Investigation and amelioration of long-term instrumental drifts in water vapor and nitrous oxide measurements from the Aura Microwave Limb Sounder (MLS) and their implications for studies of variability and trends

The Microwave Limb Sounder (MLS), launched on NASA’s Aura spacecraft in 2004, measures vertical profiles of the abundances of key atmospheric species from the upper troposphere to the mesosphere with daily near-global coverage. We review the first 15 years of the record of H2O and N2O measurements from the MLS 190-GHz subsystem (along with other 190-GHz information), with a focus on their long-term stability, largely based on comparisons with measurements from other sensors. These comparisons generally show signs of an increasing drift in the MLS “version 4” (v4) H2O record start5 ing around 2010. Specifically, comparisons with v4.1 measurements from the Atmospheric Chemistry Experiment – Fourier Transform Spectrometer (ACE-FTS) indicate a ~2–3%/decade drift over much of the stratosphere, increasing to as much as ~7%/decade around 46 hPa. Larger drifts, of around 7–11%/decade, are seen in comparisons to balloon-borne frost point hygrometer measurements in the lower stratosphere. Microphysical calculations considering the formation of polar stratospheric clouds in the Antarctic winter stratosphere corroborate a drift in MLS v4 water vapor measurements in that region and season. 10 In contrast, comparisons with the Sounding of the Atmosphere using Broadband Emission Radiometery (SABER) instrument on NASA’s Thermosphere Ionosphere Mesosphere Energetics and Dynamics (TIMED) mission, and with ground-based Water Vapor Millimeter-wave Spectrometer (WVMS) instruments, do not show statistically significant drifts. However, the uncertainty in these comparisons is large enough to encompass most of the drifts identified in other comparisons. In parallel, the MLS v4 N2O product is shown to be generally decreasing over the same period (when an increase in stratospheric N2O is 15 expected, reflecting a secular growth in emissions), with a more pronounced drift in the lower stratosphere than that found for H2O. Comparisons to ACE-FTS and to MLS N2O observations in a different spectral region, the latter available from 2004– 2013, indicate an altitude-dependent drift, growing from 5%/decade or less in the mid-stratosphere to as much as 15%/decade in the lower stratosphere. Detailed investigations of the behavior of the MLS 190-GHz subsystem reveal a drift in its “sideband fraction” (the relative sensitivity of the 190-GHz receiver to the two different parts of the microwave spectrum it observes). 20


Introduction
The Microwave Limb Sounder (MLS, Waters et al., 2006) is one of four instruments launched on NASA's Aura mission (Schoeberl et al., 2006) in July 2004 and has operated essentially continuously from launch to the present. The MLS GHz antenna looks in the "forward" direction from the Aura spacecraft and vertically scans Earth's limb from the surface to~95 km altitude every~26 s, yielding~3500 scans per day. Aura is in a near-polar (98°-inclined) sun-synchronous orbit, enabling MLS observations (Nair et al., 2012), and comparisons versus MLS data have been used as a measure of the stability of other longterm records (e.g. Adams et al., 2014;Eckert et al., 2014). Confidence in the stability of MLS ozone was further underscored in a very thorough study (Hubert et al., 2016) comparing ozone profiles from MLS and a range of other spaceborne sensors to a comprehensive set of coincident measurements from both ozone sondes and ground-based lidar. The authors concluded that the MLS "version 3" ozone record (at least through the May 2013 date then considered) was stable to within ±1.5%/decade in 5 the middle stratosphere, and ±2%/decade in the upper stratosphere, with these small drifts with respect to the lidar and sonde records not being statistically significant at 95%. Larger and statistically significant drifts compared to the lidar and sonde records were seen at lower altitudes, though they were all within ±5%/decade.
In contrast to ozone, we now have strong evidence that in "version 4" (v4) the MLS H 2 O and N 2 O products, both measured by the 190-GHz receiver, show larger and statistically significant drifts. This paper quantifies drifts in these and other species 10 measured by the MLS 190-GHz receiver, discusses factors that may be giving rise to the drifts, describes efforts by the MLS team to ameliorate those factors, and provides updated guidance to users of the affected MLS products. These drifts were first noted by Hurst et al. (2016)  other sensors that also show evidence for a positive drift in MLS H 2 O, albeit a slower one than that indicated by the frost point record. Section 2 also quantifies a drift between the MLS standard ozone product, measured at 240 GHz, and a diagnostic ozone product derived from an ozone line measured by the MLS 190-GHz receiver. Section 2 finishes by examining MLS N 2 O measurements, which are drifting to smaller values. Section 3 provides background on the MLS 190-GHz subsystem and its 20 measurements and describes insights obtained into the probable underlying contributors to the observed drifts. Section 3 also details how, in the new MLS "version 5" (v5) data record, much of the H 2 O drift has been ameliorated, and drifts in N 2 O and 190-GHz O 3 have been reduced. Finally, section 4 provides a summary and, most importantly, guidance for users of the MLS H 2 O and N 2 O datasets as to how the drifts should be factored into future studies using those data. 25 This section examines observed drifts in the MLS H 2 O, N 2 O, and 190-GHz O 3 products. MLS 190-GHz signals are also the source of the MLS HCN product and contribute to the HNO 3 product. However, the behavior of the latter two species is better considered as part of the discussion of the origin of the drifts in the first three products. Accordingly, consideration of the MLS HCN and HNO 3 products is deferred to section 3.

Comparisons with the balloon-borne hygrometer record
In contrast with the lower and middle troposphere, for which a wealth of well-characterized observations exist to validate spaceborne humidity sensors, in the upper troposphere and lower stratosphere correlative observations of water vapor are far less frequent and sparser. Although both operational radiosondes and spaceborne Global Navigation Satellite System-Radio 5 Occultation sounders are able to profile temperature well into the stratosphere, neither provide scientifically useful water vapor information in or above the upper troposphere. We note that several in situ instruments have obtained lower stratospheric water vapor observations from high-altitude aircraft, and these observations have been used for validation of the MLS water vapor product (e.g., Read et al., 2008;Weinstock et al., 2009). However, their campaign-based sampling is extremely sparse both temporally and spatially, severely hampering their application to validation of long-term variability in global spaceborne 10 observations.
The longest near-continuous systematic record of in situ stratospheric water vapor observations comes from balloon-borne frost point hygrometer instruments (Mastenbrook and Oltmans, 1983;Vömel et al., , 2016Hall et al., 2016;Hurst et al., 2011Hurst et al., , 2014Hurst et al., , 2016 launched from Boulder, Colorado. Additional frost point instruments have been routinely launched from a small number of locations, though not for as long a period. The typical uncertainty on frost point measurements of water vapor 15 mixing ratios in the lower stratosphere is better than 6% (Hall et al., 2016;Vömel et al., 2016). Most of this uncertainty is in the form of "random" error due to oscillations in the feedback loop that maintains a stable layer of frost on a temperaturecontrolled mirror. Absolute accuracy is estimated to be significantly better. Specifically, through the careful calibration of each instrument, <0.05 K inaccuracies in frost point temperatures result in systematic errors of < 1% in stratospheric water vapor partial pressures. However, biases in radiosonde pressure measurements can increase mixing ratio biases to 4% or greater in 20 the stratosphere (Hall et al., 2016).
Frost point sonde observations have underpinned the validation of stratospheric water vapor measurements from the MLS instruments Read et al., 2007) and other spaceborne sensors. As reported by Hurst et al. (2016), comparisons between the frost point record and MLS v4 H 2 O indicate a divergence between these two water vapor datasets in the lower stratosphere, commencing around 2010, with the MLS values increasing relative to the frost point record. Figure 1 presents an 25 update of the Hurst et al. (2016) comparisons between MLS v4 and frost point measurements at Lauder (New Zealand), Hilo (Hawaii, USA), and Boulder (Colorado, USA). These timeseries have been extended with additional years and are shown with the sign convention switched compared to those presented by Hurst et al. (2016), in order to be consistent with later figures in this paper. As with the original study, the MLS and frost point data are quality screened using established rules described in Livesey et al. (2020) and Hurst et al. (2016), respectively, and the frost point profiles have been smoothed by the MLS "least 30 squares operator" and averaging kernel.
The year 2010 was chosen as the starting point for the fitted drift in light of the findings of the Hurst et al. study, substantiated by the observed behavior of the MLS 190-GHz receiver subsystem as discussed in section 3. These and most other fits described in this paper are simple linear regressions of differences between individual MLS Level 2 profiles and coincident measurements. Except where noted otherwise, all of the statistical fits in this paper establish significance at 95% using a block bootstrap resampling of the fit residuals (e.g., Froidevaux et al., 2019, and references therein) that uses one-year blocks and allows for block replacement. For succinctness, rather than referring to the possibly asymmetric 95% confidence intervals from this analysis (which would require two numbers to describe), we will use ±2σ bounds to describe uncertainty in drifts, where the standard deviations, σ, are those resulting from the bootstrapping analysis. The differences between the 2σ and 2.5%/97.5% thresholds are small in the cases shown here.
Given that each frost point sonde is an independently calibrated instrument, and the calibration scheme is referenced back to National Institute of Standards and Technology (NIST)-standard temperature sensors as well as to a small archive of previously 5 calibrated mirror thermistors, it seems unlikely that this drift reflects any systematic evolution of the calibration of the frost point sonde record. That said, we note the findings of Lossow et al. (2018), particularly their Figure 8, which shows two multi-year periods during 1987-2001 when the frost point record at Boulder departed by 5-10% from both model-based and satellite-based measures of stratospheric water vapor. We believe that these 2-and 4-year periods of disparity may have resulted from atypically large, batch-dependent biases in radiosonde pressure sensors that propagated directly to the frost point 10 mixing ratios. Such pressure biases became easily correctable after Global Positioning System (GPS) receivers were added to radiosondes after 2001, and similar disparities were not observed again until~2010, when MLS water vapor retrievals began to drift.

Comparisons with other spaceborne sensors
Another source of near-continuous long-term stratospheric water vapor observations is the spaceborne Atmospheric Chemistry drifts show a minimum value at 46 hPa, above which a fairly uniformly (and typically significant) positive drift of around 2-3%/decade is found. We note that these drifts are mostly smaller than those seen in the frost point comparisons (Figure 1), particularly at the higher altitudes (e.g., 22 hPa). By contrast, the 2005-2010 period shows essentially no statistically significant 30 drifts between MLS v4 and ACE-FTS (a few levels/latitudes have > 2σ drifts, though no more than might be expected when using a~95% confidence criterion for each of~24 levels over three latitude bands). started providing a water vapor product from the lower stratosphere to the thermosphere (Rong et al., 2019). These data are estimated to have 4% precision on an individual profile below 60 km and have been shown to agree with those from other sensors to within~20% or better. Figure 3 shows estimated drifts between MLS v4 and SABER version 2.07 water vapor using the same coincidence criteria as used for ACE-FTS in Figure 2. In contrast with ACE-FTS and the frost point observations, comparisons between MLS v4 and SABER show no statistically significant drift except perhaps in a narrow region around 5 3 hPa, where a 2-3%/decade drift is seen in the southern and tropical latitude bands (a similar but not statistically significant pattern is seen in the northern latitudes also). The 2σ uncertainties on the estimated lower stratospheric drifts are significantly larger than those for the corresponding ACE-FTS comparisons. In many cases the 2σ uncertainty in MLS versus SABER drift encompasses the drift estimated from the MLS and ACE-FTS comparisons.  1-year block bootstrap 95% confidence range for the slope). Drift rates (converted from ppmv to percent) and an associated 2σ uncertainty are quoted in each panel.

Comparisons with ground-based observations
Water vapor in the upper stratosphere and lower mesosphere can be measured from the ground by zenith-viewing microwave spectrometers. Figure Table Mountain comparisons does not quite encompass such a drift. However, the large gap in the timeseries likely precludes robust interpretation of this comparison. Comparisons at other altitude regions (not shown) give similar results.

Inferences from thermodynamic and microphysical studies
Additional confirmation of a probable drift in the lower-stratospheric MLS v4 water vapor observations comes from a microphysical study building on the work of Lambert and Santee (2018), who examined MLS observations of polar winter lower stratospheric nitric acid (HNO 3 ) and water vapor, along with coincident polar stratospheric cloud (PSC) observations from the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) instrument on the NASA/CNES Cloud-Aerosol Lidar and 5 Infrared Pathfinder Satellite Observation (CALIPSO) mission. MLS and CALIOP data were used, in conjunction with a microphysical model of the equilibrium thermodynamics of supercooled ternary solutions (STSs) and ice clouds, to derive an independent measure of atmospheric temperature in the vicinity of PSCs. These estimated temperatures were then used to quantify the accuracy and precision of various reanalysis stratospheric temperature datasets. The left-hand panel of Figure 5 shows timeseries of average differences between this temperature estimate and temperature from the European Center for Colors indicate different pressure levels as indicated in the legend. MLS and CALIOP measurements before 2008 and after 2018 were not sufficiently colocated to be included in this analysis. The implied drift in reanalysis temperature, assuming no drifts in the MLS and CALIOP data, is estimated to be -0.35±0.04 K/decade. Only rarely does the Arctic winter polar vortex get cold enough to form large-scale water-ice PSCs, so this analysis is limited to the Antarctic. 15 Conversely, this analysis can be recast into one that provides insights into the MLS water vapor drift by instead assuming that the ERA-I temperatures are unbiased and, more critically, stable over the timeframe considered. The latter assumption is quite reasonable given that, over this period, a consistent portfolio of observations were used in the ERA-I reanalysis (e.g. Long et al., 2017). The microphysical framework described above can then be used to derive an independent estimate of stratospheric water vapor in Antarctic regions containing water-ice PSCs. The right-hand panel of Figure 5 shows the relationship between 20 water vapor estimated using this approach and MLS v4 H 2 O. This relationship exhibits a 6.7±0.8%/decade drift, with MLS v4 again moistening. In this case the uncertainty corresponds to a simple 1σ estimate from a linear least-squares fit (i.e., with no bootstrapping).

Water vapor summary
We have found strong evidence for a drift in the MLS v4 water vapor dataset starting around 2010. Balloon measurements 25 indicate statistically significant drifts as large as 11%/decade in the lower stratosphere. Comparisons with ACE-FTS show a smaller 2-3%/decade drift over much of the vertical range, increasing to as much as~7%/decade in the lower stratosphere.
Consideration of PSC microphysics in the Antarctic corroborates such a drift, and comparisons with ground-based microwave observations of lower mesospheric water, while not showing any statistically significant drift, are largely not inconsistent with a~2%/decade drift. Comparisons with SABER lack sufficient statistical significance to comment on the drifts. We also note 30 that Randel and Park (2019), who examined the time-lagged relationships between tropical tropopause cold-point temperature and lower stratospheric humidity, also found evidence for a positive drift in MLS v4 water vapor, peaking at 7-9%/decade in the northern hemisphere lower stratosphere, consistent with the findings above.

Ozone measured by the MLS 190-GHz receiver
Ozone has many strong lines throughout the microwave spectrum, with multiple lines in each of the spectral regions observed by MLS. The MLS ozone "standard product" is retrieved from radiances around 240 GHz, primarily selected because this region includes the strongest lines, providing information over a broader altitude range than is available from the other receivers, spanning from the upper troposphere to the upper mesosphere. Ozone profiles are also retrieved from each of the other four 5 MLS spectral regions, including from strong signals in the 190-GHz and 640-GHz receivers. Ozone signals at 118 GHz and 2.5 THz are noisier and, in the latter case, time limited, so they are not considered here. The MLS 190-and 640-GHz ozone data products can be found in the "L2GP-DGG" data files.
Initial validation of MLS stratospheric ozone focused on the version 2 data set Froidevaux et al., 2008), and average differences between that version and subsequent versions have been small, apart from some vertical oscillations  over the 100-22 hPa pressure range are less consistent in sign, though still statistically significant in some cases. However, the low abundances of ozone at these levels make reliable assessment of fractional drifts challenging.

N 2 O from MLS 190-and 640-GHz measurements
In MLS Level 2 versions 3 and earlier, the standard product for N 2 O was obtained from the 640-GHz receiver. However, this measurement had to be discontinued in June 2013 because of rapid degradation in the N 2 O-specific elements of the 640-GHz   3 Insights into causes of the drifts, and pathways to their partial amelioration

A drift in 190-GHz "sideband fraction"
The MLS instrument observes thermal microwave emission from the Earth's limb. The radiance signals from which all the MLS products discussed in this paper are derived are received by a 1.6-m primary "GHz" antenna. This antenna scans the limb vertically from the surface to~90 km altitude every~24.6 s, looking in the forward direction from the Aura spacecraft where S atm (f ) is the atmospheric spectrum (e.g., between 176 and 208 GHz for the MLS "190-GHz" signals, for which f LO = 15 191.9 GHz), and S IF (f IF ) is the resulting IF spectrum (between 0 and~16 GHz for the 190-GHz example). The terms α and β are IF-dependent "sideband fractions" describing the contribution to the IF signal from atmospheric signals below and above the LO frequency, respectively. In most cases these sideband fractions are around 0.5, while for the 118-GHz receiver, the upper sideband signals are blocked by a waveguide filter, giving α 1, β 0. In practice, factors related to the MLS calibration scheme and non-neligible thermal emission from the primary antenna and other optics, dictate that the sum of α 20 and β typically falls slightly short of unity, but that has no consequence for the issues discussed here. Figure 11 shows illustrative simulated atmospheric spectra and intermediate frequency spectra for the 190-GHz signals that are central to the MLS products affected by the drift described above. The spectral lines that give rise to the MLS 190-GHz products are listed in Table 1, which also summarizes the drift direction for each product. As summarized in the  bands 4 and 5 measure HNO 3 and ClO, respectively, whose signals are too small to be seen clearly on this plot.
compared to MLS observations in other spectral regions and to other data records. This strongly implies that a slow change in sideband fractions, with the lower sideband being increasingly favored, is at least partly responsible for the observed behavior.

Other MLS 190-GHz data products
The standard product for HNO 3 uses radiances from lines in the 190-GHz region for pressures smaller than 22

Quantifying sideband fraction
Although the MLS sideband fractions were measured as part of the pre-launch MLS calibration (Jarnot et al., 2006) upper sideband signal, by contrast, is very small at these altitudes and is largely determined by dry air and water vapor continua. Given knowledge of the atmospheric temperature profile (e.g., from MLS radiance observations in other spectral regions, and/or from meteorological analysis fields), the expected radiance signal in both sidebands can be readily estimated using a "forward model" radiance calculation and compared to the observed radiances. Such forward model calculations are routinely performed as part of the MLS Level 2 processing. 5 Figure 12 shows a timeseries of such radiance differences (observed minus calculated, i.e., the difference between the measured radiances and those predicted from forward model calculations and pre-launch measurements of α and β according to equation 1), along with a comparable calculation for the channels in the center of the 235-GHz O 3 line measured with the MLS 240-GHz receiver. The 183-GHz radiance differences have a baseline of~8 K after launch, show a significant~0.7 K drop in late 2004, and then a comparable increase in early 2006. Those changes are followed by a slow increasing trend from 10 ~8 K in early 2006 to~10 K in late 2019 (with a strong increase in the late-2018 timeframe). In contrast, 235-GHz radiance differences are stable to within~0.1 K or better, and the absolute offset is only~1 K, which is reasonable given the expected accuracy of the MLS-derived temperature fields used (which also include information from the NASA Global Modeling and Assimilation Office's "Forward Processing for Instrument Teams" data stream) and other assumptions in this calculation.
Other than a change in sideband fractions, the most plausible alternative factor that could give rise to the post-2010 drifting 5 behavior of the 183-GHz signals shown in Figure 12 would be some time-dependent variation in the MLS temperature dataset (which would need to change by an amount comparable to the~2 K change in radiance residual). We see no signs of such a drift in comparisons of the MLS temperature product with other datasets. Furthermore, such a temperature drift would affect both the 190-and the 240-GHz calculations in Figure 12 in a similar manner. The right-hand axis in each panel of Figure 12 shows the approximate sideband fraction change (compared to the pre-launch calibration value) that would account for the difference sum that is close to unity, obtained from instrument calibration.
The drop in the 190-GHz sideband fraction seen in late 2004, with a "recovery" in early 2006, is associated with a period when the MLS optical bench was maintained at a different temperature compared to that selected for the rest of the mission. 15 Similarly, the discontinuity in late 2018 corresponds to a change in an attenuator setting in the MLS band 2 signal chain. The fact that this receiver's sideband fraction appears to be affected by such changes in operating conditions implies that it is not as stable as was intended, and thus it is likely also affected by other parameters. In other words, the observed drift in sideband fractions is likely to be a symptom of aging in some more fundamental receiver parameter that is not directly measured within the MLS instrument. Further, we should not assume that the sideband fraction drift is the only consequence of such aging -20 other calibration parameters may be drifting also, in less obviously identifiable ways, contributing to the drifts seen in the MLS 190-GHz observations.
The~8 K post-launch offset in the 183-GHz radiance differences in the saturated line-center region, corresponding to ã 6.5% change in sideband fraction, is also notable. Again, this is likely to be linked, at least in part, to instrument operating temperature, as ground-based calibration was performed at a different instrument temperature than is experienced in flight. The 25 fact that no such offset is seen for the 240-GHz receiver again speaks to its greater degree of resilience and stability. Note that the 8 K radiance difference between measured and expected radiances seen in the line center occurs in an altitude region where radiances near the line center are not used in the 190-GHz retrievals, specifically because they do not convey information on water vapor, as mentioned earlier.
Sideband fractions vary across the spectral region observed (e.g., the 190-GHz lower sideband fractions established in  These results are scaled from the differences between two retrievals of a full day of simulated MLS observations. For one of these simulations, a perturbation has been applied to the 190-GHz sideband fraction; the other is an unperturbed "control" simulation. These simulations are similar to those discussed by Read et al. (2006) 5 but updated for v4. Results are expressed in terms of a percentage of the retrieved value (from the control retrieval).
The simulated changes are generally similar to the drifts reported in the comparisons described in earlier sections (thin lines in Figure 13), particularly for water vapor in the middle and upper stratosphere. In the case of 190-GHz O 3 and N 2 O, the simulated sideband fraction change impacts underestimate the magnitude of the drift signatures by around a factor of two. However, the vertical structure of these changes generally agrees with that in the observed drifts in the lower to middle 10 stratosphere, with a strong vertical gradient in the 190-GHz N 2 O impacts.

The MLS version 5 algorithms and dataset
The work to characterize and diagnose the drift in the MLS 190-GHz data products has been the principal focus of the MLS team in developing the algorithms and software for the v5 release of the MLS dataset.
The calculations underlying Figure 12 have been used to generate new time-dependent sideband fraction calibration infor-15 mation (with monthly granularity) throughout the Aura mission. These evolving sideband fractions include not only the slow time drifts, but also the jumps seen earlier in the mission, as well as the~6.5% initial offset from the values derived from pre-launch calibrations. In the absence of insights into the sideband fraction beyond the 183-GHz line center region used for this diagnosis, the same percentage changes are assumed to apply uniformly across the 190-GHz receiver's spectral range. The analysis of Figure 13 indicates that correction of the sideband fraction drift should go a long way toward improving the drift seen between MLS and ACE-FTS H 2 O in the mid-to upper stratosphere, but perhaps not the larger (~7%/decade) drifts seen at lower altitudes in mid-latitudes, nor those seen in comparison to the frost point record.

5
An additional focus of the v5 development was to alleviate a previously noted  significant dry bias in MLS H 2 O in the region below the tropopause in v4 and all previous versions. The bias is of order 20% in the tropics, smaller at higher latitudes. The change from the pre-launch value of the 190-GHz sideband fraction increased water vapor in this region by around 10%, somewhat alleviating the bias. Moreover, it was found that, of the various sources of systematic error giving rise to biases in the MLS water vapor observations , another error term with an impact fingerprint that In addition to improving the H 2 O dry bias below the tropopause, these changes had the encouraging benefit of essentially 15 eliminating the v4 bias between the MLS 190-GHz ozone product and the ozone measured by the 240-and 640-GHz receivers.
Another goal of the MLS v5 development was to reduce the clear high bias in the 190-GHz N 2 O product at 100 hPa, which rendered it unsuitable for scientific use. Development of a separate retrieval phase (focused on N 2 O) that changed the way spectral "background" signals were accounted for mitigated that issue. However, it was found that the sideband fraction offset from the pre-launch value implemented as part of the time-varying calibration, as discussed above, significantly increased 20 the N 2 O in the lower stratosphere once again. Accordingly, the N 2 O-focused retrieval phase uses a time-dependent sideband fraction that reflects only the drift, not the initial offset from pre-launch measurements. Given the proximity of the N 2 O line to the region where the analysis in Figure 12 provides insight into sideband fraction, this inconsistency is clearly unsatisfying and warrants further investigation. However, while the absolute value of lower stratospheric N 2 O as measured by MLS is subject to some uncertainty, the morphology is expected to be robust. 25 As a preliminary assessment of drifts in MLS v5, Figure     In contrast, the drifts in the Boulder comparisons are diminished less, by 0.5%/decade at 100 hPa and about 1-2%/decade at higher altitudes.
Broadly speaking, therefore, the drift in water vapor seems to be reduced by about 2-4%/decade over much of the vertical range, though there is some variation in the degree of reduction with altitude and latitude. Such a reduction is sufficient to place the MLS v5 versus ACE-FTS drifts below the level of statistical significance. The frost point comparisons, having indicated 5 stronger drifts in MLS v4 than seen in ACE-FTS comparisons, continue to indicate statistically significant, though reduced, drifts in the MLS v5 water vapor.
The fact that a statistically significant drift remains between MLS v5 water vapor and the frost point measurements, and yet no statistically significant drift is seen in comparisons between MLS v5 and ACE-FTS water vapor, is striking and, thus far, unexplained. This is consistent with the notable disparity in MLS v4 drift estimates derived from the frost point and ACE-10 FTS records (Figure 1 versus Figure 2). We note historical challenges associated with quantifying water vapor trends from the Boulder record (Kunz et al., 2013) and some debate surrounding the extent to which water vapor trends over Boulder are representative of trends more broadly (Hegglin et al., 2014;Lossow et al., 2018). However, given that, for Figures 1 and 15, only MLS profiles in close proximity to the Boulder site are being considered, the issue of Boulder's representativeness should not come into play, as it applies equally to the frost point observations and the MLS observations sampled in that locale, at least to first order. Figure 16 shows the equivalent comparisons to those in Figure 14 for N 2 O (with v4 reproduced from Figure 10). Here, v5 has reduced the magnitude of the drift, which was larger than that for H 2 O, particularly at lower altitudes. However, a substantial 5 (as large as 10%/decade) and statistically significant decreasing drift remains at pressures of 20 hPa and greater. These figures are consistent with the analysis in Figure 13, which indicates that the expected impact of sideband fraction change, while a good match to the observed drift in water vapor, only accounts for about half of the observed N 2 O drift.
It is clear that the MLS v4 water vapor product is subject to a slow positive drift, likely starting around 2010. A 2-7%/decade drift is seen in comparisons with ACE-FTS v4.1, with the largest drifts between 100 and 46 hPa. Comparisons with balloonborne frost point hygrometers indicate a larger~10%/decade drift at these levels. Drifts are also seen in measurements of N 2 O and O 3 from the same MLS 190-GHz subsystem used to measure H 2 O. Detailed study of the MLS 190-GHz radiances 5 reveals a drift (and post-launch offset) in the "sideband fractions" of around 1.3% over the last decade that accounts for a significant fraction of the observed drift between MLS and other sensors, with the remainder likely due to other symptoms of an aging receiver. Drifts in 190-GHz sideband fractions will also impact the MLS HCN product and the MLS NNO 3 product for pressures smaller than 22 hPa. However, no clear signals of such drifts are seen in comparisons to ACE-FTS.
While the existence of such a drift is clearly unfortunate, we note that it did not start to appear until around six years into 10 MLS on-orbit operations, which corresponds to the original design life of the instrument (now more than 17 years into a nominal five-year mission). Further, the analyses described above developed to characterize and understand this drift have had the additional benefit of underscoring the stability of the other MLS receivers at 118, 240, and 640 GHz. We emphasize that this drift does not affect the MLS O 3 , CO, ClO, HCl, HOCl, CH 3 Cl, CH 3 CN, CH 3 OH, BrO, HO 2 , OH, SO 2 , temperature, geopotential height, or cloud ice, standard products. We further note that this drift does not call into question the fundamental Scientists are strongly advised to use MLS v5 data in preference to v4 for all products, but particularly for H 2 O, N 2 O, HCN, and HNO 3 , in light of the work to correct the offsets and drift in the MLS 190-GHz receiver calibration. Users are still advised to apply caution when interpreting the temporal changes in v5 H 2 O, N 2 O, HCN, and HNO 3 on multi-year time scales, including long-term trends, and to undertake such studies only in close consultation with the MLS team. On the other 25 hand, studies of spatial and seasonal-to-annual variability in these products, and investigations such as the speed of the "tape recorder" (Mote et al., 1995(Mote et al., , 1996 and the impact of the Quasi Biennial Oscillation, should be largely unaffected. The MLS team is planning to continue processing of incoming data with both the v4 and v5 algorithms for the foreseeable future, as continued comparisons between the two versions provide information on the evolution of the drift and its correction.
Data availability. MLS data are available at the NASA GSFC DISC, https://disc.gsfc.nasa.gov. Frost point sonde data can be obtained from
Author contributions. NJL oversaw the MLS project and wrote much of the text. WGR led the investigation into the MLS 190-GHz drifts and the development of the MLS v5 algorithms. LF led the analysis related to the MLS ozone products and much of the analysis for nitrous oxide.
AL performed the comparisons to ACE-FTS and SABER and led the microphysical studies. MLS contributed to the microphysical studies.
MJS and LFM, along with all the aforementioned authors, contributed to the investigation of the drifts and development and testing of the v5 algorithms. RFJ provided insights in his role as MLS instrument scientist. PAW led the implementation of the MLS v5 algorithms and testing 5 of the associated software. DFH contributed data and guidance for the frost point measurements, with KAW and PES similarly contributing for ACE-FTS, and GEN for the ground-based microwave. All authors provided extensive comments and guidance on the manuscript.
Acknowledgements. We thank James Russell III and Pingping Rong of Hampton University for making the SABER H 2 O record available and for helpful comments on this manuscript. We thank the handling editor, Dr. Rolf Müller, for his helpful suggestions on the initially submitted version of this manuscript. We also thank the two anonymous reviewers and Dr. Quentin Errera for their comments, which further improved  Tech., 9, 4447-4457, https://doi.org/10.5194/amt-9-4447-2016Tech., 9, 4447-4457, https://doi.org/10.5194/amt-9-4447- , 2016