Introduction
Trace gases with a long chemical lifetime can serve as indicators for
middle atmospheric dynamics. In our study the focus is directed to
middle atmospheric water vapor and its distribution in the Northern
Hemisphere (NH). As outlined by , “the bulk of
evidence suggests that large-scale slow vertical ascent dominates mass
transport across the tropical tropopause, and that slow ascent is
required for effective dehydration”. A seasonal cycle in the amount
of dehydrated air, due to varying temperatures in the UTLS (upper
troposphere/lower stratosphere) region, leads to the so called tape
recorder effect . Water vapor in the lower
stratosphere of mid-latitudes can originate from moisture plumes above
severe thunderstorms during injection processes
. However chemical reactions like methane oxidation
are the major source of middle atmospheric water vapor. These
reactions happen in general below an altitude of 50 km
. In higher atmospheric regions the mean lifetime
of water vapor due to vertical transport and photochemical mechanisms
is similar and on the order of several weeks. As there is no other
major chemical source of H2O in the mesosphere, it serves as
an ideal tracer to study atmospheric dynamics
. Besides its chemical
characteristics, water vapor modifies the fluxes of incoming and
outgoing radiation in the atmosphere through absorption and emission
in the IR band. Another important issue concerns the chemical
interaction with ozone. Water vapor in the middle atmosphere is the
main source of the OH radical, which contributes to destruction
processes of the UV-protective stratospheric ozone layer. Therefore
having information about the distribution of water vapor is of high
scientific value.
Apart from “reverse domain filling” and “Kalman
filtering” especially “trajectory mapping”
is used for constructing trace gas maps,
validation and climatology studies from irregular (in time and space)
distributed profile measurements of either ground- or space-based
instruments. The idea of trajectory mapping is to create synoptic maps
by advecting measurements forward and backward in time using
a trajectory model that is driven by analyzed model wind
fields. Satellite data alone suffer from a poor global horizontal
resolution. With the above-mentioned methods, horizontal data gaps can
be reduced without spatial interpolation. Several investigations have
used this technique in the stratosphere, where O3 and
N2O/NOy are of primary importance
. Here we will make use of
this technique even in the mesosphere and studying water vapor in more
detail and its relation to dynamics.
In this study we demonstrate that the trajectory mapping (TM)
technique applied to ground-based water vapor profile measurements of
a small instrument network operated within the frame of NDACC (Network for the Detection of Atmospheric Composition Change) has the
ability to provide adequate information about the horizontal
distribution of water vapor, even during fast changing dynamic
conditions in the atmosphere (e.g., deformation of the stratospheric
polar vortex during a SSW (sudden stratospheric warming) event). A first approach uses a spatial
domain-filling TM technique according to . They used
the technique for studying global stratospheric ozone climatologies up
to 26 km altitude. The quality of our hemispheric
H2O volume mixing ratio (VMR) maps depends on how equally the
trajectory endpoints are distributed around the hemisphere in
a defined pressure layer. By increasing the thickness of a pressure
layer, it is possible to enhance the number of TM points and thus the
hemispheric data coverage, but the noise in the water vapor maps may
increase if vertical H2O gradients are large. For the
numerical calculation of the 3-dimensional trajectories, it is
important to have adequate 3-dimensional wind field data as
input. Wind field data sets with large errors may lead to uncontrolled
uncertainties of the exact locations of the trajectories. In fact,
found that trajectory position errors in space are
much more important than for instance tracer conservation errors. It
has been shown that 3-dimensional trajectories in the stratosphere
(e.g., as calculated by LAGRANTO) are more accurate in terms of water
vapor conservation than for example kinematic isentropic trajectory
calculations. In order to keep position errors small, highly accurate
meteorological data are needed. Operational analysis data from the
European Centre for Medium-Range Weather Forecasts (ECMWF) have been
processed to initialize the Lagrangian trajectory model LAGRANTO
. There might be concerns about trajectory qualities
above the troposphere in consideration of the question how well
high-altitude wind fields can be resolved in global numerical
models. Published work, e.g., that of , highlights
upgrades in the representation of stratospheric winds in operational
ECMWF analysis due to better assimilation schemes such as 4D-Var
. We also explore the suitability of mesospheric
trajectory calculations by creating water vapor maps for the
0.13–0.07 hPa pressure range.
The idea to use trajectory mapping of single water vapor profiles by
a network of instruments to obtain more information about the
horizontal distribution of H2O in the middle atmosphere is
not new. However, there are only few studies
that make use of trajectory
calculations (e.g., LAGRANTO) to study mesospheric
dynamics. applied the TM technique in order to
study the effect of a major SSW (sudden stratospheric warming) on the
horizontal distribution of water vapor VMR on pressure layers between
0.07–0.14 and 7–14 hPa. A simple contrasting
juxtaposition between raw (non-unified, non-interpolated) TM maps and
pressure layer averaged Aura MLS (Microwave Limb Sounder) measurements has been performed, but
a quantitative validation within superposable observations was not
applied. In our approach, Aura MLS measurements and the trajectory-mapped data are averaged in 3-dimensional domains; hence one
particular mean water vapor VMR value can be assigned to a TM- or MLS-related domain. Because the horizontal and vertical dimension as well
as the position of the TM and MLS domains is the same, a more valuable
direct comparison is achieved.
A serious problem to overcome is unknown biases between retrieved
H2O profiles from the mini network of five microwave
radiometers. Even though common instrument hardware features were
present, we expected a non-negligible bias in the water vapor VMR
between the different instruments. We calculate quasi-seasonal
correction factors, depending on the instrument location and height
above ground, by making use of NASA's EOS Aura MLS satellite
instrument as a kind of traveling standard. Such an approach has been
previously used in, e.g., . A more detailed
description of the procedure is outlined in the second part of the
paper (Sect. ).
Retrieved water vapor profiles from ground-based observations of
a network of five microwave radiometers located over the Northern
Hemisphere, where four of them are part of NDACC, are processed. The
instruments are MIAWARA (Middle Atmospheric Water Vapor Radiometer)
at Bern/Zimmerwald (Switzerland), SWARA (Seoul
Water Vapor Radiometer) at Seoul (South Korea) ,
WVMS4 (Water Vapor Millimeter-Wave Spectrometer)
at Table Mountain (California, USA) and WVMS6 at
Mauna Loa (Hawaii, USA). Additionally, data from the campaign-based
middle atmospheric water vapor radiometer (MIAWARA-C)
, gathered during the Sodankylä campaign at FMI
ARC (Arctic Research Centre of Finnish Meteorological Institute)
between June 2011 to March 2013, have been incorporated in this TM
survey. In the discussion of the results, it will turn out that
instrument locations at higher northern latitudes are mandatory to
resolve particular polar vortex structures by trajectory mapping.
The paper continues with the description of the trajectory mapping
method and construction of the hemispheric maps
(Sects. and ). In the
subsequent Sect. results and validations of four
chosen case scenarios for trajectory mapping in 2012 are
shown. Different seasonal times are covered to analyze special
seasonally caused effects in different dynamical regimes in
consideration of an existing middle atmospheric polar
vortex. Section is dedicated to a major SSW event
close to 17 January 2012, a difficult, but interesting test bed for
the TM method. A simple error estimation of the trajectory
mapping approach is provided in Sect. .
With Sect. a summary and discussion is addressed
and a short conclusion is presented in Sect. .
Data and methods
NDACC H2O microwave radiometer
network
Research stations all over the world contribute to NDACC and provide
high-quality long-term measurements of various atmospheric trace gases
in a standardized procedure. Identifying trends and changes in the
atmospheric composition, understanding their impacts and links to the
troposphere and middle atmosphere in the light of climate change are
among the most important tasks of this research compound. The NDACC
database is commonly used to validate space-based atmospheric
measurements e.g.,. In
our trajectory mapping investigation the different microwave
radiometers were cross validated against each other in a first attempt
by making use of the double differencing method, first introduced by
and applied to either satellite-to-satellite or
ground-based-to-ground-based observation validations in the study of
. The five ground-based remote sensing instruments,
listed in Table , measure the pressure broadened emission
line of water vapor molecules at a center frequency of
22.235 GHz . Water vapor profiles are
retrieved from measured spectra by radiative transfer calculations and
retrieval techniques such as the optimal estimation method
. Some specifications of the instruments
measurement techniques and details about the applied retrieval
versions of the H2O observations are provided in the next
paragraphs.
Locations of middle atmospheric water vapor radiometers used in this study with the geolocation and operational period of each station (OP). MIAWARA-C is a campaign instrument.
Station/instrument name
Latitude (∘N)
Longitude (∘E)
Altitude (m)
OP (yr)
Bern/MIAWARA
46.88
7.46
907
since 2002
Seoul/SWARA
37.54
127
52
since 2006
Mauna Loa/WVMS6
19.5
-155.4
3394
since 1996
Table Mountain/WVMS4
34.4
-117.7
2282
since 1993
Sodankylä/MIAWARA-C
67.37
26.63
190
2011–2013
The microwave radiometer MIAWARA was built in 2002 at the
Institute of Applied Physics (University of Bern) and has been continuously
operating on the roof of the building for Atmospheric Remote Sensing
in Zimmerwald close to Bern since September 2006. The vertical
resolution of the instrument varies between 11 km in the
stratosphere and 14 km in the mesosphere. A former
measurement range from approximately 7 to 0.1 hPa
could be extended to roughly
10 to 0.02 hPa with instrumental upgrades in spring 2007. An
acousto-optical spectrometer (AOS) was replaced by a digital FFT (fast
Fourier transform) spectrometer that improved the spectral resolution
from 600 to 61 kHz. Tropospheric opacity due to weather
conditions can affect the temporal resolution. With low optical depths
in the sensitive frequency region of the radiometer during dry and
cold tropospheric conditions, temporal resolutions on the order of few
hours are achievable. But temporal resolutions up to 12 h or more
are likely when warm and humid periods occur. The MIAWARA profile
retrievals used for the trajectory mapping investigation have
a constant time resolution (integration time) of 12 h, a total
bandwidth of 225 MHz and are processed with Aura MLS v3.3
observation data to initialize pressure, temperature and geopotential
height as PTZ source.
The campaign-based version of MIAWARA, MIAWARA-C
, was operated in Sodankylä at FMI ARC
(67.37∘ N/26.63∘ E, Finland) from
June 2011 to March 2013 with practically no interruptions. Because of
an almost doubled FFT spectrometer resolution of 30.5 kHz
the upper measurement limit reaches 0.015 hPa
(78 km). At best, a lower measurement limit of
35km≈7hPa can be achieved. Vertical
resolution varies from 12 to 15km . In
this study we process a MIAWARA-C retrieval version with fixed
temporal resolution of 12 h, a bandwidth of 80MHz and Aura
MLS v3.3 data as PTZ source. Even though temporal integrations
of less than 2 h are feasible for retrieval calculations, we prefer
a constant 12 h integration time due to better signal to noise
ratios.
The microwave radiometer SWARA was developed, like MIAWARA and MIAWARA-C, at the
Institute of Applied Physics at the University of Bern and has been
operational since October 2006 at the Sookmyung Women's University of
Seoul in South Korea . SWARA is in principle a copy
of MIAWARA and the same specifications apply. However, as the wings of
the spectrum are affected by baseline ripples a retrieval of water
vapor at altitudes below 38km with a reasonable
(>60%) measurement response is limited (50MHz
retrieval bandwidth). For that reason data below 4hPa are
not used. The SWARA H2O profiles for TM have a fixed temporal
resolution with an integration time of the calibrated spectrum (Level
1 data) of 24 h. The same PTZ information source (Aura
MLS v3.3) as for MIAWARA applies.
Two further instruments from NDACC are used. Specifically, we make use
of measurements from the ground-based Water Vapor Millimeter-wave
Spectrometer (WVMS6) at Mauna Loa (HI, USA), and from
WVMS4 at Table Mountain (CA, USA). In this study we will make
use of data from both of these instruments up to 68km
(0.05hPa). showed that retrievals
from the WVMS4 instrument down to 26km were in good
agreement with satellite measurements, and we will use these
retrievals down to 10hPa. The WVMS6 instrument now
continues the water vapor record at Mauna Loa (since March 2011)
previously recorded by the WVMS3 instrument. WVMS6 retrievals have
not been validated below 40km, and we will restrict their
use here to the range 5–0.05hPa. These instruments have
a vertical resolution of around 15km and a measurement
error of about 9%. The WVMS data are provided in the
downloadable NDACC files (NASA Ames Format for Data Exchange) as daily
averages between midnight and midnight in local time and are linked to
an altitude grid with a spacing of 2km. Dependent on the
altitude region, temperature and pressure for the retrieval calculations
come either from a MLS climatology (upper stratosphere and mesosphere),
NCEP (National Centers for Environmental Prediction) data
(lower stratosphere) or MSIS (Mass Spectrometer Incoherent Scatter Radar)
model (thermosphere).
In principle, a 6-hour temporal resolution of the profile data (to
fit the time spacing of the meteorological input fields of the ECMWF
analysis data) would be optimal for trajectory mapping in the proper
sense of unifying the data sets with fewer interpolation steps. Due to
different locations and altitudes of the instruments (different
climatological conditions) such a uniform and high temporal data
resolution is not realizable. To resolve fast changing water vapor
distributions associated with polar vortex movements adequately by
trajectory mapping, at least a 24 h temporal resolution of the
retrieved H2O profiles should be used.
The vertical H2O a priori profile information needed in the retrieval
calculations of the instruments MIAWARA, MIAWARA-C and SWARA is based on the
same climatology. The a priori is taken from a monthly mean zonal mean
climatology using Aura MLS v2.2 data between 2004 and 2008. The NDACC
retrievals of the instruments in Hawaii (WVMS6) and on Table Mountain (WVMS4)
are all also run with an Aura MLS based climatology as a priori. Particularly, it
is based on v3.3 data taken from August 2004 to March 2011 within
±2∘ latitude and ±30∘ longitude of each observation
site. For each day of the year, the data are averaged over ±5 days.
Summer season (NH) mean relative differences (black lines)
of the water vapor difference profiles between five ground-based
water vapor radiometer stations (panels from left to right: (1) Bern, (2) Seoul, (3) Mauna Loa, (4) Table Mountain, (5) Sodankylä)
and Aura MLS. The mean difference profiles are calculated from
measurements in the period from August 2010 to September 2014,
considering the following months: April, May, June, July, August,
September. The number of individual profiles considered per
instrument is represented by n. Dashed red lines show the SD ±σ of all n differences and the yellow stripes depict the
±10 % areas. Data below pressure levels indicated by
the horizontal black lines (grey areas) were not used due to overly
high a priori contributions in the water vapor retrievals.
Same as Fig. but for the winter season
(NH), considering the following months in the period from
August 2010 to September 2014: October, November, December,
January, February, March.
In the following more details about the a priori contribution or
measurement response of the individual instruments
are given. Different features observed in the measurement responses
between 0.05–10hPa resulted in adjusted data omissions
to keep the a priori influence on the H2O retrievals small.
This is of special importance since the trajectory-mapped data are
compared to Aura MLS for validation. At the Mauna Loa observation site
(WVMS6) the validation of the data variations down into the
lower stratosphere is still missing, which is why it is grayed out
in Figs. and and data
from below 5hPa are not used in the four TM case studies.
The instruments providing measurements down to 10hPa
(MIAWARA and WVMS4) have a priori contributions of less
than 25% (20% for WVMS4) at this level.
In case of SWARA and MIAWARA-C there is a transition from
2 to 4hPa, where the a priori contribution drops from
∼50% (at 4hPa) to ∼20%
(at 2hPa). At higher altitudes the measurement responses
are widespread above 80%, considering data from 2012.
Accordingly, data from SWARA and MIAWARA-C are only used between
0.05–4hPa as indicated in Figs.
and . It is stated that SWARA H2O
retrievals sometimes show a priori contributions up to 40%
between 0.05–0.3hPa in the summer months,
which is tolerable.
We conclude that the fact that Aura MLS H2O climatologies
serve as a priori profiles in the five ground-based instrument
retrievals, which are compared to MLS data after application of our TM
method, is of minor relevance as we do account for bad profile
sections and therefore confine the comparison of TM data
where the contribution of the a priori is most of the time low.
Data harmonization – satellites as traveling standard
Although several significant studies have been performed to validate
ground-based water vapor microwave radiometer instruments and to find
biases relative to space-borne measurements
,
a harmonization is still strongly needed for combining middle
atmospheric H2O VMR measurements from different
instruments in view of our specific purpose.
The water vapor product from Aura MLS is retrieved from radiance
emission measurements near a center frequency of 183GHz as
outlined by . The gathered data are very valuable to
our trajectory mapping investigation because of the near-global
(82∘S to 82∘N) coverage and
daily observations above all five ground-based microwave radiometer
locations (see Table ) within our defined horizontal
coincident displacement criteria of 800km (E/W) ×400km (N/S). Besides the water vapor product we also refer
to temperature observations when we discuss the January 2012 sudden
stratospheric warming event in Sect. . In this study
we use the v3.3 data product which shows substantial improvements
compared to the previous v2.2 product by getting rid of small scale
vertical variations in the water vapor retrievals
. The whole vertical range is between 316 and
0.002hPa, and the accuracy varies roughly between 4 and
11% in the 10–0.01hPa pressure
regime. A 3.2 to 10km vertical resolution again
in-between 10–0.01hPa goes along with a spatial wide
horizontal resolution in the range of 300 to 680km. We
use MLS data if according to the data quality documentation
quality thresholds are preserved within the
addressed vertical range.
The double differencing method is a useful technique in the case of
a two-instrument network to obtain the bias between instruments by
making use of one traveling satellite, which observes the same air
columns above the ground-based observation sites. A larger network, such as
ours, requires a different strategy. Mean relative difference profiles
are calculated for every single ground-based instrument to Aura MLS in
a quasi-seasonal (6-month periods) manner to take the bias into account. All individual profiles are linearly interpolated in logarithmic
pressure to cover the same vertical extent (10–0.05hPa)
with the identical number of 1000 grid points, which serve later as
trajectory starting points.
The mean relative difference profiles (Figs. and
) reveal that nearly all relative H2O
deviations are within ±15%. An exception is WVMS6
located in Hawaii with deviations exceeding 15%. As the
yellow bands indicate, the instruments at Bern, Seoul, Table Mountain
and Sodankylä have biases of less than 10% to MLS in
most of the regarded altitude ranges. During both the April to
September and the October to March periods, the mean difference water
vapor profiles of all measurement sites show an abrupt increase of
5–10% difference to MLS from around 2 to
1hPa. Despite small variations in the seasonal behavior of
the SD (standard deviation) σ the profile structures stay
comparable. The largest uncertainty (σ≈20%)
in the mean difference can be assigned to MIAWARA-C observations
above the polar winter stratopause. Several hundreds of profiles
are processed in the time period from August 2010 until September 2014
to calculate the mean difference H2O profiles, serving as
height-dependent correction factors to harmonize the trajectory-mapped H2O VMR values during the synthesis of hemispheric
maps (see Sect. ).
In summary, all averaged ground-based measurements of the five instrument
locations result in a negative bias to Aura MLS throughout all of the studied altitudes; i.e., the water vapor mixing ratio measured by MLS
is consistently higher than that measured by ground-based instruments.
The trajectory model
Numerical trajectory simulations in the middle atmosphere were
performed with LAGRANTO , a software tool consisting
of UNIX shell scripts and FORTRAN programs to analyze Lagrangian
aspects of atmospheric phenomena. The program requires a time series
of 3-dimensional wind fields in NetCDF files. Possible error sources
are interpolation steps and uncertainties in meteorological input
data. Interpolation errors develop when the model wind field from the
model time and grid resolution is interpolated to an actual trajectory
location in the 4-dimensional continuum. With respect to the true
atmospheric state, errors will remain in the initial model conditions,
limiting the accuracy of the wind fields.
The 3-dimensional wind vector data are from the European Centre for
Medium-Range Weather Forecasts (ECMWF). A daily model run (cycle 37R3
– T1279) provides meteorological data sets on a 6 hourly spaced time
interval from midnight to midnight. The operational model analysis
provides 91 vertical model levels from the surface up to
0.02hPa. A regular latitude/longitude grid with
a resolution of 1.125∘×1.125∘ is used for the
horizontal plane. created middle atmospheric
H2O trajectory maps of synoptic scale also with 6 hourly
ECMWF data, but at a higher horizontal resolution of 0.5∘×0.5∘. Our results suggest that the horizontal model
resolution is not a key factor and more accurate trajectory maps can
be produced with a lower LAT/LON resolution, even during dynamical
extreme events. In general, errors connected to wind field
interpolations are of significant relevance. According to
errors (e.g., gridded ECMWF model variables) related
to spatial interpolations are much smaller than those related to
temporal interpolations, which are the primary limiting factor of the
accuracy of trajectories. The interpolation of the horizontal wind
components is less error-prone than for the vertical motion
w. described in their study that a doubling of the
temporal model resolution from 3 to 6 h can result in up to
40% mean relative interpolation errors of the vertical
wind component.
Most modern trajectory models use the second-order iterative
Petterssen scheme in their dynamical core for
solving the trajectory (Eq. ).
DxDt=u(x,t).
The Petterssen scheme (Eq. ) has a truncation (numerical
dispersion) error proportional to Δt2 where Δt is
the numerical time step. This error occurs, when higher order terms in
the Taylor expansion are neglected. In Eq. () x0
describes the initial position vector, whereas x1,
xn correspond to the positions after 1 respectively n
iteration steps.
x1=x0+Δt⋅u(x0,t)xn=x0+12Δt[u(x0,t)+u(xn-1,t+Δt)].
The external velocity field u(x,t) and a second-order
semi-implicit discretization in time and space are needed to compute
the future trajectory position. If the Courant–Friedrichs–Lewy (CFL)
criterion (c=u⋅Δt/Δx<1) is
fulfilled, the computed solution is numerically convergent. Obviously,
the Courant number c depends on the numerical time step Δt
and the wind field variable u. Earlier simulations performed
by revealed that a time step of around
300s is short enough for calculations with the ECMWF
operational data set. There, observations of two ground-based water
vapor microwave radiometers were used for trajectory mapping and it
was possible to find the approximate location and extension of the
stratospheric polar vortex with the obtained H2O
distribution. But irregularities in the measurements and a sparsely
distributed observation network could not match the quality of
synoptic maps from satellite observations.
The applied trajectory mapping method uses the following
assumption. It is assumed that an air parcel's water vapor volume
mixing ratio stays constant while moving along a 3-dimensional 10-day
trajectory. Hence, turbulent mixing, photolysis, chemical
reactions and phase changes of H2O are not taken into
account. as well as already
showed with trajectory studies that a time period of 10 days for
forward or backward trajectories is a reasonable timescale in the
stratosphere. We will make use of 10-day trajectories up to
0.05hPa.
Trajectory mapping – synthesis of hemispheric H2O
maps
LAGRANTO initializes trajectory calculations with starting points of
an air parcel in a latitude, longitude and pressure level coordinate
system. To generate synoptic maps with trajectory mapping, the
definition of a pressure layer with a certain thickness Δp is
required. In order to increase the number of trajectory arrival points
from instrument observations (trajectory starting points) inside
a defined pressure layer, different implementations such as increasing
the TM pressure layer thickness, the number of vertical trajectory
starting points with interpolation or an extended instrument network
might be considered. Keeping in mind that a maximal vertical
measurement resolution of ∼10km is realistic, the
number of vertical starting points can only be enlarged by
interpolation of the water vapor profiles. The largest uncertainties
during the LAGRANTO calculations are likely arising from the
interpolation of the vertical wind component. It is important that the
starting points of the trajectory calculations in LAGRANTO are within
the ECMWF vertical model grid (91 levels). Otherwise the
interpolations to the starting point positions are
impossible. Horizontal interpolations of the ECMWF wind field data to
the actual trajectory positions are bilinear; vertical ones linear
with pressure and time interpolations are also performed linearly.
The water vapor profiles of the instrument network are interpolated to
a pressure grid with a logarithmic equispaced subdivision of 1000
pressure levels between 10–0.05hPa. This is equal to
a ∼37m vertical grid point spacing. According to
instrument features and retrieval versions different altitude data
cut-off limits in the stratosphere are used as described in
Sect. and illustrated with the bias correction
plots of Figs. and . The
interpolated volume mixing ratios on the grid points are used to
create raw trajectory maps. Altogether 20 days of water vapor
profile measurements from each instrument are taken. As the temporal
resolution varies between 12 and 24 h, one or two profiles per day
are obtained accordingly. A mean time is assigned to each profile,
which results in a H2O profile time series. Because the
wind field data of the ECMWF model that go into LAGRANTO have temporal
resolution of 6 h, we further temporally interpolate the prior
pressure interpolated profile time series of every single radiometer
onto the time grid of the ECMWF model.
Now trajectories can be calculated every 6 h starting from every
grid point of the processed profiles. For 20 days and with four profiles per day plus the initial profile on the TM target time (4⋅20)+1=81 profiles are created over one ground-based
instrument location. If measurement gaps extend over more than 96 h
the mixing ratios of the corresponding interpolated water vapor
profiles are disregarded. The remaining data are needed to synthesize
a trajectory map at the center of the 20-day time period of
considered measurements. Forward trajectories are calculated for the
first 40 water vapor profiles of each measurement site and the
corresponding backward trajectories for the last 40 profiles. The
profile number 41 is already at the right position in time for the
trajectory map. If all trajectories are summed for a 20-day period
and five ground-based stations, we count 4×105
trajectories.
The volume mixing ratios from the grid points of the profiles
belonging to the trajectory start points were assigned to the new
calculated trajectory end points, assuming that the H2O
mixing ratio stays constant. While we are calculating 3-dimensional
paths through the atmosphere, the trajectory end points can rise or
sink in altitude. A simple filtering is done to separate out the
mixing ratios of points within the different defined pressure layers
(12–8, 3.5–2.5, 1.3–0.7 and 0.13–0.07hPa)
in order to get a simple raw TM map, consisting of single points in
3-dimensional space, at 12:00 UTC. A problem with a thick
pressure layer can be that a less homogeneous trajectory map is
produced if large vertical gradients in H2O exist. The
trajectory origin of the single points can be outside the previous
defined pressure layers due to rising or descending
trajectories. Later in Sect. we only refer to the
middle of the pressure layers (10, 3, 1 and
0.1hPa). The chosen layers ensure that at least one MLS
measurement on the native vertical resolution is situated at or close
to the middle of a layer.
On the basis of raw trajectory maps alone, a quantitative verification
to other measurements would be difficult. With the idea of uniquely
defined 3-dimensional domains, where the trajectory-mapped
H2O VMR data are averaged, a proper solution for following
verifications was found. Depending on whether domains are in the
stratosphere or mesosphere, the horizontal expansion varies from
2.5∘×2.5∘ to 5∘×5∘
(LAT×LON) in the domain-averaged TM maps. With
a doubling of the horizontal domain-averaging size at the stratopause
region and above, we accounted for the altitude increasing uncertainty
of middle atmospheric ECMWF wind fields by an increased blurring of
trajectory endpoint positions. The vertical extent is in agreement
with the previous mentioned pressure layer thickness. The horizontal
scale of the domain averaging is in general smaller than the
characteristic correlation length of water vapor in the middle
atmosphere. A short time series (several months in 2012) of MIAWARA
water vapor profiles has been processed with an auto-correlation
function (ACF) according to in order to compute
characteristic correlation times. A timescale between 1–3 days
is necessary to decrease the correlation coefficient by a factor of
e. Assuming a mean horizontal wind velocity of
10ms-1 a correlation length scale on the order of
860–2600km results. came up with similar
correlation time (length) scales for ozone in the stratosphere with
2–6 days (1000–2000km).
Due to averaging TM data inside the domains, some noise in the water
vapor maps is reduced. On the other hand, as an example, a vertical
averaging of H2O within the defined domains can produce
a bias due to the fact that a point A, say at 3.5hPa, will
have a mixing ratio lower than point B, say at 2.5hPa, just
because it is lower in the atmosphere. We correct for most of the bias
in this by subtracting the a priori vertical profile from the
measurements prior to the trajectory calculations and then adding the
a priori for the middle of the pressure layers at the end.
In addition, we plot the edges of the polar vortex at the lower and
upper limit of the current pressure layer. We adapt the definition and
calculation of the vortex edge from , which is
effective from 10 to 0.01hPa and uses the highest
absolute wind speed along a geopotential height (GPH) contour
(averaged) together with a minimal border length of
15×103km on a pressure level. Additionally the GPH
contour has to enclose a low-pressure system and must be everywhere
north of 15∘N.
Case A: stratospheric H2O VMR
[ppm] within 12 to 8hPa (a–c),
3.5 to 2.5hPa (e–g), respectively
mesospheric H2O VMR within 1.3 to 0.7hPa
(i–k), 0.13 to 0.07hPa (m–o) on 28
February 2012 12:00 UTC. Harmonized trajectory-mapped ground-based
measurements from five stations (black circles) corresponding to
Bern, Seoul, Hawaii, Table Mountain and Sodankylä are displayed
in raw (first column), with 2.5∘×2.5∘ (first
and second row) or 5∘×5∘ (third and fourth row)
domain-averaged format (second column). Pressure layer and
domain-averaged Aura MLS v3.3 observations are shown in the third
column. Relative difference maps of coincident domains between TM
and Aura MLS data are shown in the fourth column. All charts
indicate the position and vertical displacement of the polar vortex
edge within the given pressure layers by the solid black (upper p
limit) and dashed black (lower p limit) lines. Grey areas
indicate no data coverage.
Case B: same as Fig. , except for 22
September 2012, 12:00 UTC close to the fall equinox.
Case C: same as Fig. , except for 21 June 2012, 12:00 UTC close to the June solstice.
For all four TM case studies related domain-averaged Aura MLS water
vapor maps were produced for comparison. The domains in the MLS
observation plots coincide with the TM domains, because of their
global definition. Aura MLS measurements of a whole hemisphere cannot
show one particular point in time. To gather all information needed
along the orbit tracks, 1 day passes by. We account for the
different time offsets between trajectory-mapped data and Aura MLS
measurements by linearly interpolating the domain-averaged MLS data
along the orbit track onto the trajectory mapping target time by using
the measurements of the previous and the following day. The domain
sizes of MLS and TM are identical. Incidentally matching domains and
their H2O volume mixing ratio Q are directly compared in
relative difference maps (e.g., fourth columns in
Figs. to and
). The relative difference Drel is
calculated according the following Eq. ().
Drel=QTM-QMLSQMLS.
Results
In this section we present our results for a total of four H2O
trajectory mapping case studies in 2012. We abbreviate
the case studies with capital letters from A to D. The days of two
case scenarios (A and D) belong to northern hemispheric winter time,
where polar vortex structures were formed and non-zonal water vapor
distribution occurred (Sects. and
). With case study D (17 January 2012), one particular
selected date related to a sudden stratospheric warming is included,
where the zonal mean temperature at 10hPa increased by more
than 25K just north of 60∘N in a few days
and the prevailing westerlies changed direction to become
easterlies. All conditions of a major SSW
cf. were fulfilled. An idealized ECMWF wind
modification case study D⋆ is included, trying to
improve trajectory positions in the mesosphere on 17 January 2012.
With case B, a trajectory mapping day on the fall equinox of 2012 (9 September) is investigated. Performing trajectory calculations around
the equinox is particularly interesting because the zonal wind
direction in the stratosphere and mesosphere reverses to the winter
westerlies and the polar vortex forms again.
Case study C in Sect. shows the performance of
TM on a northern hemispheric summer day (21 June 2012) without a polar
vortex. The transport of atmospheric constituents over the
ground-based instrument locations is mainly governed by zonal
winds. In summer the prevailing advection route is from east to west
in the stratosphere and mesosphere.
TM in polar vortex regimes – case study A and B
Northern hemispheric Aura MLS v3.3 zonal mean temperature in
[K] (upper panel) and ECMWF zonal mean zonal wind in
[ms-1] for the 10hPa pressure level. The
time period extends from mid-December 2011 to mid-March 2012. The
left dashed black line indicates the date of TM case D and the right
one TM case A.
Figure shows TM case study A with all
investigated altitude ranges (from top to bottom), as will be the case in
Figs. , and
. The positions of the polar vortex edges within
the altitude range of the respective plot are indicated by solid
(lower altitude limits) and dashed (upper altitude limits) black
lines. At the vortex edges high gradients in potential vorticity are
present and the mean zonal wind speeds become fastest. A typical just
slightly disturbed polar vortex pattern centered around the North Pole
is visible in the stratosphere. Looking at the relative positions of
the ground-based observatories (black circles) with respect to the
vortex edge at that time (28 February 2012, 12:00 UTC), it is
obvious that MIAWARA-C (Sodankylä) is situated inside the vortex,
whereas the four other instruments (Table ) are
outside. Referred to the 3hPa pressure level, Bern and
Table Mountain are rather close to the comma-shaped vortex edge and
Seoul lies furthest away. The vertical displacement of the polar
vortex is marginal and mostly on the order of one domain size
(2.5∘/5∘). Isentropic tracers like water vapor tend
to stay trapped inside the polar vortex system, which has a limited
air-mass mixing across the edge . Trajectory-mapped H2O on 3hPa is found to be approximately
1ppm lower inside the plotted stratospheric vortex compared
to Aura MLS water vapor maps. The vortex edge can qualitatively be
determined from the TM trace gas distribution on 3hPa and
from the data gap shape on 10hPa, but not as clearly as
from Aura MLS H2O observations. The data gap inside the
10hPa vortex is due to the cut-off criterion for the
MIAWARA-C profiles with an overly high a priori contribution. More than
one observing site inside such a large and isolated quasi-zonal
stratospheric wind systems would be advantageous in determining more
precisely borders of air masses.
The nature of satellite observations has the advantage of more
uniformly covered measurements around the globe which TM sometimes
cannot provide; rather, randomly spread data points and various data
gap sizes occur, especially in the stratosphere. For instance,
a larger H2O data gap (beside the vortex area) extends
between Alaska and China just south of the stratospheric vortex as
Fig. b reveals. The coverage over Europe is in
contrast quite good. With increasing altitude, the water vapor
observations from Sodankylä almost reach the inner mesospheric
vortex edge in case A (28 February 2012), but data from MIAWARA-C
still mainly contribute for the part of the map which is framed by the
vortex. Contrarily in the two higher altitude regions
(Fig. j and n) the hemispherical coverage of water
vapor data is much better in the synthesized domain-averaged TM map
than in the Aura MLS map (Fig. k and o). A main
advantage of the synthesized water vapor maps is a temporal coherent
data set without any post-processing, covering over 95%
of the Northern Hemisphere on the 1 and 0.1hPa pressure
level. The area covered by TM data is remarkable and similar to that
in the TM case scenario B for the 0.1hPa and D for the 1
and 0.1hPa pressure level (see Figs. n
and j and n). The minimum H2O mixing
ratios inside the 1hPa (stratopause) and 0.1hPa
vortices coincide very well. An eastward shift in the position of the
lowest VMR domains is apparent between the Aura MLS and TM water vapor
maps for the stratopause region. A mesospheric vortex position at
0.1hPa can be seen in both TM and direct satellite
observation techniques (Fig. n and
o). A perceptible difference between Fig. m and n
in areas with low water vapor is likely due to the applied correction
for vertical averaging in the domain TM map
(cf. Sect. ).
TM case B is represented in Fig. . The trajectory
mapping target time was 22 September 2012, 12:00 UTC, the second day
of equinox in the year 2012. A rapid increase in planetary wave mode 1
usually occurs near the fall equinox in the Northern Hemisphere, and
their propagated and transferred momentum drives the zonal west wind
circulation in the middle atmosphere . Regarding the
considered pressure layers, inside-the-vortex measurements from the
ground-based instrument network were always available on the TM target
date, if the measurement response down to 10hPa was high
enough. Nevertheless there is the possibility that subsiding air from
higher altitudes can provide TM data down to lower altitudes. The
polar vortex edge detection algorithm worked for the equinox scenario
(B) and the reformed vortex after the summer season. For the
mesospheric vortex (see Fig. m), a large vertical
gradient of the edge is evident, where a 3-dimensional interpretation
would be cone-shaped.
The hemispheric coverage of the TM H2O VMR data of case
study B in the stratosphere and on stratopause level
(Fig. a, e and i) is not as good as for the
previous case study A, but similar for the 0.1hPa pressure
level. Some larger measurement gaps between 12 September and 2 October 2012 in the Table Mountain and Hawaii data lead to a reduced
number of the TM values of water vapor in case B. The visual impression
of the comparison between Fig. f and
g gives the result that TM can reproduce the H2O VMR within a few percent of relative
difference along the inner black vortex contours, as the relative
difference maps to Aura MLS observations (Fig. h)
prove. Horizontal meridional H2O VMR gradients match well
between Aura MLS and TM maps for the 3 and 1hPa
altitude. At 0.1hPa trajectory-mapped water vapor tends to
be too low on the order of 5–10%, as is obvious in
Fig. p, where bluish domains are prevalent. In
general, horizontal gradients in water vapor were found to be higher in
the stratosphere than in the mesosphere near equinox.
To sum up so far, Aura MLS water vapor observations show a quite good
and widespread agreement over the analyzed pressure levels of case
scenarios A and B in relation to the generated TM maps, which have
more noise (variability between neighboring domains) in water
vapor. Reducing the noise of TM data without smoothing algorithms is
difficult. More accurate wind field data with a higher temporal
resolution could improve the TM quality in the sense of decreasing
noise in the horizontal water vapor distribution and trajectory
position errors.
TM in a non-polar vortex regime – case study C
The prevailing wind direction in the middle atmosphere reverses seasonally,
in winter the winds are mainly eastward and in summer westward. Enhanced
gravity wave activity during NH winter leads to a deposition of angular
momentum in the middle atmosphere and decelerates the zonal flow. Meridional
transport of atmospheric constituents in summer is limited, because of
missing wave-induced forces driving north–south circulations
. As a consequence, trace constituents in the NH summer
strato- and mesosphere become fairly evenly distributed around latitudinal
bands within weeks because there are no dynamical barriers to atmospheric
transport as provided by the wintertime stratospheric polar vortex. Circle
like structures in the TM water vapor maps of Fig. ,
close to the June solstice, exemplify the situation. With only five
measurement locations, whereof two (Seoul and Table Mountain) are almost at
the same latitude, the hemispheric coverage compared to Aura MLS is poor. If
we intend to reduce the water vapor gaps in summer time, ground-based
observations of more latitudes from tropical to polar regions would be
necessary. Stations located at various longitudes are much more important in
the winter than in the summer period. A spatial interpolation of
H2O in environments where the horizontal gradient is small
(Fig. c and g) or even absent
(Fig. k) could be taken into consideration to fill in
the gaps. For filling in data gaps in the satellite observational record,
analysis or reanalysis data from e.g., ECMWF could be used. Thus an increased
number of possible comparison domains would be created. This concept has not
been implemented, because the accuracy of moisture fields in ECMWF model could
be problematic in the upper atmosphere. Some studies e.g.,
found that the ECMWF model produces an unrealistically moist mesosphere, which is
not present in the MLS observations. And more importantly, there is no stratospheric
H2O data assimilated in the ECMWF integrated forecasting system (IFS),
and we think that using the model data above the troposphere is not an alternative
regarding the TM map validation.
In the stratosphere of case study C the difference between the highest
(high LAT) and lowest (low LAT) mixing ratios is on the
order of 1ppm, what is confirmed by TM
(see Fig. b and f). The 10hPa trajectory map
shows that WVMS6 and MIAWARA-C instruments are not able to provide valuable
scientific information for this pressure layer (12–8hPa).
The H2O VMR distribution at the 1 and 0.1hPa
level is quite uniform and reaches values between 7 and
8ppm as 3 months later in the case of scenario B. With
increasing altitude, the water vapor coverage over the Northern
Hemisphere is found to increase in the TM plots
(Fig. a, e, i and m). The short-scale
H2O variation (noise) is obvious in the Aura MLS map in
Fig. o.
Performance during the January 2012 SSW – case study D
Case D: same as Fig. , except for 17
January 2012, 12:00 UTC close to the maximum temperature increase
at 10hPa during a major SSW in the Northern Hemisphere.
We restrict the temporal description of the major SSW of January 2012
to Aura MLS zonal mean temperature measurements and ECMWF zonal mean
zonal winds on the 10hPa pressure
surface. Figure shows a strong negative temperature
gradient north of 40∘N right before the end of
December 2011 in connection with the stratospheric polar vortex. At
the beginning of January 2012, the temperature gradient started to weaken
and reversed near the middle of the month. The increase in zonal mean
temperature in the polar stratosphere is about 25K during
1 week. In the same period of time, the ECMWF operational analysis
of the mean zonal wind component shows a reversal from westerly to
easterly winds. Compared to the January 2010 major SSW, described in
, the temperature increase in this case was not so
abrupt and also the duration of the easterlies at 10hPa did
not persist as long. After the SSW the temperatures decreased again in
the stratosphere north of 45∘N, but did not reach as
low values as in December 2011 (∼205K compared to
190K) owing to a less intense reformation of the polar
vortex.
Case D (Fig. ) occurs near the time of the maximum
temperature observed on 10hPa during the 2012 SSW. The
distortion and weakening of the vortex is a difficult situation for
applying trajectory mapping. By comparing the position of the vortex
edge contours between 10 and 3hPa, big differences in
size and position can be found. The mean vortex edge horizontal wind
velocities decline by almost 10ms-1 when going up in
altitude from 10 to 3hPa. Usually, in an undisturbed and
stable circulation environment in NH winters the opposite (increasing
wind speeds with altitude) is typical in the stratosphere. Regarding
the trajectory map and MLS H2O footprints on the
10hPa reference layer a good match is found
(cf. Fig. b–d as well as
f–h).
Trajectory mapping of water vapor in the mesosphere during a SSW
provides synoptic maps of slightly reduced quality and higher errors
(Fig. n and p). However in
this instance it is found to work even better than in case A with
a stable polar vortex environment. To perfectly match the noisy Aura
MLS map (Fig. o) is by chance very unlikely. The
correlation between these differences (TM-MLS) and
photolytic or chemical processes, which are entirely neglected, is not
evaluated. We note that the ECMWF winds become increasingly uncertain
with increasing altitude, and may contribute significantly to the
observed differences between MLS and the ground-based radiometers.
Pressure layer corresponding histograms of TM case study
A (28 February 2012, first column), B (22 September 2012, second
column), C (21 June 2012, third column) and D (17 January 2012,
fourth column). The number of relative difference (TM-MLS) domains in a certain deviation bin of a width of
5% between TM and Aura MLS solution is shown. From
top to bottom, the pressure layers are 12–8, 3.5–2.5,
1.3–0.7 and 0.13–0.07hPa. Vertical green lines
indicate zero deviation.
Modification of ECMWF mesospheric wind velocities – case
study D⋆
Additionally we want to briefly describe a performed sensitivity study
with the ECMWF wind field data, that might improve mesospheric
trajectory positions. Based on measurements from the ground-based
microwave Doppler wind radiometer WIRA
during different campaigns, which suggests that ECMWF wind components
(u, v and w) are overestimated in the mesosphere (above
1hPa) by the model, a constant downscaling of u, v and
w by 30% on all model grid points above
1hPa has been performed and new trajectories were
calculated with LAGRANTO for the SSW case study D to synthesize new
H2O trajectory maps (case study D⋆).
Compared to the Aura MLS water vapor maps a non-significant
improvement in the 0.13–0.07hPa pressure layer could be
detected (cf. last two columns in Table ).
Less than 4% more coincident relative comparison
domains display a difference of up to ±10 %
to Aura MLS H2O VMR domains. From this point of view it
is not possible to cross-check and prove whether the ECMWF winds
in the mesosphere are indeed too high. An assessment for effects
of potential error sources in the TM analysis, such as wind errors,
chemical reactions or a removal of water vapor by phase transitions
(e.g., mesospheric clouds), is presented in
Sect. .
The percentage of coincident domains in which the H2O from TM and Aura
MLS observations agree within 10 (20)% in each pressure layer. All four studied TM
scenarios (A–D) with applied bias correction (Y), according to Sect. , or no correction
(N) are shown. In the last column the percentages regarding the ECMWF sensitivity
case study D⋆ (30% downscaled mesospheric wind velocities for TM) are given.
TM Case
A
B
C
D
D⋆
Date
28 Feb 2012
22 Sep 2012
21 Jun 2012
17 Jan 2012
17 Jan 2012
12–8hPa (N)
77.6(99.1)
56.9(96.1)
84.9(100)
70.9(93.9)
70.9(93.9)
12–8hPa (Y)
82.6(100)
64.7(98.0)
89.0(100)
64.0(98.5)
64.0(98.5)
3.5–2.5hPa (N)
72.5(96.8)
87.0(99.7)
59.2(93.7)
56.4(96.1)
57.9(95.6)
3.5–2.5hPa (Y)
83.1(98.4)
86.0(99.1)
92.7(99.1)
91.1(99.3)
91.7(99.3)
1.3–0.7hPa (N)
37.7(94.0)
57.7(99.7)
51.4(99.3)
66.0(99.8)
74.4(99.6)
1.3–0.7hPa (Y)
87.0(99.8)
99.7(100)
98.6(100)
98.5(100)
98.4(100)
0.13–0.07hPa (N)
26.0(66.5)
16.3(84.7)
35.1(88.7)
35.1(75.1)
40.0(80.6)
0.13–0.07hPa (Y)
46.1(75.7)
80.0(99.4)
76.2(99.1)
53.3(89.3)
57.2(89.8)
Validation and statistical analysis with MLS
In addition to the relative difference maps shown before, histograms
are plotted for every individual case study to illustrate the number
of matching domains corresponding to deviation bins with a width of
5% in the pressure layers. The histograms and relative
difference maps are used for a statistical analysis and
validation. Table further summarizes the percentage of
H2O VMR domains within 10, respectively 20 %
relative difference (Drel, see Eq. ) between
TM and Aura MLS results. Further, percentages corresponding to the TM
performance without instrument bias corrections are given.
The relative differences to Aura MLS in the domain areas do not exceed
20% in most cases and altitude regions. Regarding the
whole number of domains per map, only a few outliers with deviations
Drel>|±20|% are present at pressure
levels below 0.1hPa as confirmed in the histogram charts
(Fig. ). The deviation bins with the maximum number of
relative difference domains (peak of Gaussian curve) are centered
around the zero percent line (perfect coincidence), except for the
lowest altitude in case study B and D where the TM domains show either
too high (case B) or too low (case D) H2O VMR values and
for the highest altitude in all cases where overly low mixing ratios from
TM dominate.
Table underlines the good results of the TM approach with
respect to Aura MLS observations. Referred to the bias
corrected row values of case A and B in the stratosphere and
stratopause level (1hPa), between 98.4 and 100 % of
the compared domains have a difference of less than ±20%. At least around half of all domains from the
investigated difference maps in Figs. and
d and h are indeed within ±10 %. By ignoring case B with its small statistical
significance, slightly over 82% (83%) of
the domains agree within ±10% in case A at
10hPa (3hPa). Within the mesospheric vortices
a significant number of domains show that TM resulted in too high
H2O VMR values (reddish colors in the February polar vortex
case study in Fig. p). Inside the equinox polar
vortex this feature is not present. In the histogram of case
A (0.1hPa) roughly 15 domains exist with a difference of
95–100%, located in the vortex over Greenland and
westwards thereof. While the Gaussian center line still lies close to
the zero percent line (green), the shape of the histogram spreads. The
homogeneous distributed water vapor in case study B at
0.1hPa with values around 7 to 8ppm
(Fig. o) is better confirmed by TM than in the
previous situation. In numbers we now count 46.1% (case
study A) and 80% (case study B) of the domains to be
within the ±10 % deviation limit but at least 75.7 and
99.4% are within the doubled deviation threshold of
20%.
Regarding TM case study C (21 June 2012) close to the June solstice
event, the number of coincident TM and MLS domains is reduced on the
investigated stratospheric pressure levels owing to zonal
circulation patterns compared to case study A or D. Throughout the
middle atmosphere, all relative difference domains from high to low
latitude bands show only light blue to light red colors
(Fig. d, h, l and p). The deviations in
H2O VMR are tiny, the histograms are always narrow. More
than 350 (200) 5∘×5∘ domains are within ±5 %
at 1hPa (3hPa) and nearly all
compared areas show less than 20% relative difference
(cf. Table ) on all four altitude levels.
Next we show the agreement of TM results with MLS observations in case
study D (17 January 2012), the SSW event. As it is affirmed by
Fig. , the prolonged shape of the vortex is very
well represented by the applied trajectory mapping method at the
lowest investigated altitude. Less than 2% of the
compared domains deviate more than 20% in relative
difference. Examining the variations in H2O VMR of the inner
parts of the 3hPa vortex on 17 January 2012, the outcome
has to be evaluated positively with a majority of domains that satisfy
the 10% relative difference quality criterion
(Figs. h and ). The more or less
uniform H2O distribution of 7–7.5ppm in MLS
(Fig. k) could be displayed correctly by TM
(Fig. j). There, the relative differences to MLS
never exceed 20%. At 0.1hPa the water vapor
VMR underestimation by TM is reduced compared to case A at the same
altitude. More than 53% (89%) of the
compared domains stay within a relative difference of
10% (20%). In the mean, TM domains
revealed low mixing ratios relative to MLS in the mesospheric pressure
layer as the leftward shifted peak of the histogram reveals.
Summarizing cases A to D, very good agreement between TM- and MLS-derived water vapor maps was found at the stratopause level
(1hPa∼48km). All TM domains for case B, C and
D deviate less than 20% from MLS (see Table )
and only a tiny percentage of 0.2% show larger relative
differences in case A. Of course histograms show tight bounds and the
full widths at half maxima are small (Fig. ). We assume
that low planetary wave activity in the upper stratosphere around the
selected TM dates was accountable for the low meridional gradients in
water vapor in this altitude region and hence simplified the TM method
to work well.
Histogram charts for estimating the error of TM approach in
case study A (28 February 2012, first column), B (22 September 2012, second
column), C (21 June 2012, third column) and D (17 January 2012,
fourth column) with advection of H2O profiles from Aura MLS instead of
ground-based measured profiles. The number of relative
difference (TMMLS-MLS) domains in a certain deviation bin of a width of
5% between TMMLS and Aura MLS solution is shown. From
top to bottom the pressure layers are 12–8, 3.5–2.5,
1.3–0.7 and 0.13–0.07hPa. Vertical green lines
indicate zero deviation and red dashed lines mark the standard deviations (±σ).
Error estimation of TM approach
In this section, the strategy and outcome of the investigation to estimate the error
and limitation of the trajectory mapping (TM) approach is briefly summarized.
In a first step Aura MLS profiles are taken, located near the five
ground-based observation sites. The chosen criterion for spatial coincident
of the satellite measurements is 600×300km
around the ground-based radiometer locations. The 300km
go along north–south direction, while the 600km go along
east–west direction. The unequal lengths are due to typical
water vapor gradients, which tend to be much smaller in zonal
than meridional direction. A similar way of data processing has been
applied to the obtained Aura MLS profile time series, according to
Sect. . It is evident that correction factors
to account for biases are not necessary, since the same instrument
is used to generate the H2O profiles. The histogram charts
in Fig. show the results of the
investigation. These charts are similar to the ones in Fig.
only with additional tags for the standard deviations ±σ.
For the three pressure layers at lower altitudes (12–8, 3.5–2.5
and 1.3–0.7hPa) it is found that deviations from the
coincident domain comparison never exceed 20%.
Approximately 2/3 of all domains show less than 10% deviation.
The errors become significantly higher in the mesosphere
(0.13–0.07hPa). Estimating the position of the 2/3 value
of the domains in the bar charts, it is
now roughly within ∼20% (doubled) in the mesospheric
pressure layers. Regarding the standard deviation σ, it is clear
that the uncertainties of TM are largest in the mesosphere
(0.13–0.07hPa) of case study A (28 February 2012).
It is also noticeable that TM during the SSW case D (17 January 2012) reveals less uncertainties
(σ is smaller).
Summary and discussion
We have generated NH middle atmospheric water vapor maps from five
single water vapor profile measurement sites, mainly operated in the
frame of NDACC, by use of a spatial domain-averaging trajectory
mapping technique. Forward and backward trajectories were calculated
for up to 10 days with LAGRANTO driven by ECMWF operational analysis wind
field data. A total of four TM case-by-case studies were presented and
discussed, belonging to different atmospheric circulation patterns and
seasons of the year. Apart from the SSW scenario (D) in January 2012,
we discussed (1) a stable polar vortex case (A) at the end of
February, (2) a case near June solstice (C) and (3) a fall equinox
scenario (B). For each case study four pressure layers from the
stratosphere (centered at 10hPa) to the lower mesosphere
(centered at 0.1hPa) were analyzed.
Biases between the ground-based instruments have been corrected using
coincident Aura MLS observations during a defined time period
(August 2010 to September 2014). Calculated mean relative difference
profiles of H2O served as correction factors to harmonize
the data sets. The improvements of the bias corrected synoptic maps is
very pronounced above the 3hPa pressure layer compared to
the uncorrected versions (Table ). At best (case study B,
0.1hPa), 80% instead of 16.3%
of the relative difference domains had a bias of only ±10 %. For the three upper pressure layers of TM studies A,
C and D/D⋆, the corrections led always to an
improvement of trajectory-mapped data, referring to the ±10% as well as to the ±20% regime. On
the 10 and 3hPa pressure level the applied correction
factors led sometimes to a slight worsening for cases B and D/D⋆, for instance a loss of almost 10%
of comparison domains (case D/D⋆ on
10hPa), which were associated to the 10%
quality threshold.
The mesospheric polar vortex edge was very well reproduced, with the
trajectory mapping method by the north–south gradient in water vapor
VMR on 28 February 2012 (Fig. n). In the
mesosphere, where uncertainties in the 3-dimensional wind field become
larger, leading to trajectory position errors, the quality of the TM-derived water vapor distribution is generally reduced, compared to
coincident Aura MLS observations. However, the TM data coverage is
found to be better at higher altitudes such as at the
0.1hPa level in the mesosphere. In addition, the assumption
of unchanged mixing ratios along 240h trajectories might
not be sufficient any more in mesospheric altitudes where photolysis
or chemical reactions of water vapor cannot be totally ignored. But
the analysis of the statistical outcome in the validation part
(Sect. ) shows that the VMR values of the vast
majority of spatial domains match between the TM and Aura MLS result.
After having assessed the errors of the TM method (cf. Fig. ),
it is found that there is not much of a difference observed between
Fig. and the real TM case studies. This is a
sign that a kind of optimal result has been obtained in consideration of the
errors in wind, chemistry or removal from condensed phases,
which cannot be avoided within the TM method.
Predominantly good TM results could be obtained for stratospheric
pressure layers, including the stratopause region at
1hPa. Keeping in mind that complex polar vortex
deformations occurred during the SSW time period of case scenario D
(Sect. ). Based on calm circulation patterns,
prevailing zonal winds and small meridional water vapor VMR gradients,
the summer case study C on 21 June 2012 (Sect. )
showed, beside case B on 22 September 2012, the best quantitative
result, affirmed by Table and Fig.
statistics.