Using the Modern Era Retrospective-Analysis for Research and Applications (MERRA) and MERRA-2 reanalysis winds, temperatures, and anvil cloud ice, we explore the impact of varying the cloud nucleation threshold relative humidity (RH) and high-frequency gravity waves on stratospheric water vapor (H2O) and upper tropical tropopause cloud fraction (TCF). Our model results are compared to 2008/2009 winter TCF derived from Cloud-Aerosol Lidar with Orthogonal Polarization and H2O observations from the Microwave Limb Sounder (MLS). The RH threshold affects both model H2O and TCF, while high-frequency gravity waves mostly impact TCF. Adjusting the nucleation RH and the amplitude of high-frequency gravity waves allows us to tune the model to observations. Reasonable observational agreement is obtained with a nucleation threshold between 130% and 150% RH consistent with airborne observations. For the MERRA reanalysis, we lower the tropopause temperature by 0.5 K roughly consistent with GPS radio occultation measurements and include ~0.1 K high-frequency gravity wave temperature oscillations in order to match TCF and H2O observations. For MERRA-2 we do not need to adjust the tropopause temperature nor add gravity waves, because there are sufficient high-frequency temperature oscillations already present in the MERRA-2 reanalysis to reproduce the observed TCF.
- We explore the sensitivity of the stratospheric water vapor (H2O) to high-frequency gravity waves and the cloud nucleation threshold
- Gravity waves have little impact on H2O, but increase clouds
- To match observed water vapor and cloud fraction, the global average nucleation threshold relative humidity is between 130 to 150%
Stratospheric water vapor (H2O) is determined by the processing of air at the tropical tropopause cold trap and by the stratospheric oxidation of methane. In the simplest terms, air parcels moving upward into the tropical stratosphere pass through the tropopause cold point where ice forms and falls into lower layers, dehydrating the air entering the stratosphere [Fueglistaler et al., 2009; Randel and Jensen, 2013, and references therein].
Trajectory models have succeeded in simulating many aspects of stratospheric water vapor distribution [e.g., Fueglistaler et al., 2005; Schoeberl and Dessler, 2011, hereafter SD11; Dessler et al., 2014; Schoeberl et al., 2014; hereafter S14, Wang et al., 2015; Ueyama et al., 2015]. SD11 introduced the forward domain-filling (FDF) trajectory model in which parcels are near-continuously released in the upper troposphere and fill up the stratosphere. In the FDF model, horizontal winds and diabatic heating rates determine the parcel motion, and temperature determines the water vapor abundance through a dehydration adjustment. The winds, heating rates, and temperatures are provided by meteorological reanalyses. In many previous publications [e.g., Liu et al., 2010; SD11; S14; Dessler et al., 2014], the dehydration adjustment simply requires the air parcel not exceed a fixed relative humidity (RH); we refer to this process as instantaneous dehydration (ID). In SD11 we used the Modern Era Retrospective-Analysis for Research and Applications (MERRA) reanalysis and compared our results to Microwave Limb Sounder (MLS) observations of H2O, obtaining best agreement when we slightly increased the ID threshold from 100% RH to 104%. We now understand that this agreement was the result of canceling errors. MERRA tropopause temperatures are too warm [Wang et al., 2015], and ID is too efficient a dehydration approximation (S14); these two effects roughly cancel, producing a reasonable stratospheric H2O abundance.
In S14 we introduced a simple cloud model that allowed ice to nucleate, grow, and gravitationally settle out. The net effect of cloud processing is to slow the removal of water compared to ID as previously noted by Jensen and Pfister  and Fueglistaler and Baker . The tropical tropopause cloud fraction (TCF) generated by the cloud model can also be used to quantitatively check the model results against observations.
In Dessler et al.  we showed that our model was able to accurately reproduce the observed stratospheric H2O anomalies over a 30 year period even with ID. This study outlined the dynamical mechanisms that influence the relative tropopause temperature changes and the resulting H2O response. However, a much more difficult problem, and a severe test for any reanalysis systems, is reproducing the actual stratospheric H2O concentration in addition to the anomalies.
S14 identified a number of processes that determine stratospheric H2O concentration. These processes include (1) the tropical tropopause temperature which can vary between reanalysis systems [Schoeberl et al., 2013], (2) the level of supersaturation required before cloud processes (or dehydration) are initiated (S14), (3) the level of convective penetration into the tropical tropopause layer (TTL) [S14, Ueyama et al., 2015], and (4) the amplitude of short timescale and small spatial-scale temperature oscillations produced by atmospheric gravity waves—often not resolved in the reanalysis [Jensen and Pfister, 2004; Wang et al., 2015; Schoeberl et al., 2015].
Recent results from aircraft campaigns [Krämer et al., 2009, Jensen et al., 2013, Rollins et al., 2016] show that H2O concentration in the upper tropical troposphere routinely exceeds supersaturation. Likewise, it would appear that the ubiquitous presence of gravity waves [Bacmeister et al., 1996; Podglajen et al., 2016] would contribute to the reduction in stratospheric H2O as a result of colder temperatures caused by waves [Kim and Alexander, 2015]. However, the impact of gravity waves is more complex than simply lowering the temperature experienced by an air parcel. Dehydration is not significantly enhanced by high-frequency gravity waves because such waves have high cooling rates that preferentially nucleate a large number of small ice crystals. Because of the limits on available water vapor, these ice crystals grow slowly and do not have enough time to gravitationally settle before the temperature perturbation reverses and the ice evaporates [Fueglistaler and Baker, 2006; Jensen et al., 2012; Schoeberl et al., 2015]. In the opposite extreme, if the short period gravity waves are weak, air parcel cooling rates are too low to nucleate a sufficient number of crystals to efficiently dehydrate the air [Schoeberl et al., 2015].
The purpose of this work is to report on the impact of changing the nucleation threshold and the high-frequency gravity wave amplitude on the TCF and H2O in our FDF model. We use both MERRA [Rienecker et al., 2011] and MERRA-2 [Molod et al., 2015] reanalyses in our simulations.
2 Model Description and Data Used
2.1 Trajectory Model
The FDF uses the diabatic trajectory model of Bowman and Carrie  and follows the configuration described in S14. We impose high-frequency gravity waves roughly following the approximation of Jensen and Pfister . The gravity wave temperature perturbations are assumed to take the form of a superposition of waves, and the temperature perturbation, T', is added to the reanalysis temperature field experienced by the parcel. The gravity wave temperature perturbation is defined as follows T ′ = ΣiT0(ω0/ωi)cos(ωit + ϕi) where ωi is the wave frequency, ϕ is the phase, and ω0 is the base frequency. T0 is the gravity wave temperature amplitude at ω0. For our parameterization, the base frequency is 2π/(1 day) and wave frequencies generated are between the base frequency and the Brunt-Väisälä frequency. The cloud model uses at time step of 0.005 days; and thus, we use about 350 frequencies for a 1 day interval. Our parameterization generates a wave temperature spectrum with a −2 power law on average. This power law behavior has been observed by long duration balloons in the tropics [Podglajen et al., 2016].
Our cloud model (CM) is described in S14; it is a zero dimensional model following the air parcel outputting ice mass, crystal number density, and H2O mixing ratio. The model triggers nucleation at a prescribed RH, and the number of ice particles produced upon nucleation is proportional to the cooling rate using the relationship derived by Kärcher et al. . The CM includes gravitational sedimentation, taking into account the vertical velocity associated with slow parcel ascent or descent as well as the gravity wave vertical velocities. Ice crystals are assumed to be spheres. We assume a fixed cloud geometrical thickness of 500 m based on the TTL cloud thickness distribution observed by Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) (S14). We also assume that ice falling out of the cloud slowly sublimates in subsaturated layers below the cloud. The CM produces good agreement with observational data from aircraft flights [Schoeberl et al., 2015].
MERRA provides a convective or anvil ice data product. As described in S14, for parcels that intersect convective ice the water vapor content is set to the saturation mixing ratio and a very small amount of ice is added to the parcel so that anvil clouds can be included in our cloud fraction analysis. Essentially, we assume that the convective ice simply saturates the air parcel, because MERRA's anvil ice particles are so large (30–60 µm) they quickly fall out and do not contribute to the total water budget.
For the experiments shown here, we run the model from 2003 to 2011. Starting at 1 January 2003 provides the model sufficient time to populate the stratosphere with parcels and produce a reasonable age spectrum for data analyzed in 2008. For this study we focus on results in December 2008 and January and February 2009 (DJF). Analysis of DJF 2007–2008 and DJF 2009–2010 gives similar results to 2008–2009.
MERRA dynamics and temperature biases relative to other assimilated data sets are described in Rienecker et al. , and our FDF model results using other reanalyses are evaluated in Schoeberl et al. . Wang et al.  compared MERRA tropical upper tropospheric and lower stratospheric temperatures to GPS radio occultation (RO) measurements and found that the average MERRA tropical tropopause temperature was 0.3–0.4 K warmer than the measurements. Kim and Alexander  (KA) showed that standard interpolation schemes create an additional positive temperature bias of up to 0.35 K. In Schoeberl et al.  we used the KA scheme to interpolate the MERRA data to the trajectory points between MERRA levels which are ~1 km apart. This interpolation scheme is computationally costly and would significantly reduce the number of experiments we can perform. We have not performed a KA interpolation for MERRA-2 data, but we have found that we can approximate temperature bias and the interpolation error for comparison to MERRA by simply lowering the MERRA upper tropospheric temperature by 0.5 K. As it turns out, this offset is consistent with the difference between MERRA and MERRA-2 tropopause temperatures (see below). Note that application of the KA scheme to MERRA-2 results will also slightly reduce the minimum tropopause temperature at the tropopause.
MERRA-2 has a significant number of improvements over MERRA including a fully realized quasi-biennial oscillation in the lower stratosphere [Molod et al., 2015]. MERRA-2 also assimilates GPS RO data so that the tropopause temperature bias between GPS RO and the reanalysis should be minimal. For this study, we remap the original 6-hourly MERRA-2 data with a spatial resolution of 0.625° × 0.5 to a 2° × 2° grid using bilinear interpolation while keeping the native vertical grid at 72 layers (same as MERRA). The diabatic heating is derived from 3-hourly averaged assimilated air temperature tendency at model layers. The horizontal winds and diabatic heating are linearly interpolated to isentropic levels for use by the trajectory model. Unfortunately, MERRA-2 does not output a convective cloud ice product similar to MERRA that was used in S14, so for these simulations we use the MERRA cloud ice recognizing that MERRA-2 distribution of convective ice will be slightly different than MERRA and this might alter our results. In a subsequent paper, we will address this issue in more detail.
2.3 Observations of Cirrus and Water
CALIOP lidar nighttime measurements [Winker et al., 2010] provide observations of tropical cirrus [Vaughan et al., 2009, Young and Vaughan, 2009; Avery et al., 2012]. We screen the V3 CALIOP 5 km profile data for clouds with ice water content > 0.01 mg/m3 over the profile corresponding to the lowest observed cloud ice water content (IWC) detected at night with an integrated backscatter of at least 0.001 sr−1. This threshold includes subvisible cirrus but is high enough to suppress most false positive cloud identifications. We use the same ice threshold in the analysis of the model results. We estimate the CALIOP cloud fraction precision to be about 10% based on the observation that random noise in the lidar backscatter measurements at night is consistently less than 5%.
We compare our stratospheric H2O calculations with observations from the Aura Microwave Limb Sounder (MLS) [Read et al., 2007] V4.2. The MLS precision in the lower stratosphere is about 6–15% with accuracy of 10% or less.
3 Trajectory Modeling Results
3.1 Using the MERRA Reanalysis
In Figure 1 we summarize the results for 12 experiments that provide a semisystematic assessment of the role of nucleation RH and the effects of high-frequency gravity waves. The results are shown as a scatter diagram of model area-weighted average TCF greater than 16 km from 30°S to 30°N plotted against the pole-to-pole lower stratospheric H2O, 18–30 km. Also plotted in Figure 1 are the equivalent observations from CALIOP and MLS with their estimated precision.
3.1.1 Temperature Offset
The H2O value obtained using ID at 100% RH (indicated by the arrow in Figure 1) is 3.3 ppmv. The reanalysis with a temperature offset combined with ID produces too dry a stratosphere as noted in section 1. When the CM is used, the stratospheric water vapor is much higher.
First, we focus on results where RH nucleation trigger is set at 160%—the homogeneous ice nucleation threshold at tropical tropopause temperatures [Koop et al., 2000]. There are two points, those where the MERRA temperatures are offset by −0.5 K and those without an offset. Without the −0.5 K temperature offset, the model produces a very low cloud fraction and a too wet stratosphere. The wetter stratosphere is due to the warmer tropopause in MERRA than is observed [Wang et al., 2015]. Offsetting the temperature produces closer agreement with H2O observations but insufficient TCF.
3.1.2 Nucleation Threshold
At high RH nucleation thresholds, some parcels can move into the stratosphere without dehydrating or forming clouds. It is thus physically reasonable that cloudiness is related to the amount of stratospheric dehydration. Lowering the nucleation threshold to 130–135% RH decreases stratospheric H2O and increases TCF. This brings the simulations into better agreement with observations, but TCF is still a little low. The relationship between nucleation threshold and TCF is further quantified in Figure 2a where we show the change in TCF as a function of the nucleation RH. The relationship is nearly linear with the cloud fraction decreasing with increasing RH at −0.017/10% RH.
3.1.3 Added Gravity Wave Oscillations
Figure 1 also shows that we can increase the cloudiness by increasing the amplitude of high-frequency gravity wave temperature oscillations. The values for T0 used here are similar to those from Jensen and Pfister  (0.1–0.18 K). The increase in T0 slightly increases H2O. The explanation for this counter-intuitive effect is that excessive high-frequency waves (periods shorter than a day) can reduce the dehydration efficiency of clouds [Schoeberl et al., 2015]. This occurs because the large parcel cooling rates generated by these waves generate an abundance of small ice crystals which grow and settle too slowly to allow efficient dehydration. (For example, a 2 µm ice particle will fall at approximately 0.7 mm/s in the tropical upper troposphere (125 hPa) compared to a 10 µm crystal which falls at 18 mm/s.) As a result of the gravity waves we see an increase in TCF combined with a small increase in H2O.
Figure 2b shows that the increase in TCF is nearly linear with T0. The cloud fraction roughly increases with T0 at 0.246/K. It is apparent that by adjusting the nucleation RH and the gravity wave amplitude, we can “zero in” on the observations of both H2O and TCF as shown in Figure 1. An optimal result might be nucleation threshold of 135% RH and a gravity wave amplitude of T0 = 0.08 K, for example. However, given the uncertainty of the data and the simplifications used in the CM, it might be safer to say that the average tropical troposphere nucleation threshold is likely somewhere between 130 and 150% with T0 between 0 and 0.1 K.
To summarize, adjusting the temperature offset and the nucleation threshold allows us to zero in on the right water vapor amount, while adjusting the gravity wave amplitude allows us to fine tune the cloud fraction with only a small affect on H2O.
3.2 Using the MERRA-2 Reanalysis
Figure 3 shows the results of four experiments using MERRA-2. To make comparisons with the MERRA results easier, we have also overlaid the results for T0 = 0, 160% RH MERRA minus 0.5 K for reference (orange box). Most of MERRA-2 cases are quite close to observations. We have compared MERRA, MERRA-2, and GPS RO temperatures in the tropical upper troposphere and lower stratosphere in the range ±18° of the equator and find the MERRA-2 tropical tropopause is about 0.5 K colder than MERRA. This is likely due to the assimilation of GPS RO data in MERRA-2. MERRA-2 also apparently does not require additional high-frequency gravity waves, and this point will be discussed further below. MERRA minus 0.5 K and MERRA-2 also produce similar ID results.
Figure 4 shows a quantitative assessment of the changes in model H2O relative to the nucleation threshold RH for both MERRA and MERRA-2. We use the runs without gravity waves to avoid the slight changes in H2O that gravity waves generate. For MERRA-2, the stratospheric H2O increases by about 0.018 ppmv per percent increase in nucleation RH, and the line intersects the MLS observed H2O at ~145% RH. MERRA (T0 = 0, −0.5 K) results are also shown in Figure 4 for comparison. The MERRA and MERRA-2 sensitivities are about the same suggesting that at 140–145% RH nucleation threshold, the stratospheric H2O will agree with MLS. The sensitivity to RH is 0.1–0.2 ppmv per percent RH.
4 Comparison Between MERRA and MERRA-2
4.1 Overall Cloud, H2O, and Dehydration Characteristics
Below we compare two of the simulations shown in Figures 1 and 3. We choose two simulations that are close to the observational domain. For MERRA, we choose RH 135% with T0 = 0.08, for MERRA-2 we choose RH 150%. Figure 5 shows the zonal mean H2O field compared with MLS. Both model simulations are in general agreement with observations, but there are interesting differences. First, it is obvious that the tape recorder signal is moving upward too slowly in the MERRA as earlier noted by Schoeberl et al. . This problem does not appear in MERRA-2; the H2O anomaly at 24 km from the previous winter lines up nicely with MLS although the anomaly in both simulations is weaker than MLS observations. South of 60°S neither model simulates the intense dry zone below 18 km.
Figure 6 shows the model zonal mean cloud fraction (ZCF) compared to CALIOP observations. The ZCF includes anvil ice clouds from the MERRA reanalysis because when anvil cloud hydration occurs, a very small amount of ice is added to indicate that a cloud is present. As shown in Schoeberl et al. , the MERRA ZCF shows a broader meridional distribution of clouds, and the MERRA ZCF does not extend as high as CALIOP ZCF. The MERRA-2 ZCF is narrower in latitude, in better agreement with observations. Both MERRA and MERRA-2 ZCF are lower than observations at higher altitudes. This is likely the result of the Relaxed Arakawa-Schubert (RAS) [Moorthi and Suarez, 1992] convective parameterization used in both MERRA and MERRA-2. RAS tends to underestimate the height of convection in the tropics (A. Molod, private communication, 2016, and this assessment also agrees with a comparison of convective height observations derived by L. Pfister, private communication, 2016).
Figure 7 shows the two model runs cloud fraction in the 15–18 km range compared with CALIOP observations. The cloud fraction is higher in the MERRA case (compare Figures 1 and 3 cloud fractions) as expected. Nonetheless, the patterns are quite similar with more cloud over the Tropical West Pacific (TWP) than any other region.
Figure 8 shows the final dehydration point distribution for both MERRA and MERRA-2 for the cases shown in Figure 5. MERRA-2 more closely resembles the pattern of final dehydration points seen with ERA Interim reanalysis [Schoeberl et al., 2012, Figure 3] with slightly more dehydration occurring in the Northern Hemisphere than in the Southern Hemisphere. In contrast, MERRA shows more dehydration in the Southern Hemisphere and this leads to a meridional widening of the high cloud fields shown in Figure 5.
4.2 Temperature Fluctuations Near the Tropopause
To further understand the differences between MERRA and MERRA-2—especially those that give rise to the differences shown in Figures 1 and 3, we have analyzed the 85 hPa equatorial temperature fields; this is approximately the tropical tropopause and the final dehydration level in winter. For DJF 2008/2009, MERRA-2 level is 0.64 K colder than MERRA which is consistent with the GPS RO comparisons described above and the −0.5 K offset we needed in order for MERRA to model the observed H2O concentration.
The interesting fact that MERRA-2 does not need additional high-frequency temperature fluctuations to produce the observed cloud fraction suggests that the temperature fluctuations in MERRA-2 are higher than MERRA. To confirm this idea, Figure 9 shows the power spectra for temperature fluctuations over the winter period, 2008/2009; the spectrum of temperature fluctuations experienced by a Lagrangian air parcel moving along the equator at 85 hPa with the average wind speed. This spectrum is generated by converting the spatial variability in temperature to a temporal variability as experienced by the parcel (i.e., dT/dt ∼ u∂T/∂x) and neglects the spectral contribution by waves with a nonzero phase speed. We then perform a Fourier transform on the temperature perturbations to create the power spectrum.
Figure 9 shows that the MERRA-2 and MERRA spectra match quite well for wave periods longer than a day, but MERRA-2 shows significantly higher power at periods less than a day. For MERRA the power spectrum follows a −4 power law, so short period fluctuations are weaker than MERRA-2, which follows a −2 power law. This result explains why MERRA-2 does not require the addition of high-frequency gravity wave temperature fluctuations that we had to add to the MERRA reanalysis. Recent balloon observations of high-frequency gravity waves in the tropical tropopause region display −2 power law dependence [Podglajen et al., 2016], so our model assumptions are consistent with observations.
5 Summary and Discussion
Using our FDF model we have investigated two factors that control stratospheric H2O and upper tropospheric tropical cloud fraction (TCF) in the winter tropics: the nucleation RH threshold for clouds and the amplitude of the high-frequency gravity waves. High-frequency gravity waves are ubiquitous in the lower stratosphere and according to our results play an important role in determining TCF; however, they do not appear to be as important in controlling the lower stratospheric water vapor concentration.
Overall, the nucleation RH exerts the strongest control on H2O; the lower the RH value, the drier the stratosphere and cloudier the upper troposphere. This makes complete physical sense, with a lower threshold RH, more parcels are dehydrating before entering the stratosphere which lowers the stratospheric H2O and increases cloudiness.
Even after crudely correcting for the warm temperature bias in MERRA, we generally cannot achieve the observed cloud fraction without increasing the amplitude of high-frequency gravity waves. In other words, we “tune” our model to agree with observation using a RH threshold between 135–150% and T0 values up to 0.08 K.
When we repeat these experiments with the recently released MERRA-2 reanalysis, we find that (1) we do not need a temperature offset, MERRA-2 agrees much more closely with GPS RO observations, and (2) we do not need to add gravity to generate good agreement with observations. MERRA-2 final dehydration locations are closer to what we have previously found for ERA Interim [Schoeberl et al., 2013], and the MERRA-2 tape recorder phase is closer to what is observed by MLS (Figure 5). We do not need a temperature offset because MERRA-2 tropical tropopause is at least 0.5 K colder than MERRA, in better agreement with observations [Wang et al., 2015]. We do not need gravity waves with MERRA-2 simulations because the high-frequency temperature variance is higher in MERRA-2. The MERRA-2 power spectrum has exactly the −2 spectral dependence we imposed on the MERRA temperature field by adding the temperature oscillations.
Our model shows a fairly linear response to changes in nucleation relative humidity and the gravity wave amplitude. This suggests that it is possible to tune our results to observations; and thus, it might be possible to use the model to back out the global nucleation threshold RH or the effective gravity wave amplitudes. With both MERRA and MERRA-2, our results point to a nucleation threshold RH that is much higher than 100% and this is consistent with frequent observations of supersaturated air in the upper tropical troposphere [Kramer et al., 2009; Jensen et al., 2013; Rollins et al., 2016] and suggest the possibility of wide spread homogeneous nucleation of ice in the tropical upper troposphere [Barahona and Nenes, 2008]; however, we cannot rule out that cloud formation could be triggered at supersaturation values lower than 160% by heterogeneous processes [Barahona and Nenes, 2009; Cziczo et al., 2013] when a high concentration of dust or other ice nuclei is present in the upper troposphere. That said, air near the tropopause has likely been repeatedly processed by convection and scrubbed of potential ice nuclei. In support of this idea, we note that in our model, parcels initially released at 360 K have a ~90% chance of encountering convection before reaching final dehydration level—the final cloud formation point—near 375 K. At subtropopause altitudes it may be more likely that heterogeneous nucleation processes could occur since fewer air parcels have been scrubbed by convection. This suggest that the “global average” nucleation threshold RH might vary spatially and vertically, increasing toward the tropopause.
CALIOP, MLS, MERRA, and MERRA-2 data used in this study are publically available from NASA at no charge. M. Avery acknowledges funding from the CALIPSO project. This work was supported under NASA grant NNX13AK25G and by the ATTREX Project.
- 2012), Cloud ice water content retrieved from the CALIOP space-based lidar, Geophys. Res. Lett., 39, L05808, doi:10.1029/2011GL050545.
- 1996), Stratospheric horizontal wavenumber spectra of winds, potential temperature, and atmospheric tracers observed by high-altitude aircraft, J. Geophys. Res., 101, 9441–9470, doi:10.1029/95JD03835.
- 2008), Parameterization of cirrus formation in large scale models: Homogeneous nucleation, J. Geophys. Res., 113, D11211, doi:10.1029/2007JD009355.
- 2009), Parameterizing the competition between homogeneous and heterogeneous freezing in cirrus cloud formation—Polydisperse ice nuclei, Atmos. Chem. Phys., 9, 5933–5948.
- 2002), The mean-meridional transport circulation of the troposphere in an idealized GCM, J. Atmos. Sci., 59, 1502–1514.
- 2013), Clarifying the dominant sources and mechanisms of cirrus cloud formation, Science, 340(6138), 1320–1324, doi:10.1126/science.1234145.
- 2014), Variations of stratospheric water vapor over the past three decades, J. Geophys. Res. Atmos., 119, 12,588–12,598, doi:10.1002/2014JD021712.
- 2006), A modelling study of the impact of cirrus clouds on the moisture budget of the upper troposphere, Atmos. Chem. Phys., 6, 1425–1434, doi:10.5194/acp-6-1425-2006.
- 2005), Stratospheric water vapor predicted from the Lagrangian temperature history of air entering the stratosphere in the tropics, J. Geophys. Res., 110, D08107, doi:10.1029/2004JD005516.
- 2009), The tropical tropopause layer, Rev. Geophys., 47, RG1004, 10.1029/2008RG000267.
- 2004), Transport and freeze-drying in the tropical tropopause layer, J. Geophys. Res., 109, D02207, doi:10.1029/2003JD004022.
- 2013), Ice nucleation and dehydration in the tropical tropopause layer, Proc. Natl. Acad. Sci. U.S.A., 110, 2041–2046, doi:10.1073/pnas.1217104110.
- 2012), Physical processes controlling ice concentrations in cold cirrus near the tropical tropopause, J. Geophys. Res., 117, doi:10.1029/2011JD017319.
- 2006), Physically based parameterization of cirrus cloud formation for use in global atmospheric models, J. Geophys. Res., 111, D01205, doi:10.1029/2005JD006219.
- 2013), A new wave scheme for trajectory simulations of stratospheric water vapor, Geophys. Res. Lett., 40, 5286–5290, doi:10.1002/grl.50963.
- 2015), Direct impacts of waves on tropical cold point tropopause temperature, Geophys. Res. Lett., 42, 1584–1592, doi:10.1002/2014GL062737.
- 2000), Water activity as the determinant for homogeneous ice nucleation in aqueous solutions, Nature, 406, 611–614.
- 2009), Ice supersaturations and cirrus cloud crystal numbers, Atmos. Chem. Phys., 9, 3505–3522.
- 2010) Advection-condensation paradigm for stratospheric water vapor, J. Geophys. Res., 115, D24307, doi:10.1029/2010JD014352.
- 2015), Development of the GEOS-5 atmospheric general circulation model: Evolution from MERRA to MERRA2, Geosci. Model Dev., 8, 1339–1356, doi:10.5194/gmd-8-1339-2015.
- 1992), Relaxed Arakawa-Schubert: A parameterization of moist convection for general-circulation models, Mon. Weather Rev., 120, 978–1002.
- 2016), Lagrangian temperature and vertical velocity fluctuations due to gravity waves in the lower stratosphere, J. Geophys. Res., doi:10.1002/2016GL068148, in press.
- 2013), Physical processes in the tropical tropopause layer and their role in a changing climate, Nat. Geosci., 6, 169–176, doi:10.1038/ngeo1733.
- 2007), Aura Microwave Limb Sounder upper tropospheric and lower stratospheric H2O and relative humidity with respect to ice validation, J. Geophys. Res., 112, D24S35, doi:10.1029/2007JD008752.
- 2011), MERRA—NASA's Modern-Era Retrospective Analysis for Research and Applications, J. Clim., 24, 3624–3648, doi:10.1175/JCLI-D-11-00015.1.
- 2016), Observational constraints on the efficiency of dehydration mechanisms in the tropical tropopause layer, Geophys. Res. Lett., 43, 2912–2918, doi:10.1002/2016GL067972.
- 2011), Dehydration of the stratosphere, Atmos. Chem. Phys., 11, 8433–8446, doi:10.5194/ASC-11-8433-2011.
- 2012), Simulation of stratospheric water vapor and trends using three reanalyses, Atmos. Chem. Phys. Discuss., 12, 8433–8463, doi:10.5194/ASCd-12-8433-2012.
- 2013), Modeling upper tropospheric and lower stratospheric water vapor anomalies, Atmos. Chem. Phys. Discuss., 13, 9653–9679, doi:10.5194/ASCd-13-9653-2013.
- 2014), Cloud formation, convection, and stratospheric dehydration, Earth Space Sci., 1, 1–17, doi:10.1002/2014EA000014.
- 2015), Gravity waves amplify upper tropospheric dehydration by clouds, Earth Space Sci., 2, 485–500, doi:10.1002/2015EA000127.
- 2015), Dynamical, convective, and microphysical control on wintertime distributions of water vapor and clouds in the tropical tropopause layer, J. Geophys. Res. Atmos., 120, 10,483–10,500, doi:10.1002/2015JD023318.
- 2009), Fully automated detection of cloud and aerosol layers in the CALIPSO lidar measurements, J. Atmos. Oceanic Technol., 26, 2034–2050, doi:10.1175/2009JTECHA1228.1.
- 2015), The impact of vertical resolution of temperature on trajectory modeling of stratospheric water vapor, Atmos. Chem. Phys., 15, 3517–3526.
- 2010), The CALIPSO mission: A global 3D view of aerosols and clouds, Bull. Am. Meteorol. Soc., 91, 1211–1229, doi:10.1175/2010BAMS3009.1.
- 2009), The retrieval of profiles of particulate extinction from Cloud-Aerosol Lidar Infrared Pathfinder Satellite Observations (CALIPSO) data: Algorithm description, J. Atmos. Oceanic Technol., 26, 1105–1119, doi:10.1175/2008JTECHA1221.1.