Lagrangian trajectories driven by reanalysis meteorological fields are
frequently used to study water vapor (H

Stratospheric water vapor (H

When air crosses the tropical tropopause layer (TTL), it experiences
multiple dehydrations due to encounters with lower temperatures, and the final
stratospheric H

The details of the transport and dehydration process can be understood by
performing Lagrangian trajectory simulations, which track the temperature
history of a large number of individual parcels. Unlike modeling chemical
tracers, which depends strongly on the transport imposed (Ploeger et al.,
2011; Wang et al., 2014), the simulation of H

In this paper, we use a forward, domain-filling trajectory model to study the
detailed dehydration behavior of the humidity of air parcels entering the
tropical lower stratosphere. Previous analyses have demonstrated that this
model can accurately simulate many aspects of the observed stratospheric
H

This is accomplished by comparing trajectory results from using NASA Modern Era Retrospective – analysis for Research and Applications (MERRA) (Rienecker et al., 2011) temperatures on standard model levels to using temperatures with finer vertical structures, which include GPS temperatures at finer vertical resolution and the MERRA temperatures adjusted to account for finer vertical structure induced by waves (Kim and Alexander, 2013). This will help us to further understand the importance of the vertical structure of tropopause temperatures in dehydrating air entering the stratosphere.

The trajectory model used here follows the details described in Schoeberl and
Dessler (2011), with parcel positions integrated using the Bowman trajectory
code (Bowman, 1993; Bowman et al., 2013). This model has been proven capable
of simulating stratospheric H

The parcel initiation level is chosen to be the 370 K isentrope, which is
above the level of zero radiative heating (

H

In addition to H

Along each trajectory, we locate the point where air experiences the coldest
temperature as the final dehydration point (FDP), which determines the
stratosphere entry-level H

In this paper, we use MERRA (Rienecker et al., 2011) circulation to advect
parcels. This includes horizontal wind components and total diabatic heating
rates. As shown in Schoeberl et al. (2012, 2013), the trajectory model driven by
this reanalysis yields excellent estimates of H

Driven by the same circulation, trajectory runs using three different
temperature data sets are compared to quantify the uncertainties induced by
different vertical structures of temperatures: (1) using MERRA standard
temperatures on model levels (MER-T), denoted as

Owing to its high vertical resolution, GPS temperature profiles capture the
cold-point tropopause with high accuracy. In this paper, we use GPS wet
profile (wetPrf) retrieved at 100 m vertical resolution using a
one-dimensional variational technique based on ECMWF analysis. The wetPrf and
GPS atmospheric profile (atmPrf, derived assuming no water vapor in the air)
temperatures are essentially the same at 200–10 hPa, but at altitudes lower
than the 200 hPa level the errors in atmPrf could be as high as

Comparison of temperatures from raw GPS

The GPS radio occultation (RO) technique makes the data accuracy independent
of platforms. That could make the biases among different RO payloads as
low as 0.2 K in the tropopause and stratosphere (Ho et al., 2009).
Therefore, to compensate for the relatively lower horizontal resolution (relative
to that of reanalysis), we include GPS RO from all platforms. This includes
the Constellation Observing System for Meteorology, Ionosphere, and Climate
(COSMIC) (Anthes et al., 2008), the CHAllenging Minisatellite Payload (CHAMP)
satellite (Wickert et al., 2001), the Communications/Navigation Outage
Forecasting System (CNOFS), the Gravity Recovery And Climate Experiment
(GRACE) twin satellites (Beyerle et al., 2005), the Meteorological
Operational Polar Satellite – A (MetOp-A), the Satellite de Aplicaciones
Cientifico-C (SACC) satellite (Hajj et al., 2004), and the TerraSAR-X
(TerraSAR-X). There are

Each day, GPS temperature profiles are binned to 200 m vertical resolution.
Horizontally, we grid data into 2.5

Different temperature data sets used in trajectory model.

Figure 1 shows a snapshot of the 100 hPa GPS raw (panel a) and gridded (panel b) temperature on 1 January 2010, compared with MERRA temperature (panel c). It demonstrates that the gridded GPS temperature captures most of the features, although some detailed structure might be lost due to its relatively sparse sampling.

Figure 2 shows the GPS and MERRA temperatures in the TTL (panel a) and their
differences (GPS–MERRA) (panel b, extended to 31 hPa) averaged over the
deep tropics (18

Wave-induced disturbances on tropopause temperatures are underrepresented by
current reanalyses (Kim and Alexander, 2013). At the reanalysis model levels,
temperature variability at timescales shorter than

To overcome these limitations, a scheme developed by Kim and Alexander, based
on wave amplification from radiosonde observations and frequency-domain
interpolation, has been proven effective in recovering subseasonal (less than
90 days) wave-induced variability and creating wave-like vertical structures
in reanalysis temperatures (see Kim and Alexander, 2013, for more details).
Applying this scheme to MERRA temperature records yields a new MERRA
temperature data set (MER-Twave) that has finer vertical structure induced by
waves (see Fig. 3 in Kim and Alexander, 2013). The trajectory simulation
using this temperature data set is denoted as

Note that we only considered the vertical structure issue, since it is by far
a limiting factor in representing waves in the TTL. A large portion of the TTL
wave spectrum has horizontal and temporal scales much larger and longer than
reanalysis resolution; therefore, temperature behaves almost linearly
between model horizontal and temporal resolution. However, temperature
does not behave linearly in vertical space due to the fact that a significant
portion of TTL waves have vertical wavelengths shorter than

Cold-point temperature differences between MERRA adjusted
by waves and MERRA (MER-Twave–MER-T) during 2007–2013. The probability density function in black
is plotted on the left

Probability density functions of the differences between linear and cubic spline interpolations from the actual value form the GPS temperature profiles. Left: minimum saturation mixing ratio of the profile (units are percent per 0.1 ppmv); right: pressure of the saturation mixing ratio minimum (units are percent per 1 hPa). The plus signs in each line mark the bin intervals.

The wave scheme produces both positive and negative perturbations to the
MERRA temperature profiles, depending on the phase of waves. Overall, the
change in the temperature induced by waves is less than 2 K (Fig. 3), although
in rare cases it can reach 5–7 K. Importantly, however, about 80 % of
the changes in cold-point temperature are negative, with the wave scheme
lowering the average cold-point temperatures by

In our study, we included both GPS and MER-Twave data sets because they have
their own advantages and limitations. GPS provides sparse sampling in the
tropics (only

In our study, we use linear interpolation to estimate the temperature between the fixed levels of temperature data sets. However, some previous analyses have used higher-order interpolations, such as cubic spline (e.g., Liu et al., 2010), to make assumptions about the strong curvature of temperature profiles around the cold-point tropopause. In order to determine which approach is superior, we sample GPS tropical temperature profiles at MERRA vertical levels and then use the two interpolation schemes to reconstruct the full GPS resolution. Then we compare the minimum saturation mixing ratio from the recovered profiles to the minimum calculated from the full-resolution GPS profiles.

Figure 4 (left panel) shows the probability distribution of the differences between the
minimum saturation mixing ratio in the full-resolution GPS profile and in the
two interpolation schemes. On average, the linear interpolation performs
better (RMSD is 0.18 and 0.25 ppmv for the linear and cubic
spline, respectively). Figure 4 (right panel) shows the corresponding probability
distribution of the difference of the pressure of this minimum, and the
linear interpolation does better for this metric too (RMSD is 5.2
and 7.2 hPa for the linear and the cubic spline interpolation,
respectively). We have also tested higher-order spline interpolations and
find that they do not produce lower RMSE than linear interpolation. Overall,
the cubic spline interpolation tends to underestimate cold-point temperature,
making the implied H

Seasonal FDP vertical distributions (in percentage, first row
solid lines, lower

The gridded GPS temperatures have been available since July 2006, so for fair
comparison we start all trajectory runs at that time and run them forward
till the end of 2013. For each model run, we calculate statistics of the
final dehydration points (FDP) for all parcels entering the stratosphere. We
define “parcels entering the stratosphere” as parcels that underwent final
dehydration between 45

Figure 5a–c compare the FDP frequency (solid lines) and the FDP H

Vertical distributions of normalized FDP events in time-evolutional

The FDP frequency, however, shows large differences among three runs. The run
using MERRA temperature (

Note that at FDP, the coldest temperature encountered could be either at or in-between MERRA model levels, depending on the trajectory integration intervals. If we suppose our trajectory integration time step is on the order of seconds, then at some time steps, parcels would inevitably travel to each of the MERRA model levels, and therefore the encountered coldest temperatures would always be at one of the two levels in MERRA. In other words, the bimodal FDP distribution from MERRA run (Fig. 5a) could be even more peaked when choosing a smaller integration step in our trajectories. There are two reasons that we did not choose such smaller time step: (1) the wind and temperature data are only available 6-hourly or even daily resolution (GPS) so a much smaller time step introduces more uncertainties with more interpolation; and (2) considering the balance between model efficiency and computational resources.

Figure 6 depicts the vertical distributions of normalized FDP in time (panels a–c) and longitude (panels d–f) sectors for the three different runs. We see that the MERRA coarse model levels do not capture the variations of cold-point tropopause well during MAM and SON, resulting in discontinuous transition of FDP from DJF to MAM, and from JJA to SON (panel a). When using GPS temperatures (panel b) and MERRA temperatures adjusted to bear finer vertical structures (panel c), the dehydration patterns show continuous variations throughout the year. The bimodal feature is more emphasized in the longitudinal–vertical view (panel d), where we can also see that throughout the year the most frequent dehydrations occur over the western tropical Pacific region.

It is obvious that trajectory simulations using GPS temperatures
(

Figure 7a shows the tropical (18

Figure 8c also shows that compared to

It is important to point out that, despite these differences in the absolute
value of H

The dehydration of air entering the stratosphere largely depends on the
cold-point temperature around the tropopause. This may not be represented
accurately by reanalyses due to their relatively coarse vertical resolution
that reports coarser temperature vertical structure. To investigate the
impacts of this, we compare trajectory results from using standard MERRA
temperatures at coarse model levels (

Driven by the same MERRA circulation, with a 100 % saturation assumption we
find that on average

Looking at the locations of FDP, we find a bimodal distribution when using standard MERRA temperatures on model levels (Figs. 5–6). This is caused by the fact that the cold-point tropopause is constrained to be near the two MERRA model levels (100.5 and 85.4 hPa) that bracket the cold-point tropopause (Fig. 5d–f). When using the temperatures with finer vertical structures, the resultant FDP patterns appear to be more physically reasonable (Figs. 5a–c and 6).

In this paper, we perform linear interpolations for all trajectory runs. Other analyses have used cubic spline interpolation owing to the strong curvature of temperature profile around the cold-point tropopause. We investigate the performances of both schemes using GPS temperature profiles (Sect. 2.2.3) and find that while introducing new information due to its assumption in the temperature profile around the tropopause, the cubic spline scheme tends to generate unrealistically low cold-point temperatures due to cubic fitting. Therefore, the results are not necessarily realistic and, additionally, the linear interpolation is more accurate overall (Fig. 4).

It is well known that TTL temperatures regulate stratospheric humidity. In this paper, we have investigated one issue in our understanding of TTL temperatures – the effect of finer vertical structure in tropopause temperatures – and find that it is comparatively minor. This provides some confidence that the trajectory model driven by current modern reanalyses is capable of depicting the stratospheric water vapor accurately.

The authors thank Kenneth Bowman, Joan Alexander, Sun Wong, and Eric Jensen for their helpful discussions and comments. This work was supported by NSF AGS-1261948, NASA grant NNX13AK25G and NNX14AF15G, and partially by the NASA Aura Science Program. This work was partially carried out during visits of Tao Wang funded by the Graduate Student Visitor Program under the Advanced Study Program (ASP) at the National Center for Atmospheric Research (NCAR), which is operated by the University Corporation for Atmospheric Research, under the sponsorship of the National Science Foundation. Edited by: P. Jöckel