The transport of water vapour from the troposphere to the stratosphere through the cold tropical tropopause layer (TTL) is a critical atmospheric process. On the one hand, it directly influences the radiative balance as water vapour is the most important greenhouse gas in this region. On the other hand, it constitutes an important contribution to the water vapour budget in the stratosphere. There are multiple transport pathways. The slow ascent within the ascending branch of the Brewer–Dobson circulation is thought to be the dominating pathway . This ascent is accompanied by large horizontal transport patterns e.g.. They cause a large part of the air trajectories to pass over the western Pacific, where often the lowest temperatures are encountered and consequently the final dehydration is attained. In addition, the convective lofting of ice particles into the stratosphere is thought to be an important pathway e.g.. Once the ice particles reach the stratosphere they evaporate and enhance the amount of stratospheric water vapour. Overall, the stratospheric water vapour entry mixing ratios amount to only 3.5–4.0 ppmv on an annual average , which is a reduction of 3 orders of magnitude compared to the mixing ratios commonly observed at the Earth's surface.

The δD-H2O (hereafter simply denoted as δD) isotopic ratio (see Eq. ) between the minor water vapour isotopologue HD16O (hereafter HDO) and the main isotopologue H216O (hereafter H2O) is a valuable tool for atmospheric research e.g.. This is because it is sensitive to physical, chemical and radiative processes, which can be used to infer the history of air parcels. As such it can help to discern the relative importance of slow ascent and convective ice lofting for the transport of water vapour into the stratosphere . Several parameters are of interest in the tropical lower stratosphere for these investigations; for example, the absolute value of δD, the slope of the HDO–H2O correlation, any long-term trends in δD or the regularity and the size of its annual variation. The slow ascent into the stratosphere can, to a first order, be idealised as a pure Rayleigh fractionation process. Correspondingly, a δD value of about -900 ‰ for water entering the stratosphere would be expected. Observations, however, show typically δD values between about -700 ‰ and -500 ‰ e.g., clearly indicating an importance of other processes like the convective lofting of ice or the mixing of air masses and supersaturation effects.

Based on MIPAS observations from Envisat (Environmental Satellite), reported a steeper slope for the HDO–H2O correlation than expected from Rayleigh fractionation. Discussing different processes influencing the slope they concluded that convectively lofted ice provides the most plausible explanation for the observed steepening of the slope. In their work only focused on altitudes between 25 km (∼25 hPa) and 30 km (∼12 hPa). Below 25 km (∼25 hPa), they argued, the MIPAS results are affected by a dependency on the lowermost altitude where retrievals are possible. This lowermost altitude is primarily determined by cloudiness but also aerosols, increasing water vapour absorption and the atmospheric temperature . The link to results at altitudes above occurs through error propagation in the MIPAS global fit retrieval approach . We will refer to this dependency as “start altitude effect” (see more in Sect. ).

An investigation by addressed future trends of water vapour in the tropical lower stratosphere, comparing simulations from chemistry–climate models and a trajectory model. A substantial increase of about 1 ppmv until the end of the 21st century was found. About 50 % to 80 % of this trend could be ascribed to a warming of the TTL in this time period . The remainder was attributed to an increase in evaporation of convectively lofted ice. Such a trend would inevitably show up as a trend in the δD ratio. An observational analysis of such a possible change in the past was performed by , employing almost two dozen balloon-borne observations by the Mark IV interferometer , covering the time period between 1991 and 2007 and latitudes from 35 to 65∘ N. They found a small positive trend in the δD values of water entering the stratosphere; however, this trend was not statistically significant. The authors concluded that there was no solid indication that the amount of ice entering the stratosphere had changed.

The annual variation of HDO and H2O in the tropical lower stratosphere is characterised by the atmospheric tape recorder signal . The tropical tropopause (cold point) temperature and its annual variation is imprinted in water vapour. The resulting signal is transported upwards and maintained up to about 30 km (∼12 hPa). For δD, the question of whether there is also such a tape recorder signal found as for HDO and H2O arises. If not, or if the pattern is obviously disrupted, it would indicate the importance of other processes than slow ascent such as the convective ice lofting. The reported observations show different answers to this question. discussed MIPAS observations and found a pronounced and coherent tape recorder signal in δD up to 30 km (∼12 hPa). Slightly above the tropopause the annual variation had an amplitude of about 60 ‰ to 70 ‰, which is close to the expected variation based on the annual variation of the tropopause temperature. It can be expected that these results are affected by the start altitude effect described above, as they occur below 25 km (∼25 hPa). Nonetheless, the tape recorder signal was robust in multiple retrieval setups. analysed observations of the ACE-FTS instrument on board SCISAT (Science Satellite) and found a seasonal variation in the lower stratosphere up to 20 km (∼56 hPa) and assigned this to the Asian and American monsoon systems. They excluded the possibility that this annual variation is a tape recorder signal. In addition, the amplitude of the annual variation in δD was considerably smaller than that observed in the MIPAS observations. Recently, compared multiple δD data sets from MIPAS, ACE-FTS and the SMR instrument aboard the Odin satellite and concluded that all showed characteristics that typically are associated with the tape recorder signal. They iterated that pronounced quantitative differences exist, which would yield different interpretations of the relative importance of convection for lower stratospheric water vapour.

On the modelling side, derived an amplitude of about 25 ‰ for the annual variation of δD slightly above the tropopause, using a conceptual two-dimensional model incorporating slow ascent, extratropical mixing and convection effects. implemented HDO into the EMAC (ECHAM/MESSy Atmospheric Chemistry) model and showed a coherent tape recorder signal in δD up to about 27 km (∼19 hPa). Slightly above the tropopause the amplitude of the annual variation in δD amounted to about 20 ‰ . A sensitivity study showed that the methane oxidation determines the upper end of the tape recorder signal. Methane is much less depleted in deuterium than water vapour upon the entrance of the stratosphere, i.e. δD-CH4 ≈-80 ‰ . The water vapour produced in situ in the lower stratosphere has, therefore, high δD values which overshadow the actual tape recorder signal in the EMAC simulation .

In this work we reassess the differences in the annual variation of δD in the tropical lower stratosphere between the MIPAS (IMK/IAA) and ACE-FTS data sets. For that, we consider newer data set versions than employed in the studies of and and find that the same discrepancies exist. We discuss multiple aspects that could give rise to the observed differences, primarily focusing on the MIPAS data set. The MIPAS results in the altitude range below 25 km (∼25 hPa), which have not been included in scientific analyses so far, are considered in particular. The paper is structured as follows. In the next section we briefly describe the data sets used in this study and how they have been handled. In Sect.  we reassess the observational discrepancies by showing the tropical lower stratospheric δD time series and by comparing the amplitudes and phases derived for the annual variation. In Sect.  we critically consider possible reasons for the discrepancies between the observational data sets. One focus is on the influence that clouds exert on the MIPAS results. In addition, we show that the MIPAS results are influenced by an artefact originating from different vertical resolutions of the retrieved HDO and H2O data. We demonstrate that the consideration of the MIPAS averaging kernels in the correct way largely removes any discrepancies between the MIPAS and ACE-FTS data sets and confirm a δD tape recorder signal with an amplitude of about 25 ‰ in the lowermost stratosphere. A summary and conclusion of our result will be given in Sect. .

In this study we focus on data sets derived from observations of Envisat/MIPAS and SCISAT/ACE-FTS. As noted in the Introduction section, we here employ newer data sets than used in the original works of and to reassess the observational discrepancies in the annual variation of δD in the tropical lower stratosphere. As a complement, we show Odin/SMR results to cover the entire observational database from satellites since the new millennium. In addition, also results from the EMAC simulation are considered. Below, the data sets are briefly described. For a detailed description of the satellite data sets, the reader is referred to the work of . After the data set description the data handling is explained.

In this section the observational discrepancies in the annual variation of δD in the tropical lower stratosphere between the MIPAS and ACE-FTS data sets are reassessed using three figures. Figure  shows the time series of the data sets in the form of contour plots. In Fig.  the focus is on the time series at 70 hPa, not only for δD but also HDO and H2O for the sake of attribution. In addition, fits from a regression model to quantify the amplitude and the phase of the annual variation (see Eq.  in the Appendix) are presented in this figure. Figure  shows subsequently the derived amplitudes and phases for the annual variation, again for δD, HDO and H2O. In all three figures, results from the SMR observations and the EMAC simulation are shown as a complement.

Figure  shows the time series of the δD anomaly with respect to the climatological annual mean for the different data sets. Please note that the time axes vary depending on the data set. For the EMAC simulation, the time period from 2000 to 2005 is arbitrarily considered.

Other time periods change details but not the overall picture. Gaps in the observational time series are removed for visual benefit. This concerns primarily the ACE-FTS data sets that have tropical coverage typically only during four months, i.e. February, April, August and October. In 2013 and 2014 there is also some coverage in May and November due to the shifting orbit of SCISAT. This coverage is limited to the very beginning of these months and comprises the northern tropics in May and southern tropics in November. Due to the gap removal, the individual monthly averages appear to be valid for longer periods of time in the figure. Also for the SMR data set such gaps occur in early 2005 and 2006. A corresponding figure with these observational gaps is provided in the Supplement (Fig. S1). In addition, a figure showing only 2 years of data for the ACE-FTS, SMR and EMAC data sets is supplied (Fig. S2) to provide a more comparable picture to the time coverage of the MIPAS data set.

The δD results from the different data sets exhibit profound differences. The most coherent temporal behaviour can be observed in the MIPAS data set. It shows a very distinct tape recorder signal in the tropical lower stratosphere, consistent with the results reported by using an earlier MIPAS retrieval version. Relative to the MIPAS data, the ACE-FTS data set shows a weak annual variation. This is consistent with the results reported by for the older retrieval version 2.2, where the annual variation was attributed to the monsoon systems. The annual variation is larger below 50 hPa than higher up. The SMR observations provide information on the δD variation roughly above 50 hPa. At lower altitudes no H2O data can be retrieved from SMR measurements of the 489 GHz band. The variation seems to not be very coherent. During some periods, glimpses of a tape recorder signal may be observed, as for example in 2007 and 2008. In contrast, the EMAC simulations show a coherent tape recorder signal which is clearly detectable up to at least 25 hPa. The annual variation is clearly smaller than in the MIPAS data set, fitting much better to that of the ACE-FTS data set.

To characterise the annual variation quantitatively, i.e. its amplitude and phase, we regressed the time series from the individual data sets with a simple regression model that contained an offset, a linear term and the annual variation (see Eq.  in the Appendix). The method by was used to derive the regression coefficients using the standard mean error of the monthly means as statistical weights. Autocorrelation effects and empirical errors as described in are not considered for simplicity. As an example, Fig.  shows the fits (light colours) of the tropical, monthly mean δD (Fig. 2a), HDO (Fig. 2b) and H2O (Fig. 2c) time series (dark colours) from the different data sets at 70 hPa. In the Supplement a corresponding figure (Fig. S3), focusing on data at 30 hPa, is provided. Figure  summarises the derived amplitudes (Fig. 3a, c, e; see Eq.  in the Appendix) and phases (Fig. 3b, d, f; see Eq.  in the Appendix) as a function of altitude for all data sets. The top row considers the results for δD, while the results for HDO and H2O are shown in the middle and bottom rows, respectively. The error bars indicate the 2σ uncertainty level of the derived quantities. The amplitudes of the annual variation in δD exhibit clear differences between the MIPAS and ACE-FTS data sets up to about 30 hPa. The data sets from both instruments show the largest amplitudes in the lower stratosphere at about 75 hPa. However, at this altitude also the differences between the MIPAS and ACE-FTS data sets maximise (about 50 ‰). This result is affected by the start altitude effect described in the Introduction section. Above, the amplitude decreases significantly. Between 50 and 30 hPa the amplitude for the MIPAS data set is relatively constant, with slightly more than 20 ‰. Slightly below 40 hPa the ACE-FTS data set shows essentially no annual variation, which may be an artefact. However, this behaviour is also observed for the older version 2.2 data set used in the work of , which in addition covers a shorter time period. Higher up, the amplitude increases again to 10 ‰ close to 30 hPa. The SMR data set exhibits a very large amplitude at its lower boundary. It quickly decreases towards 35 hPa where a good agreement with the ACE-FTS data sets can be observed. Above 30 hPa the amplitudes derived from the individual data sets are much closer. The EMAC results compare best to the ACE-FTS results below 30 hPa, even though some important differences exist. The simulation exhibits the maximum amplitude at about 90 hPa, and it does not show a local minimum around 40 hPa as the ACE-FTS observations do. Around 20 hPa the simulation exhibits the best agreement with the ACE-FTS and SMR results.

For all data sets the annual variation exhibits the typical phase progression associated with the tape recorder signal. Only the ACE-FTS data set shows a slightly special behaviour between 60 and 40 hPa. In the lower part the phase is relatively constant, while above 45 hPa the behaviour could be interpreted as a phase jump (even though the uncertainties are fairly large). It clearly relates to the low amplitude of the annual variation. Higher up, the phase progression is again as expected for a tape recorder. While the vertical structure of the phase qualitatively agrees, there are substantial quantitative differences among the data sets, similar to the amplitude of the annual variation. Below 60 hPa the difference between the MIPAS and ACE-FTS data sets typically amounts to 2–3 months. Here, the ACE-FTS data set systematically indicates an earlier occurrence of the annual maximum, comparing better with the EMAC simulation. Also, between 35 hPa and 15 hPa there are clear differences between the MIPAS and ACE-FTS data sets. They range between 1–3 months, with the latter data set again showing a preceding phase. Above 20 hPa the spread among the data sets is quite large (up to 4 months); however, this applies also to the uncertainties of the phase estimates.

The results for the annual variation of HDO qualitatively indicate a good consistency among the data sets. The most obvious quantitative differences between the MIPAS and ACE-FTS data sets occur around 75, 50 and 20 hPa. The SMR data set exhibits distinctively larger amplitudes below 50 hPa than the other data sets, with few exceptions. Above, the SMR results are close to that of the ACE-FTS and EMAC data sets. The amplitude derived from the EMAC simulation is smallest below 60 hPa. Phase-wise the consistency among the data sets is better for HDO than δD; often the differences are within 2 months. The behaviour that the annual maximum occurs earlier in the ACE-FTS data set than the MIPAS data set persists also for HDO. The EMAC phase precedes all other data sets except above 20 hPa.

The smallest spread among the different data sets in terms of the H2O annual variation amplitude is observed above 40 hPa. Below 50 hPa the amplitudes derived from the SMR data set are substantially larger than those from the other data sets. This behaviour is the primary reason for the corresponding differences observed in the δD data. Below 50 hPa there are also significant differences between the MIPAS and ACE-FTS data sets. The differences amount up to 0.4 ppmv and peak close to 85 hPa. Below 70 hPa the ACE-FTS amplitudes are close to those derived from the EMAC simulation. Higher up, the vertical gradient in the amplitude is smaller in the EMAC simulation, leading to clear differences up to about 40 hPa. The comparisons for the phases of the annual variation in H2O largely resemble the results obtained for HDO.

As shown in the last section, the δD time series in the tropical lower stratosphere and the derived characteristics of the annual variation exhibit clear differences among the data sets from the MIPAS and ACE-FTS (as well as SMR) satellite observations (and the EMAC simulation). In the following discussion we investigate more thoroughly possible reasons for the differences in the annual variation of δD in the tropical lower stratosphere, with a clear focus on the MIPAS data set. We will consider the three following aspects:

the differences in the MIPAS data set in terms of the temporal sampling of the tropics as well as the inferior vertical resolution relative to the ACE-FTS data set;

Time series of tropical lower stratospheric δD exhibit distinct discrepancies. This concerns in particular the characteristics of the annual variation derived from the MIPAS and ACE-FTS data sets, making it difficult to draw robust conclusions relevant for water vapour transport into the stratosphere. For the amplitude of the annual variation, these differences occur prominently below 30 hPa, with the MIPAS data set consistently indicating larger amplitudes. The maximum difference is observed slightly below 70 hPa where the amplitude derived from the MIPAS data set is about 50 ‰ larger than for the ACE-FTS data set (which shows an amplitude of around 25 ‰). Besides the amplitudes, also the phases of the annual variation exhibit obvious differences between the data sets. In the lowermost stratosphere the differences between the MIPAS and ACE-FTS data sets typically range between 2 and 3 months, with the annual maximum occurring later in the MIPAS data. Around 40 hPa a good agreement is observed, but higher up the comparisons deteriorate again. All data sets (including SMR and EMAC), at least, exhibit the typical phase progression associated with the tape recorder. This is in contrast to the conclusions of that attributed the annual variation in the ACE-FTS data to the monsoon systems. In this work we find some atypical phase behaviour for the ACE-FTS data set between 60 and 40 hPa, which may hint to other influences or a mixture of different signals. However, higher up, the phase progression is again as expected for the tape recorder.

In this work we considered a number of possible reasons for the observational discrepancies between the MIPAS and ACE-FTS data sets. This comprised, on the one hand, differences in the temporal coverage of the tropics and in the vertical resolution between the two data sets. On the other hand, we investigated impacts of special characteristics of the MIPAS retrieval.

Differences in the temporal coverage of the tropics between the two data sets are not a key factor for the observational discrepancies. The MIPAS data have a lower vertical resolution than the ACE-FTS data. Adapting the ACE-FTS data to this vertical resolution by smoothing them with a Gaussian window (with a FWHM of 6 km) exhibits only little influence on the amplitude of the annual variation of δD. For the phase, this adaption yields an influence, however, that increases the differences between the MIPAS and ACE-FTS data sets.

One characteristic of the MIPAS retrieval is a start altitude effect, i.e. an intercorrelation between the lowermost altitude where sensible retrievals are possible (typically determined by cloudiness) and results at higher altitudes. This effect arises from asymmetric kernels and is, as such, not restricted to the data sets considered in this work nor MIPAS results themselves. The start altitude effect exerts an influence on the δD isotopic ratio and the HDO and H2O volume mixing ratios up to about 30 hPa in the MIPAS retrieval. Besides that, the start altitude itself has an annual variation. Both aspects combined result in a profound impact on the amplitude of the δD annual variation below about 40 hPa. At 100 hPa a crude estimate indicates a contribution to the amplitude of the annual variation of about 40 ‰. The temporal variation of the impact is, however, distinctly different from a tape recorder signal.

The reassessment of the MIPAS data indicated that some of the discrepancies to the ACE-FTS data set may originate from the H2O retrieval, which was specifically set up for the joint analysis with HDO . It was found that there is a remaining mismatch in the vertical resolution of the MIPAS HDO and H2O data. The resolution of the HDO data is consistently better in the altitude range from 100 to 10 hPa by 0.3 to 1.4 km. Simple convolution tests with the higher vertically resolved EMAC data, adapted so that there is initially no annual variation in δD, show that such a mismatch actually causes an artificial tape-recorder-like signal in δD. We expanded these tests by convolving the model data with the actual MIPAS HDO and H2O averaging kernels and a priori data to obtain realistic results specific to our problem. These tests comprise more MIPAS data set characteristics than just the vertical resolution mismatch and, importantly, also include the start altitude effect and its impact. Also, these tests exhibited artificial tape-recorder-like signals, but they were even more pronounced. After the correction of the original MIPAS data for the artefact, a tape recorder signal still remains but with a reduced size. On average, the peak amplitude amounts to about 25 ‰ in the lowermost stratosphere, in good agreement with the ACE-FTS and EMAC data sets. In the opposite direction, a convolution of the original EMAC data yields characteristics for the annual variation in δD that are quite similar to those derived from the original MIPAS data itself. This considers in particular the larger amplitudes and the direction of the phase shift. Accordingly, the observed differences between the MIPAS and ACE-FTS data sets can be attributed to the characteristics of the MIPAS retrieval (embedded in the averaging kernels) making it essential to consider them.

Finally, both the simple and more advanced convolution tests show that the vertical resolution mismatch in the MIPAS data artificially enhances the slope of the HDO–H2O correlation in the lower stratosphere. This concerns also the altitude range between 25 km (∼25 hPa) and 30 km (∼12 hPa) considered in the work of . Accordingly, the slope is closer to that predicted for the Rayleigh fractionation process and reduces the necessity to involve strong contributions from convectively lofted ice, as was concluded from the original evaluation of the MIPAS dataset . In a future retrieval version the mismatch in the vertical resolution of the MIPAS HDO and H2O data needs to be fixed.