In the framework of the second SPARC (Stratosphere-troposphere Processes And their Role in Climate) water vapour assessment (WAVAS-II), the amplitudes and phases of the annual, semi-annual and quasi-biennial variation in stratospheric and lower mesospheric water were compared using 30 data sets from 13 different satellite instruments. These comparisons aimed to provide a comprehensive overview of the typical uncertainties in the observational database which can be considered in subsequent observational and modelling studies. For the amplitudes, a good agreement of their latitude and altitude distribution was found. Quantitatively there were differences in particular at high latitudes, close to the tropopause and in the lower mesosphere. In these regions, the standard deviation over all data sets typically exceeded 0.2 ppmv for the annual variation and 0.1 ppmv for the semi-annual and quasi-biennial variation. For the phase, larger differences between the data sets were found in the lower mesosphere. Generally the smallest phase uncertainties can be observed in regions where the amplitude of the variability is large. The standard deviations of the phases for all data sets were typically smaller than a month for the annual and semi-annual variation and smaller than 5 months for the quasi-biennial variation. The amplitude and phase differences among the data sets are caused by a combination of factors. In general, differences in the temporal variation of systematic errors and in the observational sampling play a dominant role. In addition, differences in the vertical resolution of the data, the considered time periods and influences of clouds, aerosols as well as non-local thermodynamic equilibrium (NLTE) effects cause differences between the individual data sets.
Water vapour is the most fundamental trace constituent in the troposphere but
continues to play an important role in the stratosphere and lower mesosphere.
In the lower stratosphere, in particular in the tropics, water vapour is the
most important greenhouse gas, strongly affecting global warming at the
Earth's surface
In the stratosphere and lower mesosphere water vapour has two major sources.
One is the transport of water vapour into the stratosphere that primarily
happens through the cold tropical tropopause layer. Here a large fraction of
water vapour is removed due to freeze-drying, leading to a typical entry
mixing ratio of about 3.5 to 4.0 ppmv in the lower stratosphere
Due to the importance of stratospheric and lower mesospheric water vapour, a
major research focus over the last two decades has been to understand
long-term changes of this constituent in this altitude region. In the past
many observations have indicated an increase in lower stratospheric water
vapour
A complete understanding of water vapour changes also requires good
knowledge of short-term variability, such as the annual and semi-annual
variation or the variation caused by the quasi-biennial oscillation (which we
denote here as QBO variation). The basic aspects of these shorter-term
variations in stratospheric and lower mesospheric water vapour have been
investigated in a number of studies
Overview over the water vapour data sets from satellites used in this study.
Within the second SPARC water vapour assessment (WAVAS-II) a suite of 40 data
sets (not including data sets of minor water vapour isotopologues) has been
considered, focusing on observations from 2000 to 2014
The data sets were initially screened according to the criteria recommended
by the data providers, which are summarised by
In the comparison of the variability derived from the different data sets, we
focused on two parameters, namely the amplitude and the phase (which we
collectively denote as variability characteristics). To determine these
parameters, multi-linear regressions were employed using the entire time
periods of the individual data sets (see last column of
Table
The amplitude of the annual variation
For the comparison of the semi-annual variation the regression model in
Eq. (
Example of the annual variation characteristics as a function of latitude and altitude based on the MLS data set. The left panel shows the amplitude and in the right panel the phase is shown. The phase is represented by the month in which water vapour exhibits its annual maximum in the regression fit. The light grey and white lines indicate the mean tropopause (2000–2014) derived from the MERRA reanalysis data. The red and black boxes mark characteristic features of the annual variation in water vapour that are described in more detail in the text. The colour variation of the boxes is for better contrast. White areas indicate that there are no data.
In the analysis of the QBO variation only those data sets were considered
that cover a time period of at least 28 months, which is the average period
of this variability pattern
To provide a measure for the typical spread among the data sets, the standard
deviation for the amplitudes and phases were derived for each variability
pattern. We recognise that a Gaussian distribution is not to be expected,
neither for the amplitudes nor for the phases. Still, we assume that the standard
deviation could serve as a good proxy of the typical uncertainty in the
observational database. Outliers among the individual data sets can,
however, render this purpose meaningless, in particular for the amplitudes. Outliers
especially occur at altitudes close to the lower and upper boundaries of the
individual data sets, where measurement uncertainty is large. To avoid such
outliers the amplitudes were screened using once again the median and the
median absolute deviation (see Sect.
In contrast to the amplitude, the phase
The phases corresponding to the screened amplitudes were discarded as well.
Thus
In this section, the comparison results for the annual, semi-annual and QBO
variation will be described. For every pattern we start with an example of
the typical latitude–altitude distribution for the amplitude and phase to
briefly describe the most prominent features. In the Supplement, such plots
are provided for all data sets considered in this work. Thereafter the
variability characteristics are shown at specific latitudes and altitudes.
These figures combine all data sets, allowing for a direct comparison and the
estimation of typical uncertainties. Finally, a summary is provided in the
form of standard deviations for the amplitude and phase, as described in
Sect.
The main focus will be on the stratosphere, with some discussion of the lower
mesosphere. Upper tropospheric results are visible in many figures, but will
not be discussed. The mean tropopause for the time period 2000–2014 based
on MERRA (Modern Era Retrospective-Analysis for Research and Applications,
Figure
This feature, the “atmospheric tape recorder” The amplitude of the annual variation exhibits a distinct local maximum in the southern tropical and sub-tropical middle
and upper stratosphere. The phase plot indicates that this variation has its annual maximum in August or September. In the
Northern Hemisphere a weaker counterpart can be found (not marked). The annual cycle of vertical transport within the
Brewer–Dobson circulation essentially explains the annual cycle in water vapour. However, to explain the amplitude of
this feature contributions from chemistry, meridional transport and eddy transport are also necessary. All these contributions
actually act to reduce the amplitude. The interhemispheric differences are caused by a combination of differences in transport
and chemistry. A full characterisation of this feature will be given by This amplitude maximum is due to the annual cycle of dehydration caused by PSC ice particles This high amplitude in the annual variation here is caused by the annual cycle in the vertical transport at
high latitudes. During winter, the middle atmospheric water vapour maximum is shifted down to about
10 hPa A large amplitude in the annual variation can be found in the lower polar mesosphere in both hemispheres.
The reasons for this feature are very similar to those for key feature #4, except that feature #5 occurs at
altitudes above the middle atmospheric water vapour maximum. During winter, dry air from the upper mesosphere
descends within the polar vortex, while during summer and early autumn moist air from below is transported
upwards. In addition, in summer more water vapour is produced from methane oxidation due to higher insolation
As visible in the Supplement (see Fig. S1), these features are well depicted in most data sets even though quantitative differences exist. Depending on adequate observational coverage, all features (like in the MIPAS and MLS data sets) or only a subset of them can be found.
The annual variation in the inner tropics (5
Here, we focus on the annual variation characteristics around the Equator
(key feature #1), in the Antarctic (key features #3–5) and at a pressure
level of 1 hPa (key feature #5, slightly above key feature #2). In
Fig.
As Fig.
Figure
The annual variation as observed by the individual data sets at the 1 hPa pressure level over the entire latitude range.
Figure
The typical uncertainties among the different data sets for the
amplitude and phase of the annual variation. The upper panel shows the
standard deviation for the amplitude in absolute terms
Figure
The lower panel of Fig.
As Fig.
Figure
In the tropical upper stratosphere the semi-annual variation is the most important mode of annual variability.
The larger amplitudes in water vapour are due to vertical and meridional advection induced by the stratospheric semi-annual
oscillation in the zonal wind (SSAO). This oscillation is forced by a combination of processes. The easterly accelerations
are due to the meridional advection of easterlies from the summer hemisphere across the Equator as well as the eddy
momentum deposition from breaking planetary waves which have been ducted in the tropics. Momentum deposition from the
interaction of vertically propagating ultra-fast Kelvin waves and internal gravity waves with the mean flow cause the
westerly accelerations This maximum is related to the dehydration due to PSCs (see key feature #3 for the annual variation) and the annual
cycle in the vertical transport. At the beginning of the year, the water vapour volume mixing ratios show a minimum in
these altitudes. Towards wintertime water vapour increases due to the downwelling of moist air. Soon after temperatures
allow the widespread formation of PSCs, dehydration sets in, causing a deep minimum in water vapour during austral winter
and spring. After September, the water vapour volume mixing ratios increase again as the PSC influence reduces.
Typically in November, a small maximum can be observed before the upwelling in summer decreases water vapour again.
Thus, this feature is not a semi-annual variation in the classical sense, as also noted for the corresponding key feature #3 of the annual variation. In the sinusoidal regression approach employed here this semi-annual variation also acts as a correction of the non-sinusoidal shape of the annual variation. Due to the annual cycle in the vertical transport in the polar regions, the altitude of the middle atmospheric
maximum in the vertical water vapour distribution shifts from roughly 10 hPa in winter up to the stratopause in
summer (see key feature #4 for the annual variation). This shift gives rise to a semi-annual variation at altitudes
in between. Maxima in the annual variation occur in spring and autumn in the transition of the vertical winds from
winter to summer conditions and vice versa. Increasing amplitudes can be observed towards the middle mesosphere in the polar regions. A strong
maximum can be observed in summer (key feature #5 of the annual variation) due to the upwelling of moist
air but there is also a small maximum or plateau in late winter and early spring. In the Arctic, a large
contribution to this behaviour arises from sudden stratospheric warmings that have been relatively frequent
since the new millennium Above key feature #1, the amplitude of the semi-annual variation decreases significantly. Higher
up in the middle mesosphere the amplitudes increase again. The maxima typically occur around equinox,
thus are phase-shifted relative to the prominent semi-annual variation in the upper stratosphere
The semi-annual features highlighted in Fig.
The characteristics of the semi-annual variation in the inner
tropics (5
For the semi-annual variation, we detail characteristics in the tropics (key
features #1 and #5), the Arctic (key features #3 and #4) and at a
pressure level of 2.4 hPa where the semi-annual variation is strong in the
tropics (key feature #1). Figure
As Fig.
Figure
The latitude distribution of the semi-annual variation at 2.4 hPa,
close to the pressure level where the tropical semi-annual variation peaks as
shown in Figs.
The latitude dependence of the semi-annual variation characteristics at an
altitude of 2.4 hPa (key features #1 and #3) is shown in
Figure
As Fig.
Based on the combination of results from all data sets,
Fig.
Example of the variability characteristics for the QBO variation
based on the MIPAS-IMKIAA V5R NOM data set. The phase is derived as the shift
of the QBO regression fit for which the correlation with the Singapore
(1
Various characteristics of the QBO variation in water vapour are shown in
Fig.
In the tropical lower and middle stratosphere, the QBO variation exhibits a slightly enhanced amplitude.
This region comprises the area where the signal of the QBO is strongest in the zonal wind. QBO variations in the
vertical wind, affecting the transport of water vapour as well as the advection of QBO induced anomalies in the
stratospheric entry water vapour, contribute to this feature A much larger amplitude is found in the tropical upper stratosphere. This variation is, to a large extent,
due to transport of water vapour by the QBO-induced anomalies of the residual circulation An enhanced amplitude is found in the polar regions. In the Antarctic, the maximum can typically be observed
close to 10 hPa. In the Arctic the variation is smaller and the maximum has a tendency to occur higher up than in
the Southern Hemisphere. This feature reflects the influence of the QBO on the polar regions, probably induced
through QBO variations in the downwelling conditions during winter and the upwelling conditions during summer.
As for key feature #1, the phase shift has a pronounced altitude dependence.
The QBO variation in the tropics (5
For the QBO, we focus again on the tropics (key features #1 and #2) and
show an example at an altitude of 7.5 hPa which addresses key feature #3.
The characteristics in the tropics are shown in Fig.
The latitude distribution of the QBO variation at a pressure level of 7.5 hPa.
A final example is given in Fig.
As Figs.
As for the other variability patterns, Fig.
The amplitudes and phases of the annual, semi-annual and QBO variation in stratospheric and lower mesospheric water vapour from 32 data sets were compared to provide estimates of the typical uncertainties in these essential quantities in the observational data record. In this section, the results are first summarised. Thereafter, possible reasons for differences among the data sets are discussed and the sensitivity of the results is addressed by considering other approaches to derive the variability characteristics.
Overall, good agreement in the latitude–altitude distributions of the
variability characteristics was found, in particular for the amplitudes for
all modes of variability considered. The key features for the individual
variability patterns briefly described in
Sects.
To summarise the results for the amplitudes of the different variability
patterns, standard deviations among all data sets were derived in absolute
terms but also relative to the average over all data sets
(Figs.
The phases of the different variability patterns exhibited the smallest
uncertainties around the key features and larger uncertainties elsewhere.
This strongly resembles the pattern found for the relative standard deviation
of the amplitudes. To summarise, we derived the standard deviations over the
phase difference to the reference data set, which was chosen to be the MLS data
set. Taking all latitudes and altitudes above the
tropopause into account again, the occurrence rate of phase differences within
There are a number of reasons that potentially explain the differences
observed among the data sets. In the following subsections the most prominent reasons are considered: (1) different measurement time
periods, (2) different sampling in time and space, (3) temporal variations of
systematic errors, (4) differences in vertical resolution, (5) influences of
clouds and aerosols and (6) NLTE effects, which are considered in the
following subsections. To quantify the impact of some of the aspects listed
above we performed sensitivity studies (not shown). Changes are always given
as sensitivity test results minus the results presented in
Sect.
In our analysis we always used the entire time periods covered by the
individual data sets. Differences in those, combined with interannual
variability and changes in the QBO, impact the results. This aspect was
investigated by comparing results where the entire measurement periods of the
data sets were used and results where only observations between January 2006
and December 2011 were considered. In total, 18 data sets (ACE-FTS, MAESTRO,
MIPAS V5R NOM, MIPAS V5R MA, MLS, SCIAMACHY and SMR) contributed to this
sensitivity study. The expected improvement of the consistency among the data
sets was only found on occasion. Based on the absolute standard deviation,
the spread among the data sets increased slightly overall for the annual
variation, both in terms of amplitude and phase. An improvement in terms of
amplitude was found in the tropical and subtropical upper stratosphere
(partly related to key feature #2), while other regions predominantly showed
an increased spread among the data sets. For the phase only the upper
stratosphere between 50 and 70
To derive the time series for the individual data sets, the data were
averaged in monthly and 10
We investigated the influence of an incomplete coverage throughout the year
as one aspect of the sampling problem. A typical example is the limited
coverage of ACE-FTS in the tropics, only providing measurements in February,
April, August and October. Using this example, we analysed the sensitivity of
the annual variation by subsampling the MIPAS V5R NOM and MLS data sets to
these four months. The largest amplitude sensitivity was found in the
Antarctic. In the lower stratosphere (related to key feature #3) and in the
lower mesosphere (related to key feature #5), an increase of more than
0.25 ppmv was found. Contrary to this, in the middle stratosphere (related
to key feature #4) a decrease of the amplitude (
All data sets have systematic errors. If these errors vary with time they can
affect the variability characteristics derived from the data set. Systematic
errors are commonly associated with uncertainties in the instrument characterisation and spectroscopic parameters. Some of the differences
observed for the SMR data sets in this analysis can, for example, be attributed
to systematic errors that vary in time. SMR is a heterodyne instrument in
which the received signal
The data sets considered in this work have different vertical resolutions.
Full details are given in the WAVAS-II data set overview paper by
In addition, we performed simple tests with model simulations to study the
influence of the vertical resolution on the characteristics of the annual
variation. The high vertical resolution (
The presence of clouds can introduce errors into the water vapour retrievals from satellite observations. For this study, PSCs in the Antarctic during winter and spring are the most significant clouds impacting variability amplitudes and phases. There are obvious differences in the magnitude of the dehydration due to these clouds that clearly influence the annual (key feature #3) and the semi-annual variation (key feature #2). Upper tropospheric clouds also contaminate retrievals and impacts can be seen even in the lower stratosphere. This is, on one hand, due to the finite altitude resolution of the satellite data sets and, on the other hand, due to the propagation of errors. The measurement technique itself plays an important role for the cloud influence. Observations in the microwave region (as by MLS or SMR) are less affected by clouds than observations in the infrared (e.g. ACE-FTS, HALOE, MIPAS). The latter set of observations cannot measure above a given optical depth; hence the measurements and subsequent results will be biased towards cloud-free situations (i.e. biased dry). Similarly the observation geometry plays a role in the influence of clouds. Observations by solar occultation (e.g. ACE-FTS, HALOE, SAGE) will be less influenced than lunar (SCIAMACHY lunar) and stellar (GOMOS) occultation measurements due to a much stronger measurement signal. The same relation is valid if comparing solar occultation to emission (particularly MIPAS) and solar scattering (SCIAMACHY limb) measurements. Beyond that there are even differences in the cloud screening among the MIPAS data sets from the different processors. Hence, given the variety of measurement techniques and observation geometries, differences are unavoidable in the regions where clouds exist and assert an influence. Since cloudiness varies over the year and is also affected by the QBO, it impacts all modes of variability in water vapour addressed in this study.
In a similar way, the observations are affected by aerosols. This primarily concerns
the troposphere and lower stratosphere and the impact varies again
among the data sets. On one hand there is a data set like the SCIAMACHY limb
data set for which clear effects from aerosols have been found
NLTE effects are important in the upper stratosphere and the mesosphere,
especially during the daytime. The MIPAS data sets are influenced by these
effects. However, with the exception of the MIPAS-IMKIAA V5R MA data set,
none of them explicitly consider this influence (except by selecting
spectral information unaffected by NLTE for the retrieval).
There are many ways to derive the characteristics of the different
variability patterns. This implies a sensitivity of the results based on the
chosen approach, as already pointed out in the approach section
(Sect.
The regression models for the semi-annual
(Eq.
Comparison of amplitude and phase for the annual variation as
derived with the climatology approach (top row; see
Sect.
Another common approach to derive the characteristics of the annual variation is based on multi-year average (climatological) distributions, typically considering monthly zonal means. From this set of data, the amplitudes and phases can be determined via the size and the time of the annual extrema. To be consistent with the results from the regression approach the amplitude of the annual variation is given by
To test the sensitivity of the approach for the semi-annual variation, the
results obtained with the regression model for the QBO variation
(Eq.
For the sensitivity of the QBO variation, a test was performed in which the
regression model used the normalised Singapore winds at 30 and 10 hPa
as QBO proxies (instead of 50 and 30 hPa as in
Eq.
In summary, the sensitivity studies showed that the amplitudes of the different variability patterns were affected by the chosen approach, but the typical latitude–altitude distributions remained rather consistent. The phases were more sensitive to the approach, but both the overall distribution and the actual values were affected. In general, the most sensitive region was the polar region. This result needs to be kept in mind when comparing the variability characteristics based on different approaches.
Satellite data sets of water vapour in the stratosphere and lower mesosphere were compared with respect to the amplitude and phase of the annual, semi-annual and QBO variation. The comparisons indicated a rather consistent picture of the latitude–altitude distribution of the amplitudes of the three variability patterns. Quantitatively, however, there were obvious differences in the amplitudes derived from the individual data sets. Larger differences were typically observed at high latitudes, close to the tropopause and in the lower mesosphere. At low latitudes larger differences were found for the QBO variation. Depending on variability pattern, latitude and altitude, the spread among the data sets amounted to several tenths of 1 ppmv and in extreme cases they even exceeded 1 ppmv.
In the problematic regions, the standard deviation over all data sets
exceeded 0.2 ppmv for the annual variation. For the two other variability
patterns, standard deviations exceeding 0.1 ppmv were found to be
characteristic for these regions. The relative standard deviation (using the
average amplitude over all data sets as the reference) revealed a different
picture. Many regions with large spreads often coincided with large
variability, resulting in relatively small relative standard deviations
(
There are multiple reasons that give rise to the observed differences between the individual data sets. The most important contributions arise in general from temporal variations of systematic errors and differences in the temporal and spatial sampling. Other reasons include differences in the vertical resolution of the data, the time periods available for analysis as well as the influence of clouds, aerosols and NLTE effects. Beyond that different choices in the retrieval approach can have an effect. There are various ways to derive the characteristics of the different variability patterns. Different approaches lead to quantitative differences in the amplitude and phase estimates, which need to be considered in comparisons with other results. The latitude–altitude distribution of the amplitude is quite insensitive to the derivation approach, while the phase is more sensitive.
Overall the results provide valuable constraints on the characteristics of
shorter-term variations in stratospheric and lower mesospheric water vapour
for subsequent modelling or observational studies. A prominent example is the
simulation of the atmospheric tape recorder that still exhibits fundamental
differences compared to the observations, with respect to the amplitude but
especially the phase
Data are available upon request.
The authors declare that they have no conflict of interest.
The Atmospheric Chemistry Experiment (ACE), also known as SCISAT, is a Canadian-led mission mainly supported by the Canadian Space Agency and the Natural Sciences and Engineering Research Council of Canada. We would like to thank the European Space Agency (ESA) for making the MIPAS level-1b data set available. SCIAMACHY spectral data have been provided by ESA. The work on the SCIAMACHY water vapour data products has been funded by DLR (German Aerospace Center) and the University of Bremen. The SCIAMACHY limb water vapour data set v3.01 is a result of the DFG (German Research Council) Research Unit “Stratospheric Change and its Role for Climate Prediction” (SHARP) and the ESA SPIN (ESA SPARC Initiative) project and were partly calculated using resources of the German HLRN (High-Performance Computer Center North). Stefan Lossow was also funded by the SHARP project under contract STI 210/9-2. He would like to thank D. Lindelof and C. Cuse for their support: this is the beginning of the end. We want to express our gratitude to SPARC and WCRP (World Climate Research Programme) for their guidance, sponsorship and support of the WAVAS-II programme. We thank E. Remsberg, H. Pumphrey and an anonymous referee for valuable comments on the manuscript. The article processing charges for this open-access publication were covered by a Research Centre of the Helmholtz Association. Edited by: S. Buehler Reviewed by: H. C. Pumphrey and one anonymous referee