Long-term Time-series of Arctic Tropospheric BrO derived from UV-VIS Satellite Remote Sensing and its Relation to First Year Sea Ice

Arctic Amplification describes the rapid increase of the air temperature in the past three decades in the Arctic, which impacts on physicochemical conditions, the ecosystem and biogeochemistry. Every polar spring, the BrO explosion, a series of chemical reactions that release bromine molecules to the troposphere occurs over sea ice covered regions. This autocatalytic mechanism depletes boundary layer and tropospheric ozone, thereby changes the oxidizing capacity of the atmosphere and facilitates the deposition of metals (e.g. Hg). In this study, we present a 22 year consolidated and consistent 15 tropospheric BrO dataset, derived from four different UV-VIS satellite instruments and investigate the BrO evolution under the impact of Arctic Amplification. The retrieval data products from the different sensors are compared during periods of overlap and show good agreement. By studying the sensor merged time-series of tropospheric BrO vertical column densities, we find an increase in the magnitude of BrO explosion events under the impact of Arctic Amplification with an upward trend of about 1.5% per year. Furthermore, the areas where BrO plumes frequently appear have changed, extending over larger 20 regions in the Arctic during more recent years. Comparison to sea ice age data suggests that the reported changes in tropospheric BrO are linked in a complex way to the increase of first-year ice extent in the Arctic.

sea ice has declined dramatically in recent years as the portion of thick multi-year ice decreases (Richter-Menge et al., 2017). Sea ice is replaced by open ocean, which being darker, reduces surface reflectivity in the Arctic. As a result, more of the incoming solar radiation is absorbed by the ocean and its biosphere (e.g. phytoplankton). Consequently, temperature of the ocean and the air in the boundary layer increases, creating a positive feedback loop. This is one of the most pronounced effects associated with Arctic Amplification (Hansen et al., 1997;Kirtman et al., 2013). These evolving conditions impact on 5 many other physical, chemical and biological processes, as well as the eco system in the Arctic.
Bromine monoxide (BrO) plays a significant role in the atmospheric chemistry of the Arctic. During polar springtime, episodes of strongly enhanced amounts of BrO have been observed in the boundary layer . The formation of these intense plumes of BrO results in tropospheric ozone depletion. Tropospheric ozone depletion and the link to halogen chemistry events was first discovered 30 years ago (Barrie et al., 1988) and has been the subject of many studies 10 and campaigns over the past three decades (Toohey et al., 1990;Tuckermann et al., 1997). The release of reactive halogens and the decrease in O 3 impacts on the oxidizing capacity of the troposphere. The photolysis of O 3 in the UV-B leads to the formation of the most important tropospheric oxidising agent, the hydroxyl radical, OH. While in plumes of tropospheric BrO, the oxidising agents O 3 and OH are reduced, the reactions of BrO play also other important roles in atmospheric chemistry. For example, BrO efficiently reacts with elemental mercury. This oxidation initiates a process whereby deposition 15 of mercury to snow and ice increases. This results in Hg entering the food chain (Lu et al., 2001;Schroeder et al., 1998). The rapid and sudden appearance of BrO plumes over the Polar Regions has been called the "bromine explosion" Platt and Lehrer, 1997). It is explained by an autocatalytic multiphase chemical cycle, which occurs on cold saline surfaces (Fan and Jacob, 1992;Sander and Crutzen, 1996). The triggering mechanism for the release and the production of gas phase plumes of BrO remains a matter of scientific controversy (Jones et al., 2009). However, there is the 20 general consensus that the potential sources are (a) rich in sea salts and relatively cold (conditions occurring in potential frost flowers regions; Rankin et al., 2002;Kaleschke et al., 2004;Sander et al., 2006), (b) surfaces covered with liquid or frozen brine (Sander et al., 2006), (c) associated with blowing snow (Yang et al., 2008;Blechschmidt et al., 2016;Frey et al., 2019), (d) surface snow packs (Pratt et al., 2013;Peterson et al., 2018) and young sea ice regions (Wagner et al., 2001;Simpson et al., 2007;Peterson et al., 2016). A pH lower than 6.5 is required for efficient bromine activation (Fickert et al., 1999;25 Halfacre et al., 2019).
Organic sources of BrO, such as oceanic bromocarbons, have also been discussed in the literature (Salawitch et al., 2006, and references therein) but the relatively long lifetimes result in a slow release of Br throughout the Arctic troposphere and are currently not considered to explain observations of the bromine explosions. Starting with molecular bromine in the gas phase, the sequence of autocatalytic chain reactions describing the bromine explosion reactions can be written in its simplest 30 form as: Br 2 + hv (350 nm < λ < 500 nm) → 2Br (R1) https://doi.org /10.5194/acp-2020-116 Preprint. In short, the autocatalytic multiphase chain reaction releases molecular bromine, Br 2 , to be photolysed (R1). The resulting bromine atoms remove rapidly tropospheric O 3 (R2). The resultant BrO reacts with HO 2 to form HOBr (R3). This enters the aqueous phase or quasi liquid layers on cold brine or snow and ice, where it reacts with halogen ions to release Br 2 to the atmosphere (R5). The efficiency of such a chain reaction depends on the chain length in the atmosphere. This depends on the relative rate of chain propagation and chain termination reactions of the chain carriers i.e. the Br and BrO. The bromine 10 explosion slows through the depletion of O 3 in the air mass, or through reactions of Br or BrO with formaldehyde or NO 2 : Br + HCHO → HBr + HCO (R6) BrO + NO 2 + M → BrONO 2 + M (R7) BrONO 2(g) = BrONO 2(aq) There are also cycles involving chlorine ions, also initiating a catalytic loss. The involved reactions are explained in more 15 detail elsewhere (e.g. Simpson et al., 2007;Sander et al., 2006).
It was shown that BrO plumes can be transported far from their initial formation areas, due to release from snow packs (Peterson et al., 2018) and blowing snow (Giordano et al., 2018) favored by high wind speeds associated with cyclones (Begoin et al., 2010;Zhao et al., 2015;Blechschmidt et al., 2016. Peterson et al. (2017 studied the vertical transport of BrO and it's recycling on aerosol particles, stating that BrO can be sustained, reformed and transferred on aerosols. Using a 20 qualitative model, Jones et al. (2009) have shown, that both stable boundary layer, very low near surface wind speed conditions, as well as unstable boundary layer, high wind speed conditions increase the number of reactants in the air and hence favor the bromine explosion.
The Polar Regions are some of the most remote and hostile places on the planet. Consequently, satellite remote sensing is a unique method to study bromine chemistry in the Arctic. Richter et al. (1998) and Wagner and Platt (1998) were the first 25 studies on satellite observations of BrO plumes in Polar regions, using the GOME instrument (Richter et al., 1998;Wagner and Platt, 1998). Based on observations from the same instrument, Hollwedel et al. (2003) derived a six year time-series of Arctic and Antarctic total vertical column densities (VCDs) of BrO. This was the first scientific effort to study the evolution of BrO in the Polar regions. The transport of BrO plumes, which represents a photo-stationary state with production and loss processes and their capability of depleting ozone, far away from the initial release area was studied with satellite remote 30 sensing (Ridley et al., 2007;Begoin et al., 2010). The relationship between BrO release and young sea ice was also discussed (Wagner et al., 2001). Van Roozendael et al. (2004) and Jacobi et al. (2006), used SCIAMACHY observations for Arctic BrO and compared them to GOME data, Theys et al. (2011) andSihler et al. (2012) used GOME-2 for the same purpose.
https://doi.org /10.5194/acp-2020-116 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License. Seo et al. (2019a) presented the first BrO retrievals from the TROPOMI instrument. Studies have also used satellite remote sensing, to link them to their sources and triggering meteorological conditions, in order to better understand this complex and significant phenomenon. For instance, Choi et al. (2018), Begoin et al. (2010), Toyota et al. (2011) andJones et al. (2009) investigated links between tropospheric BrO and blowing snow. Blechschmidt et al. (2016) investigated a BrO explosion event using GOME-2A and associated its long lifetime with continuous release of bromine molecules from blowing snow 5 along the front of a polar cyclone.
Major changes in meteorological parameters, e.g. increasing air temperature (Serreze and Barry, 2011), decreasing mean sea level pressure over northeastern America and increasing pressure over Eurasia (Ogawa et al., 2018;McCusker et al., 2016), increase in cyclone frequency and intensity (Akperov et al., 2019), stronger surface winds (Mioduszewski et al., 2018) and changes in sea ice conditions (e.g. reduced sea ice extent; Stroeve et al., 2012), increased first year sea ice fraction and 10 therefore decreased sea ice thickness (Richter-Menge et al., 2017)) occur due to Arctic Amplification. It is therefore likely that the intensity, frequency and spatial distribution of bromine explosions in the Arctic are changing. Consequently, the objectives of this study are a) to derive the first consolidated and consistent long-term Arctic tropospheric BrO dataset using satellite remote sensing and b) to investigate and understand changes or trends in BrO and the relation to sea ice based on the new dataset. As described above, an earlier effort was carried out by Hollwedel et al. (2003), but they did not extract the 15 tropospheric BrO column from their 6 year GOME dataset. Moreover, Choi et al., (2018) used an 11 year Arctic tropospheric BrO dataset from the OMI instrument,, and investigated the link to young sea ice and sea salt aerosols released from blowing snow. However, the OMI sensor suffers from instrumental degradation, known as row anomaly (Class et al., 2010). In our study, we use measurements from four ultraviolet -visible satellite instruments to derive tropospheric VCDs of BrO over the Arctic, covering a time-span of 22 years, starting with GOME observations in 1996 and ending with GOME-20 2A and GOME-2B in 2017. We first retrieve BrO slant column densities (SCDs) for each instrument and then separate tropospheric VCDs from stratospheric BrO amounts. This paper is structured as follows: in section 2, a short technical description of the instruments used in this study is presented. In section 3, the methods applied for retrieving geometric and tropospheric BrO columns from the satellite sensors and the additional datasets used for stratospheric separation and sea ice flagging of tropospheric BrO are introduced. In 25 section 4, an analysis results of the long-term data is presented, including time-series and maps of tropospheric BrO VCDs and comparisons with sea ice age and a statistical analysis of possible trends of tropospheric BrO VCDs. The paper ends with a summary and conclusions (section 5).

Instruments
The technical attributes of the satellite sensors that are used, together with information about instrumental degradation are 30 described in the following. A short summary of characteristics of the sensors is given in Table 1.

GOME
The GOME (Global Ozone Monitoring Experiment)  instrument was launched in 1995 on ERS-2. It was a nadir viewing scanning spectrometer, observing solar back-scattered radiation upwelling from the Earth's surface and atmosphere. It measured continuously in four spectral channels from 240 to 790 nm, with a spectral resolution between 0.2 and 0.4 nm. For 27 days per month, the spatial resolution of the forward scans was 40x320 km 2 , and the swath 960 km. For 5 approximately 3 days per month, the instrument operated in narrow swath mode with a swath width of 240 km and ground scenes having a spatial resolution of 40x80 km 2 . ERS-2 was in a sun synchronous orbit with a 10:30 local time equator overpass in the descending node. GOME was the first satellite instrument, whose measurements of the ultraviolet and visible upwelling radiance were appropriate for the retrieval of many key trace gases . The instrument was launched in July 1995 and lost its global coverage in 2003, due to data rate limitation (the onboard storage capability of the 10 instrument was disabled and as a result data could only be transmitted directly to ground stations) (Bracher et al., 2005).

SCIAMACHY
SCIAMACHY (SCanning Imaging Absorption SpectroMeter for Atmospheric CHartographY) (Bovensmann et al., 1999) was a satellite spectrometer onboard Envisat. It was launched into space in 2002. The main advantage of SCIAMACHY compared to GOME was its broader spectral coverage, ranging from 210 nm to 2380 nm, allowing the observation of many 15 trace gases in the near infrared and short wave infrared spectral regions. SCIAMACHY's spectral resolution was 0.2 nm to 0.5 nm, and its spatial resolution in the spectral region used for BrO retrievals was 30x60 km 2 , with a 960 km swath width in nadir. The overpass of Envisat over the equator was at 10:00 local time. SCIAMACHY observed the Earth in nadir, limb and occultation geometries, providing a wealth of trace gas data on the atmosphere from the surface to the upper atmosphere (Bovensmann et al., 1999). In April 2012, Envisat lost contact with the ground station and as a result, the mission had to be 20 terminated.

GOME-2
The series of GOME-2 (Global Ozone Monitoring Experiment-2) (Callies et al., 2000) instruments were developed as the successors of GOME. There are currently three instruments in orbit, one launched onboard Metop-A in 2006, one on Metop-B in 2012, and one on Metop-C in 2018. Here, we use data from GOME-2A and GOME-2B. All GOME-2 instruments share 25 the same attributes and sense the Earth's backscattered radiance and extraterrestrial solar irradiance in the ultraviolet and visible part of the spectrum (240 nm to 790 nm). They have a spectral resolution between 0.2 nm and 0.4 nm, while the footprint size is 80x40 km 2 and a much wider swath (1920 km) than the previous instruments. GOME-2A changed its swath to 960 km and footprint to 40x40 km 2 in June 2013. The GOME-2 instruments are crossing the equator at 09:30 local time (Callies et al., 2000). All three instruments are currently in operation. 30 https://doi.org /10.5194/acp-2020-116 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License.

Instrumental Degradation
Many space borne optical instruments suffer from a decrease in throughput in the ultraviolet spectral region, which arises from the deposition of absorbing layers on the optical surfaces such as mirror lenses or gratings. This results in a variety of effects such as loss of throughput and changing etalons in the instrument. The quality of the retrieved BrO data, which is 5 produced from the weak absorption signal in the UV region, is influenced by these degradations. The GOME, SCIAMACHY and GOME-2 teams identify any degradation in the reflectances and correct where appropriate and possible. For example organic compounds and water, emitted by the spacecraft, are photochemically transformed by UV-B and vacuum UV from the sun and most likely form long changing polymers, which have low vapour pressure and are absorbed on the mirrors. In this context, Snel (2000), showed that the GOME sensor experienced degradation in all wavelength regions but in particular 10 in the UV. Krijger et al. (2007), compared the degradation of GOME to SCIAMACHY with respect to the reflectivity. Dikty et al. (2011), investigated the impact of GOME-2A throughput loss on various time-series of trace gases. Most of the major degradation effects on the sensors are identified and documented in literature and where possible accounted for Garcia et al., 2016).

Methods 15
In this section, the methodology to retrieve BrO SCDS from the satellite measurements and to obtain tropospheric BrO VCDs and datasets used in this study are described.

Retrieval of BrO slant column densities
Similar to previous studies on atmospheric BrO from hyperspectral satellite remote sensing observations in the solar spectral regions, the retrieval algorithm is based on differential optical absorption spectroscopy (DOAS) (Platt and Perner, 1983;20 Burrows et al., 2011). DOAS is an application of the Beer and Lambert law, which describes the attenuation of electromagnetic radiation in a medium: https://doi.org /10.5194/acp-2020-116 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License.

Eq. (1),
where I is the measured intensity of the electromagnetic radiation, I o is the initial intensity, J is the total number of trace gases absorbing, j denotes a particular trace gas (e.g. BrO), σ(λ) is the cross section of the absorber at wavelength λ, ρ the density of the trace gas and s the length of the light path.
The main idea of the DOAS method is, that the atmospheric amount of the trace gases of interest can be retrieved using their 5 characteristic spectral fingerprints. This is done by separating the extinction signal into a low frequency and a high frequency part. The low frequency part is treated as a closure term and is fitted by a low order polynomial. The higher frequency term contains the absorption structures of the trace gases. The final output of the retrieval is the slant column density of the trace gas, i.e. the density of the trace gas, integrated along the light path (Platt and Stutz, 2008): For cloud free conditions, the radiance upwelling at the top of the atmosphere is described at a given wavelength by: where c is the scattering efficiency, SCD j the slant column density of the gas with index j, σRay(λ) and σΜie(λ) are the absorption cross sections of Rayleigh scatterers (e.g. primarily air molecules, molecular nitrogen N 2 and oxygen, O 2 ) and Mie scatterers ( To retrieve accurate BrO SCDs, an optimal selection of the spectral window, which maximises the information content with respect to the BrO absorption, and the selection of corresponding cross sections of other trace gases absorbing in the 20 same spectral window is the first step. We chose to use temperature dependent cross sections of ozone (dominant absorber in the UV) by Serdyuchenko et al. (2014) at 223 and 243 Kelvin and a BrO cross section (Fleischmann et al., 2004)

at 223
Kelvin. In addition, a pseudo cross section is used for simulating the filling-in of Fraunhofer lines by Raman scattering known as the Ring effect (Vountas et al., 1998), and another pseudo cross section, which deals with the issue of poor spectral sampling (Chance et al., 2005), was added for all instruments. The high spectral resolution absorption cross sections were 25 convolved by the slit function of each instrument. The reason for using these cross sections is that our study is dedicated to BrO released in the Arctic region. Tests were made for different combinations of trace gases. It was found that the omission of the explicit fitting of NO 2 and SO 2 has a minimal impact on the fit of the BrO SCD: the differences between with and without NO 2 or SO 2 in the selected spectral ranges were typically less than 1% of the BrO SCD. Therefore, only cross sections of the dominant absorber O 3 in this region, and of the trace gas of interest itself, BrO were added. Also, a four 30 degree polynomial was used, as the differences between fourth and fifth order polynomials were small on the BrO SCDs, while the quality of the fit improved greatly from a third to fourth degree polynomial.
As BrO is a relatively weak absorber, small changes in the input parameters, and especially of the fitting window of the retrieval, can lead to large changes in the quality of the fit. Although there are many good practices developed for the DOAS fitting window selection for an absorber (e.g. it must include at least two absorption peaks from the trace gas of interest, no large Fraunhofer lines, small interference from other species etc.), there is no precise and agreed methodology to determine the optimal selection. As a result, different spectral fitting windows for BrO retrievals have been used in previous studies. 5 For example, Richter et al., (1998), used a 345 -359nm wavelength region for GOME, Afe et al., (2004)  The annual cycle in the fitting RMS for the Arctic results from changes in the sun's position and the impact on the upwelling radiance. In spring and autumn the solar zenith angle is larger, the scattering and attenuation increases, and as a result, the 5 radiance signal is low. In the Arctic, SCIAMACHY has the lowest fitting RMS of all instruments. GOME-2A shows a rapid increase in RMS until 2009, and a smaller upward trend in the following years. A similar systematic increase in RMS occurs for GOME-2B. The RMS for the Pacific region shows a similar behaviour for GOME-2A and GOME-2B, as both increase strongly with time. GOME appears to have lower RMS values on average than the GOME-2A and GOME-2B instruments, presumably because of the lower spatial resolution, but the RMS shows large variability. Generally, daily mean RMS values 10 are below 2.0x10 -3 , for all instruments. There is no clear trend in the SCD of BrO over the Pacific region for any of the instruments.
The use of a reference area over the Pacific as background spectrum is an alternative to the use of solar irradiance measurements, which removes systematic errors arising from interfering instrumental structures in the solar irradiance measurements. The region of the Pacific is selected because of its relatively small annual cycle of BrO (Richter et al., 2002). 15 In this way, the quality of the fit improves greatly as problems in the radiances mainly cancel out. This also has other https://doi.org /10.5194/acp-2020-116 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License.
benefits. For example, GOME-2A is currently drifting in orbit and there are periods during which solar measurements can no longer be carried out. Consequently, the Pacific background spectrum has been used instead of the Sun background spectrum for all instruments in the present study. Although trends over the clean Pacific background region resulting from instrumental degradation were minimised, residual trends had to be accounted for. This has been achieved by applying an additional Pacific correction to each instrument separately. The normalization method computes the average BrO SCD in a 5 small Pacific area (0.0 o ± 10.0 o latitude, 180.0 o ± 20.0 o longitude) and then subtracts this average from every pixel of the BrO SCD. To compensate the negative bias imposed by the method, a constant offset of 7x10 13 molec/cm 2 was added for every day (Richter et al, 2002;Sihler et al., 2012). This correction tackles offset errors that are occurring for weak absorbers, and are due to instrumental degradation (Alvarado et al., 2014).
The empirically determined set of parameters used in the fitting of BrO in the spectral windows are reported in Tables 2 and  10 3.

Tropospheric BrO vertical column densities and datasets
NO 2 and O 3 column satellite retrievals, tropopause height from meteorological reanalysis data are used for extracting the tropospheric BrO component from the retrieval (stratospheric separation). Sea ice data (age and type) obtained by satellite remote sensing was used in order to identify regions with sea ice cover and hence high surface reflectivity required for the retrieval of tropospheric BrO in this study and for data interpretation in relation to bromine sources. 5

Stratospheric Separation
In order to derive the tropospheric BrO VCD from the retrieved SCD of BrO, the method of Theys et al. (2009) was used.
Briefly, this method is based on a stratospheric BrO climatology derived with the BASCOE model (Errera and Fonteyn, 2001) and requires year, latitude, tropopause height, O 3 and NO 2 columns as input. This approach has been applied successfully in previous studies (e.g. Begoin et al., 2010;Theys et al., 2011;Blechschmidt et al., 2016;Choi et al., 2018). In 10 the present study, satellite derived O 3 VCDs from Weber et al. (2013), stratospheric NO 2 VCDs from the QA4ECV project (Boersma et al., 2017), and from the Tropospheric Emission Monitoring Internet Service (TEMIS), (Boersma et al., 2004) and tropopause heights derived from NCEP reanalysis (Kalnay et al., 1996) data were used as input. In Theys et al. (2009), a correction factor was applied, to account for the long-term reduction of bromine emissions in the stratosphere. This factor was based on ground based zenith-sky measurements of BrO over Harestua (Hendrick et al., 2008). In the present study, this  The following formula (Theys et al., 2009) is used to derive the tropospheric VCD of BrO: where SCD total is the slant column of BrO retrieved by the DOAS method, VCD strato corresponds to the stratospheric BrO VCD derived from the Theys et al., (2009) climatology, AMF strato is a stratospheric air mass factor, AMF tropo is a tropospheric 10 air mass factor. For the latter, as in Begoin et al. (2010) and Blechschmidt et al. (2016), a surface albedo of 0.9 has been assumed above sea ice and that all tropospheric BrO is well mixed within the boundary layer extending to 400m altitude.
Note that the stratospheric BrO VCD column is independent of the BrO SCD derived from the DOAS retrieval described in

Sea Ice
In order to study the connection between tropospheric BrO and sea ice under the impact of Arctic Amplification, long term sea ice data (starting from 1996) is required in this study. In addition, since the tropospheric AMF applied for the retrieval of tropospheric BrO (see previous section) assumes a surface albedo of 0.9, the sea ice data was used to remove data with no sea ice cover from the BrO data. For this purpose, the sea ice age dataset from Tschudi et al. (2019) was used. It is retrieved 5 from different passive microwave satellite remote sensing instruments and has a very high spatial resolution of 12.5x12.5 km 2 , while its temporal resolution is 7 days.

Results
In this study, the focus is the long term analysis of BrO columns during the period from March until September. This period has the best coverage by the UV-VIS satellite remote passive sensors at high latitudes. The months of October to February 10 have few or no observations due to polar night in the target latitudes. The analysis is restricted to data northwards of 70.0 0 referred to as the Arctic here. This region was chosen because the source regions for BrO explosion events are known to be associated with regions of high sea ice cover. We present the derived BrO time-series for the Arctic region, together with corresponding annual cycles, scatter plots and map plots in section 4.1. The tropospheric BrO is compared to sea ice coverage and age in section 4.2, to investigate the impact of changing sea ice conditions due to Arctic Amplification on BrO 15 amounts in the Arctic's troposphere. Finally, a trend analysis of tropospheric BrO is performed, together with a check for statistical significance of the trends (section 4.3).

Comparison between the BrO columns derived from different sensors
The geometric BrO VCD is derived by dividing the BrO SCD with a simple stratospheric AMF, which takes into account the scattering at the surface but ignores the impact of scattering within the atmosphere. Consequently, it necessarily differs from 20 the sum of the tropospheric and stratospheric column. Daily values of the geometric, stratospheric and tropospheric VCDs of BrO over the Arctic region are shown in Figure 3.

Figure 3: Long-term BrO time-series over the Arctic region: a) daily geometric BrO VCDs [10 13 molec/cm 2 ], b) daily stratospheric BrO VCDs [10 13 molec/cm 2 ] and c) daily tropospheric BrO VCDs [10 13 molec/cm 2 ]. All figures show daily averages ≥ 70.0 o N latitude. GOME data is coloured in red, SCIAMACHY in blue, GOME-2A in green and GOME-2B in brown.
The geometric VCD accounts for the differences from the different viewing geometries of the satellite sensors to a first 5 order. Qualitatively, there is good agreement during periods of overlapping measurements e.g. between GOME and  The comparison of tropospheric BrO VCDs time-series during the periods of overlap between the different instruments show a similar level of agreement to that for the geometric VCDs. The seasonality of the tropospheric time-series is also similar to the geometric one. We attribute this to the inorganic release of bromine associated with sources, which depend on sea ice and meteorological parameters. As described in the introduction, in polar spring, the combination of low temperatures on first year sea ice, presumably having sufficient brine, triggers the release of Br in the atmosphere. This occurs dominantly in 15 spring. During polar summers and early autumn, the sea ice extent reaches the yearly minimum and the temperature its   The best agreement is found between SCIAMACHY and GOME-2A, both for the geometric and the tropospheric VCDs, while the least good agreement is found between GOME and SCIAMACHY (this can be attributed to the comparably short overlapping period between these sensors).
Climatological seasonal cycles (averages over the whole period of retrievals of geometric, stratospheric and tropospheric 10 VCDs) for each instrument are shown in Figure 5.
https://doi.org /10.5194/acp-2020-116 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License. The annual cycles of the geometric and the tropospheric VCDs time-series are similar in shape. The largest VCDs occur for 5 all instruments in polar spring (March to May), followed by a strong decrease from May to June. In spring, GOME-2B VCDs are slightly larger, while GOME columns are slightly lower than data from the other instruments for the geometric VCDs. In summer and early autumn GOME VCDs are higher. For the tropospheric VCDs, the differences between the instruments become smaller.   VCDs. The transport of plumes of BrO by cyclones away from the initial release area (e.g. the 2007 case shown in Figure 5,30 investigated in more detail by Begoin et al., 2010) is also observed by the sensors. Hence, the track of cyclones transporting BrO plumes is expected to influence longer-term averaged BrO maps.

Arctic Tropospheric BrO and its Relation to Sea Ice
In this section, the evolution of tropospheric BrO and its relation to sea ice is presented. More specifically, the relationship between first year ice and tropospheric BrO is investigated. Since an albedo of 0.9 was assumed in the tropospheric AMF, a sea ice based ground scene flagging of our tropospheric BrO dataset was performed. In this way, only BrO observations above sea ice are analysed. As described in section 4.1, the individual BrO VCDs time-series from the four sensors are 5 highly consistent and the remaining inconsistencies are small. Hence, a merged tropospheric BrO dataset over sea ice was derived, by simply averaging the overlapping days between the sensors. The merged tropospheric VCDs BrO dataset is shown in Figure 8   In order to investigate the spatial evolution of tropospheric BrO in time and its relationship to sea ice age, polar spring (March to May) average maps of tropospheric BrO and corresponding sea ice age maps are displayed in Figure 9 for each year between 1996 and 2017.   Over the last years (2011 onwards), positive differences appear to the north and to the east of Greenland. Overall, and when comparing the last with the first five years, we see that the increase of BrO plumes is apparent over the last years of Arctic Amplification. 5 Figure 11 shows plots which probe the relationship between tropospheric BrO and first year ice extent. More specifically, Figure 11a shows polar spring tropospheric BrO over sea ice and first year ice extent, Figure11b shows the area extent where tropospheric BrO exceeds the threshold of 7x10 13 molecules per cm 2 . This threshold value was chosen empirically, as in previous studies (Hollwedel et al., 2003;Choi et al., 2018). First year ice extent, and corresponding scatter plots are also   The total Arctic sea ice extent has decreased over the last years, but the area covered by first year sea ice has increased. The type of ice encountered is hence changing. Due to the increasing temperatures in the Arctic, more sea ice melts every year, 10 and more fresh or young sea ice is formed each winter period. Figure 11a shows that 2008 was the first year that first year ice extent exceeded the threshold of 5x10 6 km 2 , which is also the case for almost all of the following years. During the decade from 2007 to 2017, first year ice extent was larger during polar springs than from 1996 to 2006. Although tropospheric BrO VCDs seem to have also increased in magnitude approximately from 2007 onwards, the yearly evolution between the two does not correlate strongly. For example, for years 2008 and 2016 a decrease of tropospheric BrO VCDs 15 compared to the previous year is found, while first year ice extent increased. In 2015, the highest tropospheric BrO VCDs polar spring averages are found. In the same year, the polar spring first year sea ice extent average has its lowest value after 2008. Similar conclusions can be drawn from the top right scatter plot (Figure 11c): Although most days of average BrO above 5x10 13 molec/cm 2 are occurring with first year ice extent above 4.5x10 6 km 2 , we can see such days also below this threshold. The correlation coefficient between polar spring tropospheric BrO VCDs and first year sea ice extent for the same 20 period is 0.32. The correlation between areas of enhanced tropospheric BrO VCDs and first year ice extent is even lower, https://doi.org /10.5194/acp-2020-116 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License. approximately 0.2. This is evident from both the time-series (Figure 11b) and their scatter plot (Figure 11d, bottom right). It seems that in both cases, the largest number of days with high tropospheric BrO VCDs or largest areas of BrO explosion events are occurring when first year ice extent is between 4.5 and 5x10 6 km 2 , and not when it reaches its peak. The magnitude of tropospheric BrO VDs is not closely related to first year ice extent. This agrees with the findings of Choi et al, (2018), where a similar analysis is provided based on OMI data. They found a correlation coefficient of -0.32 between first 5 year ice extent and tropospheric BrO explosion frequency. The negative sign may be attributed to the decrease of their tropospheric BrO VCDs over the latest years, which arguably is linked to degradation of the OMI instrument.

Trend analysis
Finally, a trend analysis of the tropospheric BrO VCDs time-series is performed, in order to investigate the statistical significance of the observed temporal and spatial changes in tropospheric BrO VCDs described above in relation to Arctic 10 Amplification. The tropospheric BrO VCDs time-series shows a seasonality, with a maximum every polar spring (March in most cases) and a minimum every summer. Consequently, a model which combines a linear trend and seasonal components is selected, as in other studies (Hendrick et al., 2008;Georgoulias et al., 2018), to calculate the trends appearing in our timeseries, expressed by the following formula: Eq. (7)  15 where A is the slope, B the intercept, d(t) is the modelled value of a given day, t is the day of the dataset (expressed in fractional years) and M is the time period. It should also be stated, that the number of harmonic functions was chosen based on the minimization of the residuals between the model and the dataset. Furthermore, the error of each trend was calculated, based on the formula: Eq. (8)  20 where, σ M is the standard deviation of the residuals between the model and the time-series, M is the period in years, and φ is the autocorrelation of the residuals. Finally, a trend is considered significant if the ratio between it and its error is greater than 2 (Weatherhead et al., 1998). Table 4 shows trends of tropospheric BrO over sea ice over the whole period (see Figure 8c), for individual months and for BrO explosion extent (Figure 11b). 25   In Figure 13, the trend for each individual grid box during polar spring months is displayed, for both tropospheric BrO and sea ice age. Both maps are gridded in 0.125 o x0.125 o resolution. The trend for every grid box is calculated based on the 10 method described above, i.e. in Equation 7.
https://doi.org /10.5194/acp-2020-116 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License. Figure 13, tropospheric BrO VCDs values increase inside the region of the Arctic Ocean, and also to the north and east of Greenland, where sea ice age shows a negative trend. From Figure 9, we see that from 2008 and afterwards, first

As shown in
year ice appears in the same area. However, the trends of tropospheric BrO and sea ice age do not match everywhere. The strongest negative trends in the sea ice map occur in areas where multiyear ice was dominant in the past and first year ice is forming in the latest years (Beaufort Sea and Canadian archipelago). However, in the same area, we do not see the most 5 pronounced increase in the trend map of tropospheric BrO VCDs (the increase is smaller than in other areas). The strongest increase in tropospheric BrO VCDs occurs to the northeast of Greenland close to the Fram Strait, to the north of Greenland and close to Ellesmere Island. However, sea ice age does not show a strong trend to the northeast of Greenland.

Summary and conclusions
Arctic Amplification, the rapid and intense increase of air temperature over high latitudes is expected to have drastic impacts 10 on all constituents of the Arctic ecosystem. Bromine release from young sea ice, frost flowers, blowing snow and liquid brine was first discovered about 30 years ago. This is of great significance in the Arctic atmosphere, as, through a set of autocatalytic reactions, bromine depletes tropospheric ozone very efficiently, potentially altering the oxidizing capacity of the troposphere. In this study, the first consistent and long-term dataset of tropospheric BrO over the Arctic, retrieved from four UV-VIS instruments over a 22-year time period is presented. The results demonstrate a high level of consistency of the 15 retrieved BrO VCDs between the different sensors.
Time-series of tropospheric BrO over the Arctic and corresponding trends indicate that the magnitude of tropospheric BrO explosion events has increased over the last years. This increase is pronounced in polar springs, when BrO explosion events https://doi.org /10.5194/acp-2020-116 Preprint. Discussion started: 16 March 2020 c Author(s) 2020. CC BY 4.0 License. occur. Additionally, by studying the spatial patterns of tropospheric BrO, it can be seen that the size of the areas where BrO explosions occur has increased.
In order to investigate the mechanisms behind these trends, tropospheric BrO and sea ice age time-series and spatial trends were evaluated. The overall increase in tropospheric BrO is, in theory, in agreement with the increase in first year ice coverage, which favours the release of BrO and is widely believed to be a main driver of the BrO explosion. However, our 5 analysis shows that there are areas where tropospheric BrO can be linked directly to locations of first year sea ice, while in other regions no clear relationship was found. Tropospheric BrO is not significantly correlated with first year sea ice extent, neither temporally nor spatially, suggesting that transport of BrO plumes away from their source regions complicates the analysis. Our comparisons hence suggest that the increase in magnitude and in the extent of the areas where BrO plumes appear is linked in a complex way to the retreat of multi-year ice and the appearance of first year ice in the Arctic region due 10 to Arctic Amplification. Although an overall increase of tropospheric BrO VCDs is found, there is also large inter-annual variability. While the time-series shows a positive trend on the order of 1.5% per year during polar springs, tropospheric BrO decreased in 2016. It remains to be seen if the positive trend in tropospheric BrO VCDs will continue and develop to be more pronounced in future years. Yang et al. (2010) investigated ozone loss using numerical modelling and found that blowing snow sourced bromine reduces tropospheric ozone amounts by up to 8% in polar spring in the Arctic. Based on 15 trends reported in the present study, future ones need to elaborate more deeply on the impact of increasing amounts of tropospheric BrO on ozone loss, mercury depletion and alteration of the oxidizing capacity of the atmosphere.
We note that the air temperature was reported to be the highest on record over the Arctic in 2016 (Overland et al., 2016). The appearance of plumes of tropospheric BrO VCDs and their intensity, are known to be influenced by meteorological drivers (air temperature, sea level pressure, wind speeds and cyclones) and the amounts of blowing snow 20 Seo et al., 2019b). Further investigation is required in order to link the evolution of tropospheric BrO to these drivers of tropospheric BrO release, in order to better understand the relationship between them and the effect of Arctic Amplification on bromine concentrations in the Arctic region.

Data availability
Part of the BrO data of this study are available through the World Data Center PANGAEA (https://doi.pangaea.de/10.1594/PANGAEA.906046). GOME2 lv1 data were provided by EUMETSAT. We acknowledge the free use of tropospheric NO 2 column data from GOME, GOME-2A, SCIAMACHY and GOME-2B sensors from the QA4ECV project (http://www.qa4ecv.eu/) and from the TEMIS web site (www.temis.nl). We acknowledge Mark Weber 5 and the UVSAT group of Institute of Environmental Physics, University of Bremen, for providing total ozone columns for GOME, SCIAMACHY, GOME2-A and GOME-2B. NCEP Reanalysis data was provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, through their Web site (https://www.esrl.noaa.gov/psd/). The EASE-Grid Sea ice age version 4 was provided by NSIDC from their website (https://nsidc.org/data/nsidc-0611/versions/4).

Author contributions 10
Ilias Bougoudis performed the retrieval of BrO from the different satellite instruments, collected and processed the sea ice age data, performed the analysis and wrote the paper. Anne-Marlene Blechschmidt performed the stratospheric separation.
Anne-Marlene Blechschmidt, Andreas Richter, Sora Seo and John Philip Burrows provided insights and knowledge on the study. Andreas Richter developed software which was used for processing and analyzing the data. Nicolas Theys developed the stratospheric separation method. All authors contributed to the writing of the paper. 15

Competing interests
On behalf of my co-authors, I declare that I have no significant competing financial, professional or personal interests that might have influenced the performance or presentation of the work described in this manuscript.