The MIPAS global climatology of BrONO2 2002–2012: a test for stratospheric bromine chemistry

We present the first observational dataset of vertically resolved global stratospheric BrONO2 distributions from July 2002 until April 2012, and compare them to results of the atmospheric chemical climate model EMAC. The retrieved distributions are based on space-borne measurements of infrared limb-emission spectra recorded by the Michelson Interferometer for Passive Atmospheric Sounding (MIPAS) on Envisat. The derived vertical profiles of BrONO2 volume mixing ratios represent 10◦ latitude bins and three-day means, separated into sunlit and observations in the dark. The estimated uncertainties 5 are around 1–4 pptv caused by spectral noise for single profiles as well as for further parameter and systematic errors which may not improve by averaging. Vertical resolutions range from 3 to 8 km between 15 and 35 km altitude. All leading modes of spatial and temporal variability of stratospheric BrONO2 in the observations are well replicated by the model simulations: the large diurnal variability, the low values during polar winter as well as the maximum values at midand high latitudes during summer. Three major differences between observations and model results are observed: (1) a 10 model underestimation of enhanced BrONO2 in the polar winter stratosphere above about 30 km of up to 15 pptv, (2) up to 8 pptv higher modelled values than observed globally in the lower stratosphere up to 25 km most obvious during night, and (3) up to 5 pptv lower modelled concentrations at tropical latitudes between 27 and 32 km during sunlit conditions. (1) is explained by the model missing enhanced NOx produced in the mesosphere and lower thermosphere subsiding at high latitudes in winter. This is the first time that observational evidence for enhancement of BrONO2 caused by mesospheric 15 NOx production is reported. The other major inconsistencies (2,3) between EMAC model results and observations are studied by sensitivity runs with a 1d model. These tentatively hint to a model underestimation of heterogeneous loss of BrONO2 in the lower stratosphere, a too low simulated production of BrONO2 during day as well as strongly underestimated BrONO2 volume mixing ratios when loss via reaction with O(P) is considered additionally to photolysis. However, considering the uncertainty ranges of model parameters and of measurements, an unambiguous identification of the causes for the differences 20 remains difficult. The observations have also been used to derive the total stratospheric bromine content relative to years of stratospheric entry between 1997 and 2007. With an average value of 21.2±1.4 pptv of Bry at mid-latitudes where the modelled adjustment from BrONO2 to Bry is lowest, the MIPAS data agree with estimates of Bry derived from observations of BrO as well as from MIPAS-Balloon measurements of BrONO2. 25 1 https://doi.org/10.5194/acp-2021-535 Preprint. Discussion started: 11 August 2021 c © Author(s) 2021. CC BY 4.0 License.

Instrumental uncertainties are estimated at 3% for the instrument line shape expressed as linear loss of modulation efficiency toward the maximum optical path difference of the interferometer (ILS), 1% for radiometric gain calibration (RadGain), and 300 m for tangent height knowledge (Htang). The uncertainty of ECMWF temperatures (Temp) has been set to values of 2 K below and 5 K above 35 km altitude. The uncertainty of the BrONO 2 spectroscopy has been assumed as 5%, which is on the conservative side considering the 2% (1-σ) error estimation given in Wagner and Birk (2016). Further errors refer to 130 the spectroscopic parameters of interfering gases. For those, we have assumed uncertainties of 5% for species described by cross-sections (SpecXitf), 5% for intensities (SpecINTitf) and 10% for the half widths of the line-parameters (SpecHWift).
These assumptions are within the typical errors provided in the database (Gordon et al., 2017). A further error term describing the retrieval from averaged spectra (NonLin) has been estimated on the basis of dedicated retrieval simulations as detailed in Höpfner et al. (2009). For this estimate the values used for the tangent altitude scatter of single observations were set to 400 m 135 and 300 m (1-σ) during P1 and P2, respectively.
The total error estimate, as calculated by quadratic combination of the single components is given by the blue lines in Fig. 1. Around that, the blue shading indicates the variability of the estimated errors between all latitude bands. In general, the estimated total errors vary between about 1 pptv and 4 pptv, independent of day-or night-time observations. They appear to be slightly smaller during P1 compared to P2, which is probably due to the better spectral resolution during P1. 140 As a further diagnostic measure of the retrieval, Fig. 2 shows the vertical resolution as a function of altitude. It has been calculated by dividing the retrieval grid width of 1 km by the diagonal elements of the averaging kernel matrices (Rodgers, 2004). The vertical resolution is about 3 km at 15 km altitude and becomes coarser with altitude, reaching 8 km at 35 km altitude. The vertical resolution is generally finer at the tangent altitudes and coarser at retrieval levels between the tangent altitudes. Conversely, the retrieval noise is larger at the tangent altitudes and smaller at altitude levels in between. These effects 145 are only visible when a retrieval setup is chosen where the retrieval grid is finer than the tangent altitude spacing. Since in period P1 the tangent altitude grid is fixed, this effect survives the averaging, leading to a zigzag profile of the vertical resolution. In contrast, in P2 the tangent altitude grid varies with latitude, and the zigzag features of vertical resolution average out. Figure 1. Retrieval error estimates from two 3-day periods during P1 (top) and two periods during P2 (bottom) for dark (left) and sunlit (right) conditions. Considered error sources are the uncertainties of the instrumental line shape and radiometric gain calibration (ILS, RadGain), the pointing knowledge (Htang), assumed temperature profiles (Temp), spectroscopic errors of BrONO2 absorption cross sections (SpecBrONO2) and errors in cross-sections (SpecXitf), line half-widths (SpecHWift) and line intensities (SpecINTitf) of interfering species, as well as the error due to the applied technique of retrievals from averaged spectra (NonLin) and the spectral noise of the instrument (Noise). The total error (Total Err) has been determined by quadratic combination of all single error components while the combined parameter and systematic error (Tot paraerr) considers all uncertainties except the spectral noise. The blue filled space around the total error curves indicates the areas in which 90% of the total error profile estimates fall into.
6 https://doi.org/10.5194/acp-2021-535 Preprint. Discussion started: 11 August 2021 c Author(s) 2021. CC BY 4.0 License. Figure 2. Examples for the vertical resolution of the MIPAS BrONO2 retrieval as derived from the diagonal elements of the averaging kernel matrices. Given curves are averages over all latitude bands for the given periods during P1 and P2. The zigzag during P1 is caused by the constant tangent altitude grid while during P2, the variation of tangent altitudes with latitude smear out this effect. Note that the vertical resolution is generally finest at the tangent points and coarsest in between. 7 https://doi.org/10.5194/acp-2021-535 Preprint. Discussion started: 11 August 2021 c Author(s) 2021. CC BY 4.0 License.

Atmospheric modelling
We have compared the MIPAS BrONO 2 dataset with a multi-annual simulation from the chemical climate model ECHAM/

150
MESSy Atmospheric Chemistry (EMAC) (Jöckel et al., 2010). Within EMAC, the interface Modular Earth Submodel System (MESSy) links the sub-models describing tropospheric and middle atmospheric processes to the dynamical core, the fifthgeneration European Centre Hamburg general circulation model ECHAM5 (Roeckner et al., 2006). We have used EMAC (ECHAM5 version 5.3.02, MESSy version 2.52) in the T42L90MA-resolution with 90 vertical hybrid pressure levels from the ground up to 0.01 hPa (∼80 km) and a horizontal resolution of ∼ 2.8 • × 2.8 • latitude × longitude. The sub-models MECCA 155 (Sander et al., 2005) and MSBM (Kirner et al., 2011) simulate gas-phase chemistry and polar stratospheric clouds including heterogeneous reaction rates, respectively. To reproduce realistic conditions for comparison with the observations, the model run was nudged towards the ECMWF reanalysis ERA-Interim (Dee et al., 2011) by a Newtonian relaxation technique of surface pressure, temperature, vorticity, and divergence above the boundary layer and below 1 hPa (van Aalst, 2005). We have applied a comprehensive chemistry set-up from the troposphere to the lower mesosphere with more than 100 species involved in gas-160 phase, photolysis, and heterogeneous reactions on liquid sulfate aerosols, nitric acid trihydrate (NAT), and ice particles. Rate constants for gas phase reactions have been taken mainly from Atkinson et al. (2007) and the Jet Propulsion Laboratory (JPL) compilation (Sander et al., 2011). Photochemical reactions of short-lived bromine containing organic compounds CHBr 3 , CH 2 Br 2 , CH 2 ClBr, CHClBr 2 , and CHCl 2 Br are included in the model set-up (Jöckel et al., 2016). Boundary conditions for CH 3 Br and the bromine-containing halons were taken from Meinshausen et al. (2011) and extended with the RCP6.0 scenario 165 as suggested by Eyring et al. (2013). We have used scenario 5 of Warwick et al. (2006) to describe the surface emissions of these organic bromine species. During the MIPAS measurement periods, from the model output first all data within one hour around 10 LT and 22 LT were selected. Depending on their latitude, longitude and altitude, they were then assigned to dayand night-time conditions and averaged over the observational bins of 10 • latitude and three-day periods.
For specific sensitivity investigations at low latitudes we have applied a 1d photochemical stacked box model. The chemical 170 mechanism of the 1d model is based on the SLIMCAT model (Sinnhuber et al., 2005, and references therein). The 1d model runs have been initialized with equatorial mean profiles of the EMAC simulation of all inorganic Br y species and NO 2 . For the other species as well as pressure and temperature, equatorial profiles from the MIPAS reference database have been used (Remedios et al., 2007). A comparison of the parameters of several bromine reactions between EMAC, the 1d baseline model The coverage at lower altitudes is determined by the lower limit of 15 km to confine the retrievals mainly on the stratospheric situation, by the cloud coverage and the scan pattern of MIPAS restricting the dataset mainly in the tropics. Some additional data gaps exist at high southern latitudes during winter when thick polar stratospheric clouds (PSCs) obscured the observations.  4. The lack of a similarly outstanding annual signal at tropical latitudes.
We have reduced the dataset to annual views by averaging over the whole MIPAS observational period in order to provide a clearer picture on intra-annual variabilities. The related horizontal and vertical cross-sections are presented in Figs. 5-8.

Measurement/model comparisons
The results of the EMAC model run are presented in the second column in each of the Figures 3-8. For these comparisons, we have applied the averaging kernel matrix of each retrieval to the related modelled profile. Differences between model results with and without averaging kernel application are generally below 1 pptv with only sporadic exceptions at highest altitudes near polar latitudes where differences reach up to 3-4 pptv (Figs. A1 and A2). The most obvious differences appear (1) at higher altitudes during polar winter, (2) in the lower stratosphere mainly at midand low latitudes during the entire year as well as (3) at altitudes around 30 km in the tropics during sunlit conditions. We 205 discuss these differences one by one below.

Low modelled polar winter BrONO 2
One disagreement between modelled BrONO 2 and MIPAS observations can best be observed in Figs

High modelled night-time BrONO 2 at lower altitudes
As visible e.g. in Figures 5 and 7, the model overestimates BrONO 2 volume mixing ratios with respect to the MIPAS results by up to 8 pptv in the lower stratosphere at altitudes up to about 25 km during night. Such differences are also present during day, albeit to a smaller extent (up to about 4 pptv, see Fig. 8).

220
The black line in panel "night" of the top row in Fig. 9 represents the measured mean night-time profile over the whole period at 5 • S. The related EMAC model result is shown in red. Maximum differences between both are about 8 pptv at around 20-22 km altitude. As possible explanation for these differences we will discuss below (1) measurement errors, (2) wrong modelling of the release of Br y from its source gases and, (3) wrong partitioning of Br y between its main constituents in the model simulations.

225
1. The degree of discrepancy between measurement and model of up to 8 pptv over an altitude range of around 5 km cannot reasonably be explained given the errors estimated for the MIPAS BrONO 2 retrieval (Fig. 1). This would allow for discrepancies of around 2-3 pptv, especially considering that the "noise" error term is strongly reduced by the temporal averaging over all equatorial measurements. Still one cannot rule out unequivocally any unidentified additional systematic error source in the measurements contributing to these differences. as by satellite observations (Sinnhuber et al., 2005;Sioris et al., 2006;Rozanov et al., 2011;Parrella et al., 2013). A possible contribution to a model overestimation might also be the used emission scenario of organic bromine species by Warwick et al. (2006). As has e.g. been shown by Keber et al. (2020), this scenario probably leads to an overestimation of brominated VSLS by up to 2 pptv, which is, however, not enough to explain the observed differences. Thus, it is highly unlikely that the inorganic bromine content at 20-25 km is strongly overestimated in the EMAC model calculations. As can be seen from Fig. 10, in test (a) one would require surface area densities enhanced by factors of more than 3 to deplete BrONO 2 to such an extent that the observations in the altitude range 20-26 km are met. For the idealized case of γ=1, the aerosol surface area densities (SAD) had to be adjusted up to 24 km only, however, still by a factor of more than 3 at 20 km altitude.

260
The EMAC model calculations fit quite well the observed night-time BrONO 2 maximum at 29-30 km altitude in the tropics (see Fig. 7). However, during daytime (Fig. 8), BrONO 2 is almost entirely depleted in the EMAC simulations at these altitudes while MIPAS still detects maximum values of up to 5 pptv. Possible reasons for this discrepancy discussed below might be (1) measurement errors, or (2) the impact of uncertainties in the parameters of reactions R1 and R2. Further we will address the effect of reaction R3 which is not considered in the present implementation of EMAC. 270 1. In the top row of Fig. 11, again the mean profile of BrONO 2 measured at 5 • S is again shown together with the EMAC model results in panel "day". As in the case of the night-time discrepancy, the values of BrONO 2 observed during sunlit conditions cannot be explained by the estimated retrieval errors which are about 1-2 pptv at those altitudes (Fig. 1).
However, as can be seen in Fig. 11, the mean measured profile values below about 24 km are negative indicating an unresolved issue in the retrieval at these altitudes. Still the negative values are in the range of the estimated uncertainties.

275
To investigate whether the positive values above, where the discrepancy to the model becomes apparent, might be due to an oscillatory feature caused by the negative values below, we have tested different retrieval options (increasing the regularization strength, performing retrievals of log(vmr) instead of vmr, skipping tangent altitudes below). Still in all tests, the maximum values at around 30 km appeared in the retrieved profiles, indicating them as a robust feature.
2. Sensitivity calculations using the 1d model are also shown in Fig. 11. As a baseline for these simulations we have applied 280 the setting "SAD adapted" so that the calculations during night-time fit to the observations also at altitudes below 27 km.
The resulting profile from the 1d model run also shows values which are smaller than the observed BrONO 2 mixing ratios by about 3 pptv at around 30 km altitude. Though being about 1 pptv larger than the EMAC simulations at those altitudes, the 1d results are also not compatible with the measurements.
To test the sensitivity on the production of BrONO 2 via the three-body reaction R1, we have used the JPL2019 formula- Further increasing these rate coefficients by a factor of two, which is well covered by the 2-σ uncertainty factor of 3.7 for that reaction at 220 K (Burkholder et al., 2019), lead to an additional increase by about 3 pptv (dash-dotted olive curves in Fig. 11). While these results coincide now with the observed daytime abundances of BrONO 2 at around 30 km, the 290 increase at lower altitudes does not correspond well with the observations there. Moreover, around 30 km during night the calculations overestimate the measurements by up to 2 pptv.
To test the sensitivity with respect to the photolysis of BrONO 2 (Eq. R2), we have divided the photolysis rate by 1.2, the 2-σ uncertainty as provided by Burkholder et al. (2019). The result is illustrated by the blue dashed curves in Fig. 11.
The resulting increase of only around 0.5 pptv is too small compared to the observations. Moreover, the increase appears The grey dotted curve in Fig. 11 illustrates the effect on the simulated mixing ratios when the loss reaction R3 is included in the 1d model by using the reaction coefficients from the JPL2019 compilation, see Tab. D1 (Burkholder et al., 2019).
With a reduction of up to 1.5 pptv this obviously drives the concentrations further away from the observations.

300
In conclusion, we have found no unequivocal explanation for the high measured daytime mixing ratios of BrONO 2 at around 30 km over the tropics/subtropics: (a) the differences between models and observations are outside the estimated measurement errors, (b) the uncertainties of the cross sections for photolysis (R2) of BrONO 2 are by far too small, (c) the error estimates for reaction R1 would allow a sufficient increase of BrONO 2 mixing ratios but over a too large vertical extend, and (d) any inclusion of reaction R3 opens the gap between simulated and observed BrONO 2 concentrations even more.  BrONO2 (top row), BrO (middle row) and HOBr (bottom row). Left: diurnal evolution of the "1d base" run for a period of two days. The white dashed lines indicate the local solar times of the MIPAS observations. The columns "night" and "day" contain the averaged night-time and daytime profiles of MIPAS at 5 • S as black solid lines along with error bars indicating the 2-σ estimated measurement uncertainty. The mean EMAC results are provided in red. The "MIPAS" and "EMAC" curves are the same as in Fig. 9. The other curves illustrate the results of the 1d model simulations and the columns "diff night" and "diff day" show the differences between the 1d sensitivity runs and the "1d base" run. The MIPAS dataset of BrONO 2 allows us to determine the total stratospheric content of bromine (Br y ) by using the socalled inorganic method. It is a linear correction of the observed bromine species by multiplication with the ratio between total modelled bromine (Br mod y ) and the modelled bromine species observed. This procedure has often been applied in case of daytime observations of BrO (e.g. Dorf et al., 2006a, b). Wetzel et al. (2017) was the first to apply this method to nightime (1) Using night-time observations of BrONO 2 instead of BrO measured during day to derive Br y should have the advantage that BrONO 2 during night makes up a larger fraction of Br y than BrO does during day, due to the continuous production of BrONO 2 (Eq. R1).

315
The modelled night-time ratio between BrONO mod 2 and Br mod y has been examined to decide which region (in terms of altitude and latitude) to use for determination of Br y from the MIPAS dataset (cf. Fig. C1). Values clearly exceeding 90% are simulated in mid-latitudes mainly during spring/summer/autumn and centred at altitudes around 26 km. Thus, for the analysis we have chosen data at 25-26 km altitude and 40-60 • latitude from October to March in the southern and from April to September in the northern hemisphere. Additionally, due to the low seasonal variability of BrONO 2 and to capture 320 relatively young air masses, we considered also tropical data at 29-30 km in the analysis. While not exceeding 90%, the ratio , blue) is shown in Fig. 12. The solid lines represent weighted means of all data points, where for the weighting the a-posteriori estimated random error (dotted black curves in Fig. 1) was applied. The resulting random error as indicated by the thickness of the line is very small (0.04-0.07 pptv) due to the large 330 amount of data points. The major part of the uncertainty in our estimation of total Br y , however, is due to the combined parameter and systematic error (the green line in Fig. 1) which is shown by the red shading in Fig. 12. It should be noted that this error term is a combination of a variety of estimated uncertainties each of which might also partly be random in nature with different temporal correlation lengths (e.g. correlated over each of the two measurement phases). Thus, we have made a conservative assumption in considering all those as systematic -but possibly also underestimating them by applying quadratic  Table 1). The values of Br y from the mid-latitudes of both hemispheres coincide clearly within their uncertainty ranges. Since it is unlikely that the real values of total Br y vary strongly in the stratosphere, the difference of 1-340 1.4 pptv between tropical and mid-latitudinal estimates are more probably caused by uncertainties. These may either be caused by errors in the BrONO 2 concentrations derived from MIPAS or be due to the calculation of total Br y from BrONO 2 through of Eq. 1 (or a combination of both). The first explanation would require a retrieval error component varying with latitude e.g.
due to some temperature dependence while the second one implies model uncertainty. Since the model adjustment of Br y from BrONO 2 is much larger in the tropical stratosphere (about 2.5 pptv) than at mid-latitudes (about 0.5 pptv), the second 345 explanation would affect more strongly our estimation of Br y in the tropics.
Br y obtained from MIPAS can be compared to data of Br y derived from observations of BrO, as summarized for example in   We have presented the first global dataset of BrONO 2 volume mixing ratio profiles for day and night derived from 10 • -zonally concern about its validity.
While we cannot rule out for sure that unaccounted systematic errors in the observations are responsible for these discrepancies, this seems rather unlikely given their overall fit to the model as well as the error assessment. This view is supported by the estimation of the total stratospheric bromine content from MIPAS BrONO 2 measurements for years of stratospheric entry between 1997 and 2007, i.e. around the maximum of stratospheric total bromine content (Engel et al., 2018). At mid-latitudes, 385 where the model correction to estimate Br y from observed BrONO 2 volume mixing ratios is smallest, we derived an average value of 21.2±1.4 pptv of total stratospheric Br y which fits very well to independent estimates based on observations of BrO and to Br y estimates derived from BrONO 2 observations of the MIPAS balloon experiment.
In case the inconsistencies between model and observations as discussed above are due to model uncertainties, they could also affect estimated ozone loss processes through bromine cycles. In future, our dataset of BrONO 2 from MIPAS can be 390 combined e.g. with the simultaneous daytime BrO observations from the SCIAMACHY instrument on Envisat to investigate the revealed issues about possible deficiencies in our understanding of stratospheric bromine chemistry as well as to gain more insight in possible uncertainties in the observations.