 km
simulation has a considerably higher ice sublimation rate. One of the possible
reasons is that the KFC convective scheme that is used in this
simulation brings more ice into the UTLS region which enhances the production
of water vapour through the ice sublimation process. The cause of this
overproduction of ice is likely associated with the lack of downward
transport of ice that is observed after the overshooting cloud tops reach
their maximal height in the lower stratosphere. The simple
entrainment–detrainment model used in KFC scheme may not represent the
complex processes near and above tropopause well during the convective event. One
solution is to add an element in the KFC scheme to take this
downward transport above the tropopause into account. Another solution is to increase the
ice particle size in the UTLS so that the ice can sediment faster and hence
reduce the ice water content at these levels. Another possible reason why
the 10 km model has a higher sublimation rate is the high ice sublimation
efficiency. This high efficiency is due to the different distribution of ice
water contents compared to the those of high-resolution models. This
results from the inability to resolve overshooting tops by the coarse grid
boxes (10 km
The material includes two videos showing the model-simulated transport of water vapour into the lower stratosphere and three supporting figures for budget analysis and back-tracing studies (
YH, PAV, JNSC, MKY and KW conceptualized the research goals and aims. ZQ and YH designed the experiments and ZQ carried them out. ZQ developed the model code, performed the simulations and prepared the manuscript with contributions from all co-authors.
The authors declare that they have no conflict of interest.
The authors thank Katja Winger and Sylvie Leroyer for their help with running the high-resolution GEM model, and Yuwei Wang for his help with running the RRTMG model.
This research has been supported by the Atmospheric Science Data Analysis programme of the Canadian Space Agency (grant no. 16SUASURDC).
This paper was edited by Martina Krämer and reviewed by Eric Jensen and Christian Rolf.