The spectrometer is a key component of any MWR and can significantly influence its retrieval capabilities. Since 2009, both instruments use the same spectrometer, namely the Acqiris AC240 which is a digital fast Fourier transform (FFT) spectrometer . On SOMORA, it replaced an acousto-optical spectrometer in September 2009, whereas on GROMOS it replaced discrete filter banks in July 2009. In both cases, the time series were homogenized using an overlap period of roughly 2 years, and the pre-2009 time series were corrected with respect to the FFT spectrometer time series e.g.. Whereas both instruments use the same digitizer with the same bandwidth of 1 GHz, it should be noted that the frequency resolution is 2 times higher for GROMOS than for SOMORA because GROMOS uses an in-phase quadrature (IQ) down-converter and digital sideband separation, which results in twice the number of channels . As a result, GROMOS could be more sensitive to ozone at higher altitudes. However, we do not see any significant difference in vertical sensitivity compared to SOMORA, possibly because of the high receiver noise, which could act as a limiting factor for extending the altitude coverage of the two instruments.

The AC240 is still being used in many MWRs; however, it is ageing and has recently been shown to produce a spectral bias compared to more recent spectrometers, most likely impacting ozone retrievals as well . In this contribution, we only focus on the period where both instruments use the AC240, namely from September 2009 to end of 2021. Therefore, both time series should be similarly impacted by the spectrometric bias and thus should not affect the results of the comparisons between GROMOS and SOMORA. This might, however, influence the comparisons against the satellite observations, but there is unfortunately no way to confirm the amplitude of the bias on the ozone profiles at the moment.

GROMOS and SOMORA are both total power radiometers with superheterodyne receivers. They measure the atmospheric ozone emission line around 142.175 GHz and use the heterodyne principle to down convert the incoming radiation (RF signal) to an intermediate frequency (IF) by mixing with a local oscillator frequency (LO) which allows for easier signal processing.

The operation of microwave radiometers requires continuous calibration because their receivers are never perfectly stable e.g.chap. 7. Both instruments use a so-called hot–cold calibration scheme: using a rotating mirror fixed on a path length modulator, they are continuously switching between the atmospheric observation, a hot and a cold calibration target. In both instruments, a heated black-body kept at a constant temperature (Thot≈310 K) is used as hot load, whereas liquid nitrogen (LN2) observation is used as cold load. Both instruments use a Martin–Pupplet interferometer (MPI) to suppress the contribution of the undesired sideband. The pathlength modulator is used to mitigate the standing waves between the receiver and the calibration targets, which are otherwise causing systematic baseline errors on the calibrated spectra. In parallel to the hot–cold calibration scheme, the instruments also perform tipping curve calibration as cross-validation for the LN2 calibration. Assuming linear transfer characteristics, the atmospheric spectral radiance can then be determined and further converted to brightness temperature using Planck's law e.g.chap. 6.

Despite similar designs and raw data contents, the previous calibration routines for GROMOS and SOMORA were different. Therefore, a new routine was designed to harmonize the calibration between the two instruments. The calibration essentially converts the raw spectrometer measurements to radiance intensity and integrates them together on a chosen integration time. For this new routine, the calibration results in two different data levels, namely the calibrated spectrum (level 1a) and the integrated spectrum (level 1b).

Harmonized quality control was introduced in order to identify spurious instrumental signals. It flags the most common technical problems at level 1a (e.g. noise temperature jumps, LN2 refills, LO frequency shifts) and combines them into a single instrumental flag value for level 1b .

Considering instrumental issues and technical interruptions for maintenance (e.g. for LN2 refilling or instrument repairs), GROMOS and SOMORA provided good-quality hourly spectra for 87 % and 89 % of the measurements performed between 2009 and 2021, respectively. This results in more than 80 000 h of comparable retrieved ozone profiles.

In the microwave frequency range, the pressure-broadening effect of atmospheric emission lines is used to retrieve information on the atmospheric constituent profile from the calibrated microwave emission spectra. This so-called retrieval is a well-validated technique that has been successfully applied to temperature; wind; and many trace gases like O3, CO, or H2O chap. 7. Among the different retrieval techniques, we selected the optimal estimation method (OEM) following the formalism described by . This statistical method extracts the best estimate of an atmospheric profile from a set of measurements with noise, a priori information and a forward model. In addition, the OEM enables the characterization of the error budget of the retrievals (Fig. ). In the following, we will briefly present and discuss the new harmonized retrieval setup used for GROMOS and SOMORA. More information on this setup is available in . For detailed information on the OEM or its application to ozone profiling instruments, the reader is redirected to or .

The retrieval errors presented above do not include systematic errors that can arise during the calibration or the retrievals. It is cumbersome to estimate all possible errors on such complex measurement setup and therefore, we decided to perform a sensitivity analysis on the most important error sources using two reference time periods with low (τ≈0.15) and high (τ≈1.3) atmospheric opacities. The uncertainties considered in our study are listed in Table  as well as the perturbations used for the sensitivity analysis. These were determined in different ways for each error source, deriving it either from measurement (e.g. Tcold, sideband ratio, window transmittance) or empirical values (e.g. pointing, spectroscopy).

The uncertainty budget for GROMOS and SOMORA is presented in Fig.  in the case of low tropospheric opacities. The high-opacity cases for both instruments can be seen in Appendix  (Fig. ).

In general, the sensitivity of GROMOS and SOMORA to the different perturbations is very similar. A notable exception is the higher sensitivity of GROMOS to the sideband path length, which is a consequence of its lower intermediate frequency. For both instruments, the total uncertainty is dominated by systematic errors below 2 hPa, whereas the measurement noise becomes quickly dominant above this point. In relative terms, the uncertainty is approximately 9 %–10 % for GROMOS and 7 %–8 % for SOMORA up to the stratopause and then increases significantly in the mesosphere.

In the case of high tropospheric opacity, the ozone emission line gets more attenuated by the tropospheric water vapour absorption. The AVKs get degraded, reducing the sensitivity of the retrievals and leading to higher uncertainties than at lower opacities. As can be seen in Fig. , the atmospheric temperature profile becomes the dominant contribution to the uncertainties below 1 hPa at higher opacity. This is likely due to the increased importance of the water vapour continuum retrieval, which is itself strongly dependent on tropospheric humidity and temperature. In the higher-opacity case, the total relative uncertainty in the stratosphere is 12 %–15 % for GROMOS and 10 %–12 % for SOMORA. In view of the perturbations and error sources considered in this study, these values compare well with similar ozone radiometers at other locations reported in the literature e.g..

GROMOS and SOMORA are located close to each other, have similar viewing directions, and experience similar tropospheric conditions during all seasons (Fig. ). In addition, they have similar altitude range and sensitivity and can therefore be used for direct cross-validation of their time series. The upper panel in Fig.  shows the weekly mean relative differences between GROMOS and SOMORA harmonized data series (note that the lower panel of this figure will be discussed in Sect. ). In general, GROMOS and SOMORA agree well in most of the middle atmosphere, with relative differences mostly lower than 10 % in the stratosphere and lower mesosphere (from ∼ 50 to 0.1 hPa), increasing towards lower and higher altitudes. The higher relative differences at lower and higher altitudes are partly explained by the shape of the ozone VMR profile when intensity is at its maximum in the stratosphere. In general, the lower altitudes are also the most impacted by instrumental baselines, which explains the increase in the differences below 50 hPa, whereas at higher altitudes the instrumental noise becomes the dominant factor and the sensitivity of the radiometers decreases quickly. In addition, the diurnal ozone variations typically become much larger in the mesosphere (e.g. around 20 % compared to a few percent in the stratosphere; ).

We also see some oscillatory patterns in the relative differences, some of which can be identified as clear seasonal patterns (e.g. in the lower stratosphere between 2014 and 2017). These seasonal differences are highlighted in Fig. , which shows seasonal ozone profile comparisons between GROMOS and SOMORA. The mean seasonal differences between the two instruments are lower than 10 % at all seasons and throughout most of the middle atmosphere and show a negative ozone bias from GROMOS in the upper mesosphere (p < 0.05 hPa). In the stratosphere and lower mesosphere, the ozone profiles are well correlated with Pearson's R coefficients mostly above 0.7 at most pressure levels and seasons (Fig. ). However, this is not the case during summer, where we find significantly lower correlation between GROMOS and SOMORA ozone profiles.

Figure  shows scatter plots of their differences in three pressure level domains corresponding approximately to the lower stratosphere, the upper stratosphere and the lower mesosphere (see Table  for the definitions). It shows the net difference in atmospheric opacity between the winter and the summer and highlights the higher ozone variability during the wintertime. Figure  confirms the general good agreement between GROMOS and SOMORA in the middle atmosphere and corroborates the existence of a seasonal bias between the instruments during summertime.

During the summertime, the warmer and wetter troposphere results in a higher opacity. This attenuates the ozone spectral line and thus decreases the retrieval sensitivity during summer. As discussed in Sect. , a higher tropospheric opacity also results in larger uncertainties in the retrieved ozone profile. In case of very hot and humid conditions, the troposphere can become optically thick at 142 GHz, which can prevent the retrieval of ozone profiles. It can be seen in Fig. , which shows higher tropospheric opacity in summertime than during the other seasons. However, Fig.  also shows that the difference in tropospheric opacity at the two sites remains constant, independent of the season. In addition, we investigated the correlations between GROMOS and SOMORA considering only profiles measured at low tropospheric opacity (τ⩽1) and did not see any significant changes in the results. For these reasons, we believe that the summer bias does not result from the higher tropospheric opacities affecting this season.

The reasons for the summer seasonal bias remain unclear, but we assume that they result from seasonal temperature and humidity cycles in the troposphere. Indeed, despite controlled room temperature for both instruments, the higher summer temperatures still influence the room and window temperatures and consequently the instruments (e.g. receiver noise temperature or instrumental baselines). We believe that the hardware components of GROMOS and SOMORA have different sensitivity to such influences, which could explain the seasonal patterns observed in their relative differences and the lower correlation of the ozone profiles during summer.

In addition to these seasonal effects, Fig.  highlights some sudden changes in the differences between the two instruments, most of which can be related to a specific instrumental issue on either instrument. It can be seen for instance in April 2012, where the cold load observation angle was changed on SOMORA, reducing its baseline significantly. Another example is the strong negative ozone differences during summer 2016, which were due to a frequency lock problem in GROMOS. Finally, the large flagged period starting at the end of 2019 marks the beginning of several instrumental issues on SOMORA that were finally solved by the replacement of the LO baseband converter in September 2020. All of these issues have been identified and documented and are flagged accordingly in the new ozone data series. A detailed documentation of the time series can be found together with the data.

Computing trends for GROMOS and SOMORA is out of scope of this contribution but we would still like to provide some first elements toward answering whether this harmonization can help solving the discrepancies previously found between both instruments . Therefore, we compare our new harmonized ozone time series with the previous data version of GROMOS and SOMORA.

Figure  shows the weekly relative differences between the new harmonized series (upper panel) and the previous retrievals (lower panel) from 2010 to 2021. It highlights the significant improvements introduced by the harmonization process in most of the pressure range covered by the radiometers. Among other changes, it corrects the strong positive ozone bias from GROMOS seen in the mesosphere and reduces the stratospheric ozone difference clearly visible in many years of the previous data series at ∼ 10 hPa. The differences between the previous series also showed a quite strong seasonal signal. As the previous processing was different between the two instruments, in particular in the way it was treating the tropospheric attenuation, it gives some confidence that the remaining seasonal bias in the new series is not an artefact introduced by the new retrieval method.

Although the harmonized retrievals improve most of the time period considered, it seems that the problems seen on SOMORA in 2020 are less well treated in the new processing. Indeed, in the previous processing the sine baseline periods were adapted daily during this time whereas the new processing only considered fixed periods. It indicates that the instrumental baselines on SOMORA varied significantly during this period and highlights the need to treat it carefully for further analysis.

From Fig. , it is clear that the harmonized processing significantly reduces the differences between the GROMOS and SOMORA ozone time series. However, the question remains if it can solve the discrepancies found between their respective trends. Of course, the full reprocessing of the series (including the decade 2000–2010) would be needed to fully answer this question, but we present some preliminary results showing the temporal evolution of the ozone differences between both series in Fig. . It shows the weekly mean differences between GROMOS and SOMORA with the previous and new retrieval algorithms in three pressure ranges. Ideally, these differences should be constant to guarantee similar trends from both instruments. Simple linear regressions have been performed on these data and indicate smaller drift intensities at all pressure ranges from the new data processing that are significant above 10 hPa.

As a consequence, the future trends to be derived for this decade from the new series should be in better agreement than with the previous retrievals. However, even with the new series, we still observe a drift between both instruments in the stratosphere, which calls for a careful treatment of spurious data periods for the next trends analysis, as done in .

As the main validation dataset, we use ozone measurements from the Microwave Limb Sounder (MLS) on the Aura satellite launched in 2004 . It is operated by the National Aeronautics and Space Administration (NASA) in the frame of the Earth Observing System and has been used extensively for ozone profile validation over many regions and against many other observing systems e.g..

MLS is a passive microwave radiometer observing the ozone emission line around 240 GHz in a limb sounding geometry. It follows a sun-synchronous orbit which results in two overpasses per day around 01:00 and 13:00 UTC​​​​​​​ over central Europe. In this work, we have used the latest level 2 ozone retrievals (version 5) and the recommended data screening described in . It results in ozone VMR profiles between 261 to 0.001 hPa with a typical vertical resolution ranging from ∼ 2.5 km in the lower stratosphere increasing to ∼ 5.5 km at the mesopause with an accuracy of 5 %–10 % in the stratosphere increasing up to 100 % at 0.01 hPa.

For the following comparisons, we extracted co-located MLS observations to GROMOS and SOMORA. As spatial coincidence criteria, we use ±3.6∘ in latitude and ±10.5∘ in longitude from Bern, an area corresponding approximately to Central Europe. As temporal criteria, we averaged the MWR and the MLS profiles within 3 h time windows and keep only the time windows where both MLS and the MWR have profiles with sufficient data quality.

The MLS vertical resolution of ozone retrievals is much lower than the one from the MWRs. It means that the MWRs will essentially observe a smoothed vertical profile compared to the MLS observations. Therefore, the higher-resolved MLS profiles are convolved with the MWR averaging kernels for the comparisons see. This AVKs smoothing also enables the removal of the influence of the a priori and follows Eq. (): 1xc=xa+A(x-xa), where x is the higher-resolution profile (MLS), xa is the a priori profile from the MWR retrievals, A are the averaging kernels and xc is the resulting convolved profile.

In addition to MLS, we also use the latest release of the Solar Backscatter Ultraviolet Radiometer (SBUV/2) Merged Ozone Dataset (MOD) . This dataset provides daily overpasses over many ground-based ozone measurement stations, including Payerne in Switzerland. It provides stratospheric ozone VMR profiles from 50 to 0.5 hPa merged according to the new MOD v2 Release 1 derived from SBUV and adjusted for the diurnal cycles to an equivalent local measurement time of 13:30. The vertical resolution from the SBUV retrievals is ∼ 6–7 km in the middle and upper stratosphere , which is closer to the vertical resolution of GROMOS and SOMORA in this region. For this reason, contrary to MLS, we do not apply any AVK smoothing to the SBUV measurements for the following comparisons.

Figure  shows weekly averaged GROMOS and SOMORA time series together with SBUV and MLS measurements on three pressure ranges corresponding to the lower stratosphere, upper stratosphere and lower mesosphere. It shows the consistencies of the GROMOS and SOMORA time series and highlights the good agreement of both MWRs with both satellite datasets during the last decade. As these time series are already averaged on given pressure ranges, we did not apply any AVK smoothing on the MLS data at this stage. It is also important to keep in mind that the SBUV daily dataset is adjusted to daytime (13:30), whereas both MLS and the MWRs have both daytime and night-time measurements.

In the stratosphere, clear seasonal patterns are well captured by all datasets, and the higher winter ozone variability is clearly visible at all pressure levels. On timescales of a few weeks, we can see that all four datasets are able to capture the larger ozone variations well not only in the stratosphere but also in the mesosphere where these variations become relatively small compared to the amplitude of the ozone diurnal cycle.

We can see a slight bias of the SOMORA data series in the lower stratosphere. It is especially visible before 2014 and after 2019, as has been mentioned previously. This plot also helps to identify some remaining spurious time periods in the new harmonized series (e.g. GROMOS data in summer 2016). From a qualitative point of view, we do not observe large drifts from any of the datasets with respect to the others. More work will be needed to confirm the stability from both MWRs, but it gives some confidence that both instruments can be used for trends analysis in the decade 2010–2020.

As quantitative validation, we show seasonal comparisons of MWRs profiles with the satellite datasets. In the following, we mostly focus on the MLS time series because it covers the same altitude range as the MWRs and because SBUV only provides daytime measurements. For the period between 2009 and 2021, we obtain more than 7100 co-located profiles between MLS and each MWR, giving approximately 1700 profiles per meteorological season. Figures  and show comparisons between winter (resp. summer) ozone profiles measured by GROMOS, SOMORA, SBUV and MLS. Both figures show the mean seasonal ozone profile from each dataset and the relative differences between MLS and the MWRs with and without AVK convolution. The comparisons for spring and autumn are shown in Appendix  (Figs.  and ).

Both GROMOS and SOMORA show very good agreement with MLS at all seasons and altitudes, with the exception of SOMORA during summertime. Mean seasonal relative differences between the two instruments and co-located MLS profiles are within 10 % in the stratosphere and lower mesosphere (up to ∼ 60 km), corresponding to the expected uncertainties of the MWRs. Above in the mesosphere, the relative differences between the MWRs and MLS grow rapidly and show some oscillations. For most of the mesosphere, the mean seasonal relative differences stay below 50 % for both instruments, but given the errors reported for the MWRs and MLS at these altitudes, we will focus our discussion on the region from ∼ 20 to 60 km. The relative differences with SBUV (not shown) are very similar to those with MLS and are below 10 % in the whole stratosphere for the two instruments.

Figure  again reveals the summer bias mentioned previously. Taking MLS as a reference, this plot indicates that the summer bias in the lower stratosphere is the result of an overestimation of ozone by SOMORA during this season. The reason for this could be a seasonal change in the instrumental baselines that is not taken into account in the retrieval. For both instruments, the differences with the convolved MLS profiles are still smaller in autumn and winter than in spring and summer when the absorption by the troposphere is stronger.

compared the previous GROMOS retrieval dataset to MLS between 2009 and 2016. Similar agreement was found in the middle stratosphere; however, this quickly degraded at lower and higher altitudes. This is in accordance with the results shown in Fig.  and confirms the improvement brought by the new data processing. SOMORA showed similar agreement with MLS in the range 25 to 0.1 hPa between 2004 and 2015 . Below 25 hPa, SOMORA showed a positive bias compared to other datasets that gives confidence that this bias is not related to the new data processing.

Similar comparisons between MWR and MLS has been performed at various locations e.g. and showed similar results to the ones obtained in our study. This is confirmed by the mean ozone VMR relative differences between MWR and MLS given in Table  for the middle atmosphere. Averaged over these pressure ranges and the entire time period, the differences between MLS and the MWRs are less than 5 % in the stratosphere and lower mesosphere.

Overall, SOMORA and GROMOS profiles are in better accordance with the non-convolved MLS than with the convolved MLS profiles. This can be seen for both instruments and at the three pressure ranges from the seasonal plots and in Table . It is not entirely clear why these differences are larger with the convolved MLS profiles, but it does not result from sampling differences (not shown). As it seems especially visible in SOMORA in the lower stratosphere, it could potentially arise from instrumental baselines impacting the AVKs.