We present results from multiaxis differential optical absorption spectroscopy (MAX-DOAS) and long-path DOAS (LP-DOAS) measurements performed at the North Slope of Alaska from February to April 2009 as part of the Ocean-Atmosphere-Sea Ice-Snowpack Barrow 2009 campaign. For the first time, vertical profiles of aerosol extinction and BrO in the boundary layer were retrieved simultaneously from MAX-DOAS measurements using the method of optimal estimation. Even at very low visibility, retrieved extinction profiles and aerosol optical thickness are in good agreement with colocated ceilometer and Sun photometer measurements, respectively. BrO surface concentrations measured by MAX-DOAS and LP-DOAS are in very good agreement, and it has been found that useful information on the BrO vertical distribution can be retrieved from MAX-DOAS even in cases when blowing snow strongly reduces visibility. The retrieved BrO and extinction vertical profiles allow for a thorough characterization of the vertical structure of the boundary layer during numerous ozone depletion events observed during Barrow 2009. High BrO concentrations are usually present during the onset of ozone depletion events, and BrO disappears as ozone concentrations approach zero. The finding that elevated BrO concentrations occur mainly in the presence of high extinction near the surface strongly suggests that release of reactive bromine from airborne aerosols and/or ice particles at high wind speed plays an important role. Back trajectory calculations indicate that the particles were transported from the frozen ocean to the measurement site and that the release of reactive bromine from sea ice and/or frost flowers occurs when low temperatures (<250 K) prevail in the regions where reactive bromine is emitted.
 The presence of large amounts of reactive bromine in the polar boundary layer during polar sunrise is known to have a significant impact on the chemical balance of the atmosphere [Simpson et al., 2007a, and references therein]. In the mid-1980s, it was first discovered that ozone levels in the polar boundary layer frequently drop below detection limits [Oltmans, 1981; Bottenheim et al., 1986], and that the presence of reactive bromine in the atmosphere is the likely cause for these ozone depletion events (ODE) [Barrie et al., 1988; Hausmann and Platt, 1994]. Apart from ozone depletion, reactive bromine also alters the chemical balance within the HOx and NOx families and possibly leads to an increase of cloud condensation nuclei formation by oxidation of dimethyl sulfide [von Glasow et al., 2004]. Furthermore, the oxidation of gaseous elemental mercury by bromine radicals (Br and/or BrO) can lead to an increased input of this toxic compound into the vulnerable polar ecosystems [Barrie and Platt, 1997; Schroeder et al., 1998; Steffen et al., 2008].
 The primary source of reactive bromine in the polar boundary layer is the heterogenous reaction of gaseous hypobromous acid (HOBr) with bromide ions on saline surfaces:
 Here the index (aq) denotes aqueous phase reactions at the surfaces of ice and sea salt. In this reaction cycle, first proposed to be of importance on sulphuric acid aerosols by Fan and Jacob  and later on sea salt aerosols by Mozurkewich , two Br atoms (in the form of Br2) are emitted to the gas phase for each Br atom (in the form of HOBr) reacting at the surface. This can result in an exponential increase in gaseous reactive bromine [Tang and McConnell, 1996; Vogt et al., 1996] and is therefore referred to as the bromine explosion [Platt and Lehrer, 1997]. In principle, this reaction sequence can proceed on any saline surface with sufficient acidity (pH < 6.5) [Abbatt, 1994].
 However, the exact nature of the surfaces on which these bromine release processes actually take place is still under debate. Possible candidates are the highly saline liquid layer (brine) that forms on top of newly forming sea ice, as well as frost flowers, fragile crystals that grow on top of young sea ice [Rankin et al., 2002]. Capillary forces can lead to the uptake of concentrated brine by frost flowers, which therefore represent a further potential source for reactive bromine. Indeed, it has been shown that air masses previously in contact with newly forming sea ice [Frieß et al., 2004; Jones et al., 2006] and potential frost flower areas [Kaleschke et al., 2004; Jacobi et al., 2006] are enriched in BrO. However, back trajectory calculations in the Arctic have shown that BrO enhancements are better predicted by first-year sea ice contact than by potential frost flower contact [Simpson et al., 2007b].
 It is commonly thought that a stable boundary layer which forms during calm meteorological conditions and inhibits mixing of boundary layer air with free tropospheric air is a necessary prerequisite for the buildup of a bromine explosion [Lehrer et al., 2004], and that the main sources for reactive bromine are located on the frozen ocean, where bromine enhancements and ODEs have been detected on a much more regular basis than at coastal stations, where these events are of more episodic nature [Wagner et al., 2007; Bottenheim et al., 2009; Pöhler et al., 2010]. However, indications for bromine release from blowing snow during blizzards were reported from a coastal station in Antarctica by Jones et al. . It has been shown by chemistry transport modeling that the dispersion of saline snow crystals, originating from the sea ice and dispersed at high wind speeds, and the subsequent release of halogens from blowing snow are sufficient to explain the observed bromine release in polar regions [Yang et al., 2010]. Evidence for active bromine air-snow chemistry in polar coastal regions has been found by Simpson et al.  from measurements of halide concentrations of snow in Barrow, Alaska. The transport and deposition of saline aerosols on the snowpack possibly leads to the emission of reactive bromine far inland, as observations of active bromine chemistry in the snowpack at Summit, Greenland, indicate [Dibb et al., 2010]. The phenomenon of blowing snow is usually restricted to several tens of meters above the surface. However, during transport, airborne snow grains change their size by saltation processes, and are subject to sublimation which eventually leads to the production of small aerosol particles which can be lifted to higher altitudes and transported over larger distances [Pomeroy et al., 1997].
 The horizontal extent of bromine explosions is well known from satellite measurements [Richter et al., 1998; Wagner and Platt, 1998], which show that the areas of enhanced BrO in the springtime Arctic and Antarctica extend over several million square kilometers [Wagner et al., 2001]. Recently, progress has been made regarding a better quantification of the tropospheric fraction of the observed satellite BrO column densities [Theys et al., 2009; Salawitch et al., 2010; Theys et al., 2011], and the transport of tropospheric BrO plumes has been investigated on the basis of satellite measurements [Begoin et al., 2010]. Theys et al.  have determined typical tropospheric BrO vertical column densities (VCD) of 4–8 × 1013 molecules/cm2 in the Arctic. From satellite measurements, it is possible to derive the tropospheric vertical column density, but they contain no information on the vertical distribution of tropospheric BrO.
 Measurements of the vertical profile of reactive bromine are highly desirable since these allow to investigate the dynamical and chemical processes affecting bromine chemistry, such as the transport of BrO from the boundary layer into the free troposphere, the chemical interaction of BrO with snow and ice surfaces, and the possible role of the snowpack as bromine source. However, direct measurements of the BrO vertical distribution are very sparse and restricted to airborne observations which are limited in time and space [Neuman et al., 2010; Prados-Roman et al., 2010]. From ground-based measurements, which have the potential to observe reactive bromine on a long-term basis, information on the vertical distribution of BrO has mostly been deduced indirectly from measurements of the ozone concentration from tethered balloons or ozonesondes [Jones et al., 2010], and from height-resolved back trajectory calculations in combination with measurements of the BrO column density [Frieß et al., 2004]. An estimation of the vertical extent of the BrO layer has first been performed using MAX-DOAS measurements at Alert, Canada [Hönninger and Platt, 2002]. First comparisons of long-path DOAS (LP-DOAS) and MAX-DOAS measurements, performed at Hudson Bay, Canada, were presented by Hönninger et al. .
 Here we present active and passive differential optical absorption spectroscopy (DOAS) measurements performed in the Arctic during the Barrow 2009 intensive measurement campaign. This campaign, part of the Ocean-Atmosphere-Sea Ice-Snowpack program, took place in Barrow, Alaska (71°19′N, 156°40′W) from 26 February to 16 April 2009, and encompassed measurements of the atmospheric composition, physical, optical and chemical properties of snow, snow photochemistry, as well as dynamics and meteorology of the atmospheric boundary layer.
 During Barrow 2009, for the first time BrO and aerosol vertical profiles in the boundary layer were determined simultaneously by multiaxis DOAS (MAX-DOAS) over a period of seven weeks, while a long-path DOAS (LP-DOAS) instrument directly measured the average concentration of BrO along a well-defined light path near the surface. The instrumental setup is briefly described in section 2. The retrieval of BrO and aerosol vertical profiles from MAX-DOAS measurements is subject of section 3. Aerosol extinction profiles and aerosol optical thickness retrieved from MAX-DOAS is compared to data from colocated aerosol instrumentation in section 4. Finally, the results are presented in section 5.
2. Instruments and Data Analysis
 During the Barrow 2009 campaign, atmospheric trace gases were measured by passive (MAX-DOAS) and active (LP-DOAS) instruments. Both instruments perform spectrally resolved measurements of light that has traversed the atmosphere and was subject to absorption by numerous trace gases according to the Beer-Lambert law. This allows for determining their integrated concentration (column density) using the well-established DOAS method [Platt, 1994; Platt and Stutz, 2008].
 The basic idea of LP-DOAS measurements is to send light from an artificial source (xenon arc lamp) by a telescope through the atmosphere, which is reflected back to the telescope by retro-reflectors installed in a typical distance of several kilometers. This allows for a direct determination of the average concentration of trace gases along the light path.
 In contrast, MAX-DOAS instruments measure scattered sunlight from different viewing directions. Therefore, the quantitative determination of the light path through the atmosphere and thus the estimation of trace gas concentrations is more difficult than for LP-DOAS measurements [Hönninger et al., 2004]. In particular, the effective light path depends on the state of the atmosphere, with the aerosol extinction being one of the most important factors. However, the measurements at different elevation angles α contain information on the vertical distribution of trace gases in the lower troposphere [Sinreich et al., 2005]. Moreover, information on the aerosol extinction profile and on aerosol optical properties can be gained by measuring a trace gas with a known vertical profile, such as the oxygen collision complex O4 [Wagner et al., 2004; Frieß et al., 2006].
 Since the LP-DOAS instrument used in this study has already been described by Liao et al.  for the same measurement campaign, as well as by Pöhler et al.  for measurements onboard the RV Amundsen in 2008, and details of the spectrometer/detector unit of the MAX-DOAS instrument can be found in the work by Wagner et al. , both instruments are only described briefly here.
2.1. The Long-Path DOAS Instrument
 The LP-DOAS instrument determined the average BrO concentration along two light paths defined by the distance between the telescope and retro-reflector arrays sited at 1074 m and 3023 m distance (see Figure 1). The total light path lengths were thus 2148 m and 6046 m, and the height of the light paths above ground was approximately 2 m. The long light path was used under favorable meteorological conditions, whereas the short light path was used during periods of low visibility owing to fog, blowing snow, or occasionally occurring mirages. Depending on meteorological conditions, spectra were recorded with integration times ranging between 1 and 50 seconds. The spectral analysis of BrO was performed in the wavelength range between 315.5 and 348 nm, with optimum nighttime detection limits of 0.7 parts per trillion per volume (ppt) and 2 ppt for the long and short light path, respectively. Detection limits during the day are slightly higher (2 and 5 ppt, respectively) owing to sunlight being scattered into the telescope. Further details on the spectral retrieval and error analysis can be found in Liao et al. . A comparison of the LP-DOAS data with in situ BrO measurements by a chemical ionization mass spectrometer (CIMS) during the Barrow 2009 campaign shows excellent agreement [Liao et al., 2010].
2.2. The MAX-DOAS Instrument
 The telescope unit of the MAX-DOAS instrument collects scattered sunlight from different elevation angles (i.e., the angle between the horizon and the viewing direction) using a quartz glass prism which is rotated by a brushless DC motor. Measurements were taken sequentially at elevation angles of 90° (zenith), 20°, 10°, 5°, 2° and 1°. During twilight (SZA > 87.5°), measurements were performed at 90° and 2° elevation only. The light was focused on a quartz fibre bundle (14 × 200 μm) using a quartz lens of 100 mm focal length (F#/4). This optical setup resulted in a field of view of 0.95° which was confirmed by calibration measurements in the laboratory as follows: The telescope was pointed to a lamp located at a distance of several meters and the elevation angle was varied in steps of 0.1°. Fitting a Gaussian function to the resulting intensity profile allowed for the determination of the field of view, and simultaneously for the calibration of the elevation angle, resulting in a pointing accuracy better than 0.02°. An accurate knowledge of the viewing direction is necessary to avoid systematic errors of the retrieved trace gas and aerosol profiles [Frieß et al., 2006; Clémer et al., 2010]. To avoid freezing and condensation of water vapor, the MAX-DOAS telescope housing is heated to +28°C.
 The fibre bundle conducts the light to a spectrograph/detector unit, consisting of an Acton 300i spectrometer and a back-illuminated Andor CCD with 2048 × 512 pixels. Spectra are recorded at wavelengths between 330 and 398 nm with a spectral resolution of 0.56 nm FWHM. During the day, spectra are recorded with a total integration of 60 seconds. To increase the signal-to-noise ratio during twilight, the integration time is increased to up to 200 seconds at 95° SZA.
 The MAX-DOAS spectral analysis of BrO is performed in the wavelength interval between 346 nm and 359.5 nm, using the Windoas analysis software developed by the Belgian Institute for Space Aeronomy (BIRA) [van Roozendael et al., 2003]. This relatively small wavelength window, encompassing only two BrO absorptions bands, minimizes any interference with strong ozone absorption at shorter wavelengths and allows for analyzing the spectra using fixed Fraunhofer reference spectra [Aliwell et al., 2002]. Noontime zenith sky measurements from 26 February and 5 March were chosen as Fraunhofer reference for the analysis of spectra from 26 February to 4 March and 4 March to 16 April, respectively. To retrieve differential slant column densities (DSCDs), i.e., integrated trace gas concentrations along the light path relative to the Fraunhofer reference, the cross sections of BrO (228 K) [Wilmouth et al., 1999], OClO (233 K) [Kromminga et al., 2003], Ozone (223 K and 243 K) [Bogumil et al., 2000], O4 [Hermans et al., 2003] and NO2 (230 K and 294 K) [Harder et al., 1997] are simultaneously fitted to the logarithm of the ratio of measurement and Fraunhofer reference spectrum. A polynomial of third degree removes broadband spectral structure caused by Rayleigh and Mie scattering, and a Ring spectrum calculated according to Chance and Spurr  accounts for the filling up of the Fraunhofer lines by rotational Raman scattering. As an additional retrieval parameter, an offset to the measurement spectrum was included in the retrieval to compensate for possible instrumental stray light. The oxygen dimer O4 is analyzed with the same settings as for BrO, but in the wavelength region between 350 nm and 380 nm, where an O4 absorption band with a peak wavelength of 361 nm is located. Examples for the spectral retrieval of BrO and O4 are shown in Figure 2.
 The residual root mean square (RMS) of the MAX-DOAS BrO retrieval is typically around 5 × 10−4, resulting in statistical BrO dSCD errors of less than 2 × 1013 molecules/cm2. The statistical error of the O4 dSCDs is <5 × 1041 molecules2/cm5. For off-axis measurements relative to the fixed zenith sky reference, the O4 error is smaller than 5%. Systematic errors in the retrieved dSCDs mainly arise from the uncertainty in the absorption cross sections adapted from the literature. An error in the BrO cross section of 8% was reported by Wilmouth et al. . Significant uncertainties exist regarding the absolute value of O4 cross section, which might be overestimated by as much as 20%, as well as on its temperature dependence [Wagner et al., 2002, 2009; Clémer et al., 2010]. This leads to substantial uncertainties in the aerosol extinction profiles retrieved from MAX-DOAS O4 measurements (see section 3).
 Both LP-DOAS and MAX-DOAS telescopes were installed on the outside platform at the northeast face of the Barrow Arctic Research Center (BARC) building. The light path of both instruments pointed to the northeast, nearly parallel to the coastline in a distance of ≈500 m to the frozen ocean. The light paths and viewing directions of the instruments are depicted in Figure 1. MAX-DOAS measurements started on 26 February 2009, whereas LP-DOAS measurements started on 13 March. Both instruments measured continuously until 15 April 2009.
2.3. Complementary Data
 For the interpretation of our results, we use complementary data provided by the Barrow Atmospheric Baseline Observatory of the NOAA Earth System Research Laboratory (ESRL, http://www.esrl.noaa.gov/gmd/obop/brw) and the Barrow facility of the U.S. Department of Energy's Atmospheric Radiation Measurement (ARM) program (http://www.arm.gov/sites/nsa/C1). Both ESRL and ARM sites are located approximately 2 km east of the DOAS measurement site (see Figure 1). We use meteorological data (wind, temperature, pressure, relative humidity), in situ surface ozone concentrations from a TEI 49C ozone monitor, and ozonesonde data measured by ESRL. The aerosol extinction profiles retrieved from MAX-DOAS O4 measurements (see section 3) are compared to backscatter profile measurements performed at the ARM site by a Vaisala ceilometer model CT25K (section 4). The ceilometer has a maximum vertical range of about 7.5 km, and backscatter profiles are provided on a 30 m vertical grid. Aerosol optical thickness (AOT) retrieved from MAX-DOAS is compared to data from a CIMEL Sun photometer instrument operated on the ARM site.
 The history of air masses is investigated using back trajectory calculations (see section 5.2 calculated using the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model [Draxler and Hess, 1997; Draxler, 1999] together with the Global Data Assimilation System (GDAS) meteorological data set on a 2.5° × 2.5° grid from the National Center from Environmental Prediction (NCEP). Information on sea ice cover and sea ice properties is derived from the EUMETSAT Network of Satellite Application Facilities Ocean & Sea Ice product (OSI-SAF), which provides fractional sea ice cover and sea ice properties (discrimination of first-year or multiyear sea ice) using multisensor data from SSM/I and ERS-2 scatterometer [Eastwood et al., 2009].
 Most of these retrieval algorithms are based on the optimal estimation technique [Rodgers, 1990, 2000], which determines the most probable atmospheric state (the trace gas concentrations or aerosol extinction at a series of discrete altitude intervals) given a set of measurements y (BrO or O4 dSCDs at different elevation angles) and an a priori state vector xa. This so-called maximum a posteriori (MAP) solution is determined by minimizing the cost function
 Here, F(x, b) is a forward model which describes the measurement y as a function of the atmospheric state x. The vector b represents additional forward model parameters (e.g., atmospheric pressure and temperature profiles as well as aerosol microphysical properties) which are not retrieved. The a priori state vector xa serves as an additional constraint, which has to be considered because the information content of the measurement is usually too low to allow for a full reconstruction of the atmospheric state on the basis of the measurements only. Sε and Sa are covariance matrices describing the uncertainties in the measurement and the a priori state, respectively. The vertical resolution of the retrieval is quantified by the so-called averaging kernel matrix A = ∂/∂x, which represents the sensitivity of the retrieved profile as a function of the true atmospheric profile.
 We use a two step approach as suggested by Sinreich et al.  for the retrieval of aerosol and trace gas profiles. First, aerosol extinction profiles are retrieved from the measured O4 dSCDs. In a second step, these aerosol extinction profiles serve as forward modeling parameter for the retrieval of BrO vertical profiles. Note that the retrieval of the aerosol extinction is based on the absorption band at 360 nm, whereas the BrO analysis is performed at a wavelength window centered around 352.5 nm, where Rayleigh extinction is about 10% higher. While the resulting difference in visibility with respect to Rayleigh scattering is accounted for by the radiative transfer models, a wavelength dependence of the aerosol extinction coefficients has been neglected. However, this effect becomes small as soon as large particles, such ice and snow crystals, are present, since the scattering coefficient becomes almost independent of wavelength.
3.1. Aerosol Retrieval
 The aerosol retrieval algorithm is already described in detail by Frieß et al. . In brief, the measurement vector y for the aerosol extinction profile retrieval consists of O4 dSCDs at 360 nm measured sequentially at different of elevation angles (see section 2.2) during 15 minutes, from which the mean zenith sky dSCDs during the same time interval are subtracted. The retrieval is performed using an iterative nonlinear optimal estimation algorithm based on the Levenberg-Marquardt method [Levenberg, 1944; Marquardt, 1963]. The radiative transfer model SCIATRAN [Rozanov et al., 2005] serves as the forward model. Since the radiative transfer near the surface is frequently affected by blowing snow, the aerosol optical properties are set to values typical for clean ice crystals, with a single scattering albedo of 0.999982 and a Henyey Greenstein phase function with an asymmetry parameter of g = 0.89 [Dominé et al., 2008, and references therein]. The aerosol extinction profile is retrieved in the lowermost 2 km of the atmosphere on a 100 m vertical grid. The diagonal elements of the measurement error covariance matrix Sε are set to the square of ten times the errors in O4 dSCDs. Assuming a higher measurement error than obtained from the O4 spectral retrieval turned out to be necessary because otherwise the constraints posed by measurement would be too strong and horizontal gradients or short-term fluctuations in the aerosol distribution would lead to artifacts in the retrieved extinction profiles. On the basis of comparisons between MAX-DOAS and Sun photometer AOT measurements in Beijing, China, Clémer et al.  have reported that the absolute value of the O4 cross section may be overestimated by as much as 25% and that the measured O4 dSCDs should be scaled accordingly. However, from our data best agreement between modeled and measured O4 dSCDs, but also between retrieved AOT and Sun photometer data as well as aerosol profiles and ceilometer backscatter profiles is achieved if the O4 cross section remains equal to the originally reported value by Hermans et al.  (see section 4). This difference between the O4 observations in Beijing and in the Arctic is possibly caused by a temperature dependence in the O4 absorption cross section, which has not been accounted for in both studies.
 While aerosol extinction is almost negligible and the radiative transfer is dominated by Rayleigh scattering at low wind speeds and clear-sky conditions, blowing snow occurring during blizzards can cause an extreme increase in extinction, with visibilities as low as only several meters. Therefore the retrieval of aerosol extinction profiles in the Arctic is very challenging, and convergence of the retrieval algorithm cannot be achieved on the basis of a fixed a priori aerosol profile. As suggested by Clémer et al. , this problem can be overcome by using a variable a priori profile, which is implemented here as follows. In the first iteration of the retrieval, an extinction profile exponentially decreasing with height is used, with a scale height of 1 km and a surface extinction of 0.05 km−1. In following iterations, the a priori is set to the retrieved profile of the previous iteration, smoothed using a boxcar average with a width of 300 m. In each iteration, the a priori error, i.e., the square root of the diagonal elements of the a priori covariance matrix, is set to three times the a priori extinction, while the nondiagonal elements of Sa are set to zero. This approach ensures that convergence is achieved even if the atmospheric state is very different from the initial a priori state. However, the retrieved extinction profile is not optimal in the sense that it does not represent a maximum a posteriori solution, and the measurement errors and averaging kernels are difficult to interpret since they do not represent the true statistical uncertainties. Since a detailed error discussion of this aerosol retrieval algorithm is already provided by Frieß et al. , the validity of the retrieved aerosol profiles is tested here only on the basis of a direct intercomparison with backscatter profiles from the ARM Ceilometer instrument and AOT from the ARM Sun photometer (see section 4).
3.2. BrO Retrieval
 In contrast to the computationally expensive nonlinear aerosol retrieval, the retrieval of BrO vertical profiles represents a linear problem that can be solved in a single iteration. The forward model is represented by a weighting function matrix K = ∂F/∂x consisting of box–air mass factors (AMF) which are modeled by the Monte Carlo radiative transfer model McArtim [Deutschmann et al., 2011; Frieß et al., 2010]. Since the extinction profile varies with time, the use of lookup tables is not possible but Box-AMFs need to be recalculated for each individual retrieval using the extinction profiles derived from the aerosol retrieval in the first step. An a priori BrO profile exponentially decreasing with altitude with a scale height of 500 m and a surface volume mixing ratio (VMR) of 10 ppt is chosen. The a priori error is set to twice the a priori VMR. As an additional constraint on the smoothness of the retrieved BrO profiles, the off-diagonal elements of the a priori covariance matrix are chosen to decrease exponentially with the distance between the layers, with a correlation length of 250 m.
 The correlation between BrO surface mixing ratio from LP-DOAS and from the MAX-DOAS profile retrieval is shown in Figure 3. The data points are color coded for the surface extinction retrieved from MAX-DOAS. Good agreement is only found for small to moderate extinction values <0.5 km−1, whereas BrO surface mixing ratios are significantly overestimated by MAX-DOAS at high extinction. This illustrates the limitations of this technique under conditions of low visibility. For a surface extinction <0.5 km−1, a linear fit yields a regression coefficient of R = 0.78 and a slope of 0.75, which is caused by the aforementioned overestimation of the BrO concentration by MAX-DOAS at high extinction. The bias between MAX-DOAS and LP-DOAS is only 1.3 ppt. At this point, it is important to emphasize that LP-DOAS and MAX-DOAS represent very different measurement techniques for BrO, and that a simple correlation analysis needs to be interpreted with caution. LP-DOAS observes the average concentration along a well defined light path, whereas MAX-DOAS observes the average concentration in a layer of several hundred meters thickness (see the discussion of the averaging kernels in section 3.3). The vertical resolution varies with time, since it is a function of visibility. Furthermore, both instruments do not measure exactly the same air masses, and small delays in the observed temporal variability can reduce the correlation significantly (see the example shown in Figure 12).
3.3. Performance of the Retrieval and Vertical Resolution
Figure 4 shows a comparison between measured and modeled O4 and BrO dSCDs, simulated for the retrieved aerosol and BrO vertical profiles under clear-sky conditions. This example illustrates that the measurements can be reproduced well by the radiative transfer models. Slight disagreements between modeled and measured O4 dSCDs occurring at low elevation angles around 8:00, and at 20° elevation around 10:00, are most likely caused either by residual clouds or by mirages, which were frequently observed during the campaign. Figure 5 shows examples for the averaging kernels of the aerosol and BrO retrievals. Figure 5 (top) shows results for very low aerosol loading with an extinction of less than 0.15 km−1 in the boundary layer, corresponding to a visibility of ≈15 km at 350 nm. The averaging kernels peak at the surface with values of 0.94 and 0.72 for the aerosol and BrO retrievals, respectively, indicating that the retrieved aerosol profile has a slightly better sensitivity to the true atmospheric state than the BrO profile. The measurement is only sensitive to BrO and aerosols in the lowermost 1000–1500 m. Due to the lack of information on the BrO profile at high altitudes, the retrieved mixing ratio above ≈1500 m will be equal to the a priori (which is close to zero owing to the exponential decrease of the a priori profile). Thus elevated BrO layers at such high altitudes can hardly be detected by MAX-DOAS. Figure 5 (bottom) shows the sensitivity of the retrieval to the atmospheric state in the presence of blowing snow, when the extinction near the surface is as high as 9 km−1, corresponding to a visibility of less than 500 m. The aerosol averaging kernels are not much affected by blowing snow, and the aerosol retrieval remains almost equally sensitive as under clear-sky conditions. In contrast, the averaging kernels of the BrO retrieval during blowing snow are distinctly different to the clear-sky case. Only the surface layer remains sensitive to BrO, whereas the averaging kernels for higher layers are close to zero at all altitudes. That means that there is only one quantity that can be retrieved, namely a BrO concentration in the surface layer, which is actually representative for a weighted mean from the ground up to an altitude of ≈500–750 m. The reduced sensitivity of the BrO retrieval during blowing snow can also be seen from the reduced information content, expressed as degrees of freedom for signal (DFS), which quantifies the number of independent pieces of information that can be retrieved. For the examples shown in Figure 5, the DFS of the BrO retrieval decreases from 2.1 during clear sky to only 1.0 during blowing snow, whereas the DFS of the aerosol retrieval even shows a slight increase from 2.4 to 2.7.
4. Aerosol Retrieval Validation
 In this section, AOTs and aerosol extinction profiles retrieved from MAX-DOAS are compared to Sun photometer and ceilometer data, respectively. While Sun photometer data can be compared quantitatively to AOTs from MAX-DOAS, ceilometer data are only available as range corrected signals (in the following referred to as backscatter profiles), which can only be compared to extinction profiles from MAX-DOAS in a qualitative way. Unfortunately, the Sun photometer data during the Barrow 2009 campaign are sparse, and no Sun photometer measurements were taken before 3 April. Sun photometer measurements require direct sunlight and are restricted to short periods around noon owing to the low elevation of the Sun in early April.
 The comparison of the AOT from MAX-DOAS and Sun photometer shown in Figure 6 shows good agreement, in particular for the quality filtered Sun photometer data (cloud screening and quality control algorithms according to the AERONET standard, shown as green circles).
 The range of AOT values from the Sun photometer during this period is restricted to 0.1–2.4. In contrast, MAX-DOAS measurements can be performed even if very high extinction is present, and retrieved AOTs range from 0.05 to almost 50. Figure 7 shows the correlation between AOT from Sun photometer (quality filtered) and from MAX-DOAS. A linear fit yields a regression coefficient of R = 0.57, and a slope of 0.75 ± 0.17. Unfortunately, Sun photometer data are very sparse (only 43 data points). Therefore a robust statistical comparison cannot be performed on the basis of the existing data.
 Backscatter profiles from the ceilometer instrument are available for the whole measurement period of the Barrow 2009 campaign. The ceilometer profiles have a vertical resolution of 30 m, which is much better than for MAX-DOAS (see the averaging kernels in Figure 5). Therefore a realistic comparison between both instruments requires that the ceilometer profiles with a much higher vertical resolution are degraded to the vertical resolution of the MAX-DOAS instrument [Rodgers and Connor, 2003]. This is achieved by applying the MAX-DOAS averaging kernel A to the ceilometer profiles xc:
c represents a smoothed version of the backscatter profile xc, or more precisely the profile that the MAX-DOAS instrument would have measured if the true profile would have been the ceilometer profile. Figure 8 shows a comparison between the ceilometer and MAX-DOAS profiles for the whole measurement period from 27 February to 15 April 2010. The top plot of each set profiles in Figure 8 shows the original range corrected backscatter signal from the ceilometer instrument from 0 to 2 km. Frequent increases in the backscatter signal near the surface occur, which are correlated to high wind speeds (see section 5) and are mainly caused by blowing snow. The middle plot of each set of profiles in Figure 8 shows the ceilometer backscatter profiles degraded to the vertical resolution of the MAX-DOAS instrument according to equation (2). As expected from the shape of the averaging kernels, this procedure causes a smaller signal at high altitudes as well as an attribution of the backscatter signal at high altitudes to lower layers. Strong backscatter signals seen by the ceilometer at high altitudes (e.g., on 9–10 March as well as 1 and 9 April), cannot be detected by MAX-DOAS. In these cases, the extinction layer retrieved by MAX-DOAS exhibits a smaller vertical extent. To account for these differences in vertical sensitivity, the degraded backscatter signal in the MAX-DOAS backscatter plots in Figure 8 can be directly compared to the extinction profiles from MAX-DOAS shown in the bottom plots of each set of profiles. The extinction profiles from MAX-DOAS are in excellent agreement with the degraded ceilometer measurements regarding the vertical structure, the relative magnitude of the signals, and the temporal variability. Note that the color scales of both ceilometer backscatter and MAX-DOAS extinction are shown on logarithmic scale, and agreement between both data sets is achieved over an extinction range of 4–5 orders of magnitude. A notable difference between both data sets is that MAX-DOAS extinction profiles sometimes indicate uplifted layers with lower extinction in the surface layer than above, e.g., from 12–14 April, whereas the ceilometer profiles exhibit a high signal down to the ground. However, although the ceilometer data are range and sensitivity corrected, it is subject to higher uncertainties in the lowermost 50 m owing to the limited overlap between laser beam and the field of view of the receiving optics.
 In summary, the very good agreement between extinction profiles retrieved from MAX-DOAS and backscatter profiles from the ceilometer, for cases ranging from an almost pure Rayleigh atmosphere to blizzard conditions with visibilities of only several meters, demonstrates that MAX-DOAS represents a reliable and very robust technique for the retrieval of vertical extinction profiles and that it is particularly well suited for cases with extremely high extinction in the boundary layer, when Sun photometer measurements are not possible. In contrast to ceilometer or lidar measurements, which only measure the backscatter signal, extinction profiles can be directly inferred from MAX-DOAS measurements. A further advantage of MAX-DOAS is that it is most sensitive for the extinction at the surface, where lidar and ceilometer instruments cannot measure owing to the lack of overlap between laser beam and field of view of receiving optics. The main limitations of MAX-DOAS are the lack of sensitivity for altitudes >2 km (and for even lower altitudes in the presence of blowing snow), and the much lower vertical resolution compared to ceilometer and lidar instruments.
5. Results and Discussion
Figures 9 and 10 provide an overview of the DOAS measurements performed during the Barrow 2009 campaign. Numerous ODEs occurred from late February to mid April 2009 (numbered from 1 to 12 in the surface ozone time series). Some ODEs exhibit only a partial destruction of ozone (events 3, 6, 9 and 12), while for others a total destruction of ozone down to a VMR of less than 5 parts per billion per volume (ppb), and sometimes down to the detection limit of 1 ppb, occurs (events 1, 2, 4, 5, 7, 8, 10 and 11).
 A calibration of the ESRL ozone monitor versus a WMO standard reference photometer revealed very good agreement with the standard, with a bias of only 0.15 ppb [Zellweger et al., 2008]. However, values of <0.5 ppb were not observed by the ESRL ozone monitor. This either indicates that events with less than 0.5 ppb of ozone do not occur, or that for very low ozone concentration the ozone monitor data have an offset of 0.5–1 ppb.
 The third plot each set of profiles in Figures 9 and 10 show a comparison of the BrO surface VMR by the LP-DOAS instrument (black symbols), and retrieved from the MAX-DOAS measurements (green). The former is directly measured approximately 2 m above the ground, whereas the latter represents an estimate for the average BrO concentration in the lowermost 100 m of the atmosphere. Despite this difference in the way air masses are sampled, both data sets show very good quantitative agreement during calm weather conditions, e.g., from 27–29 March and from 6–8 April. Even when high wind speeds of about 10 m/s lead to high extinction by blowing snow (e.g, from 30–31 March and 9–12 April), the diurnal variation of the BrO surface concentration from MAX-DOAS is still in good qualitative agreement with the LP-DOAS measurements, although the high variability in the BrO dSCDs caused by the strong fluctuations of the atmospheric conditions leads to a larger scatter of the data and some outliers with unrealistically high BrO VMR retrieved from MAX-DOAS. The very good overall agreement between MAX-DOAS and LP-DOAS provides high confidence that both surface VMR and vertical extent of the BrO layer can be retrieved realistically from the MAX-DOAS measurements.
 In contrast to BrO measurements performed offshore on the frozen Arctic Ocean [Pöhler et al., 2010], the occurrence of enhanced BrO concentrations and ozone depletion appears to be much more irregular at coastal sites. Most ODEs observed during Barrow 2009 are initially coincident with enhanced BrO, but sometimes partial ozone depletion is also observed in the absence of BrO, in particular if the duration of the ODE is short (events 4 and 9). These air masses were possibly subject to ozone depletion by BrO a long time ago, and reactive bromine has been removed by dry or wet deposition. After high concentrations of BrO had been reached via reaction (R1) at the onset of ODEs, BrO disappears again as soon as ozone drops to values close to or below 1 ppb (events 2, 7, 11, and at the end of event 12), a behavior that has already been observed during previous measurements in Barrow [Simpson et al., 2007b]. This finding can be expected from photochemistry since in the absence of ozone Br cannot be converted to BrO by reaction (R2) anymore, and the Br/BrO steady state is shifted toward Br (which is subsequently converted to HBr by reaction with, e.g., HO2 or HCHO) as O3 concentrations approach zero. This behavior can be seen very well on 14 April (see Figure 11), when ozone suddenly drops from 2–3 ppb to the detection limit of 1 ppb at 10:00 local time (LT). A similarly sharp drop in BrO surface concentration from 10 ppt to ≈2.5 ppt is observed both by LP-DOAS and MAX-DOAS. The MAX-DOAS BrO profiles show that this reduction in BrO occurs within the entire boundary layer from 0 to 1000 m. This situation persists for about 5 h, until ozone and BrO increase simultaneously to 2 ppb and 10 ppt, respectively. Although the wind direction gradually changes from south to north in the morning of this day, there seem to be no sudden changes in meteorology (wind direction and speed, temperature, humidity, aerosol load) that could trigger this event.
 Most events of enhanced BrO are accompanied, at least initially, by the presence of a high extinction in the boundary layer, whereas low extinction is usually coincident with low BrO amounts. A striking example for this correlation between BrO and extinction is shown in Figure 12 for the period from 11 to 13 April, when highest BrO VMR during Barrow 2009 of 36 ppt were measured by LP-DOAS. During this period, easterly and southeasterly winds with velocities of 5–10 m/s lead to blowing snow with a surface extinction of up to 1 km−1. Apart from some outliers of the MAX-DOAS data, probably caused by a strong temporal variation of the extinction, the BrO surface VMR of both MAX-DOAS and LP-DOAS are very well correlated with the surface extinction, in particular on 13 April when both surface extinction and BrO VMR are highly variable. This finding strongly suggests that, in this situation, the release of BrO occurs in situ via heterogeneous reactions on the surface of airborne particles, instead of BrO being released directly on the sea ice surface and then transported to the measurement site. Note that on 13 April, BrO and surface extinction from MAX-DOAS are shifted in time relative to the LP-DOAS measurements by 30–45 min (see Figure 12). Given a wind speed of 3–5 m/s, the corresponding spatial distance of 5.5–13.5 km reflects the fact that MAX-DOAS observations are representative for air masses at larger distances than LP-DOAS, which yields the average VMR between the telescope and the retro-reflector.
5.1. Comparison With Ozonesondes
 As can be seen in Figures 9 and 10, the vertical extent of the BrO layer is usually less than 300 m, with the exception of 12 April, where the BrO layer has a vertical extent of ≈500 m. In contrast, the extinction profiles indicate that, depending on the wind speed, blowing snow and/or aerosol particles produced by a sublimation of snow grains, can be present in a layer of up to 1 km thickness. This difference in layer height can, however, be partly caused by the different height resolution of BrO and aerosol retrieval (see section 3). A more detailed picture of the vertical structure of the boundary layer is provided by Figure 13, where potential temperature and ozone profiles from ozonesondes are compared to BrO and extinction profiles from MAX-DOAS for four ozone soundings during low ozone episodes. The thick vertical bars at the bottom of the BrO profiles from MAX-DOAS (Figures 9 and 10, third set of plots in each profile) represent the 15 min average of the BrO surface VMR from LP-DOAS. This again illustrates the very good agreement between BrO surface VMR from MAX-DOAS and LP-DOAS measurements.
 On 18 March, a strong surface inversion is present from the surface up to an altitude of ≈75 m, followed by a nearly neutral stratification up to an altitude of 320 m and stable conditions with a positive potential temperature gradient above. Ozone is moderately depleted in the lowermost 500 m, with O3 mixing ratios of ≈20 ppb. Both MAX-DOAS and LP-DOAS measure a BrO surface VMR of 13 ppt, but much higher BrO amounts of more than 40 ppt are present in an elevated layer between 100 and 200 m altitude. At moderate wind speeds of 3–5 m/s, the extinction has a maximum of 0.25 km−1 at the same altitude as the BrO concentration maximum. A similar potential temperature profile was present on 22 March, when a severe ozone depletion with mixing ratios of less than 4 ppb persisted in the entire boundary layer up to an altitude of 400 m. Again, the highest BrO amounts (≈20 ppt) are detected by MAX-DOAS in an elevated layer between 200 and 300 m, whereas a relatively low surface concentration of only 4.5 ppt is measured by LP-DOAS. A strong surface inversion with a vertical extent of 25 m is coincident with slightly higher ozone mixing ratios of 1.5 ppb than above. A very shallow layer of ozone depleted air, with surface concentrations of 7.3 ppb and a vertical extent of ≈200 m, is present on 8 April. The increase in potential temperature with height indicates a stable stratification, and calm conditions with wind speeds of less than 5 m/s at the time of the ozone launch lead to a small extinction of less than 1 km−1. Given that the ozonesonde has passed the lowermost 200 m during 90 s, and a response time of the electrochemical ozonesonde of ≈20 s [Davies et al., 2000], the true vertical extent of the ozone depleted air is likely to be somewhat smaller. In this situation, the BrO surface concentration from the MAX-DOAS profile inversion (3.0 ppt) is significantly smaller than from LP-DOAS (9.6 ppt). This example illustrates the limited ability the MAX-DOAS technique to determine steep gradients in the vertical concentration profile as quantified by the averaging kernels discussed in section 3. On the other hand, given a BrO VCD of V = 1.1 × 1013 molecules/cm2 as retrieved from MAX-DOAS and a BrO surface concentration of ρ = 3.0 × 108 molecules/cm3 from LP-DOAS, a vertical extent of the BrO layer of z = V/ρ ≈ 350 m can be estimated, which is in the same order of magnitude as the vertical extent of the ozone depleted air masses observed by the ozonesonde. The ozonesonde from 26 March was launched during the most severe ODE with near-zero surface ozone for more than 3 days (event 7 in Figure 10). A neutral stratification (∂Θ/∂z ≈ 0) with wind speeds of ≈5 m/s during the time of the ozonesonde launch represents favorable conditions for a well mixed boundary layer, and the entire boundary layer is completely depleted in ozone (VMR < 1 ppb) up to an altitude of 560 m. This ODE is characterized by a high extinction of up to 10 km−1 in the boundary layer. Since the wind speed is actually too low to generate blowing snow, and also because the extinction is increased up to high altitudes of several hundred meters, it is likely that the high extinction is caused by aerosol particles produced by sublimation of blowing snow. As discussed in section 3, the MAX-DOAS measurements are only sensitive for surface layer where MAX-DOAS and LP-DOAS BrO VMR agree well, and the retrieved BrO vertical profile at higher altitude is highly uncertain as a result of the high extinction present in the boundary layer. Similar conditions as on 26 March were present on 12 April: the potential temperature is constant up to an altitude of 400 m, and high wind speeds of up to 9 m/s lead to blowing snow and/or aerosols. The MAX-DOAS retrieval indicates a maximum extinction of 0.6 km−1 between 200 and 300 m, and BrO VMR increases from 27 ppt at the surface (measured by LP-DOAS) to a maximum of 44 ppt between 100 and 200 m.
5.2. The History of Air Masses
 A necessary prerequisite for the enrichment of air masses with reactive bromine is the contact with saline surfaces. As already discussed in section 1, possible sources for reactive bromine are newly formed sea ice and frost flower fields. Therefore it is instructive to investigate the history of the air masses arriving at the measurement site using back trajectories in combination with sea ice data, a technique that was first employed by Frieß et al. , who used it for the interpretation of BrO measurements in Antarctica, and was later applied by Simpson et al. [2007b], who showed from measurements in Barrow that the observed BrO amount is correlated better with a contact of air masses with first-year sea ice than with potential frost flowers.
 Here we determine the history of air masses arriving at the Barrow measurement site at a given altitude and time using 72 h back trajectories, calculated with the Hysplit model for each hour of the campaign for arrival altitudes from the surface up to 4 km in steps of 100 m. It was assumed that an exchange between the surface occurs whenever the altitude of the air parcel is below a threshold value of z0 = 500 m, which is the typical vertical extent of the Arctic springtime boundary layer (see section 5.1). Using OSI-SAF data on fractional sea ice coverage ρice and sea ice properties (first-year or multiyear sea ice), properties of the air parcel which might influence bromine release were determined. (1) The sea ice contact time tice is the sum of the time the air parcel spent over trajectory points where first-year sea ice is present and z < z0, multiplied with the fractional sea ice coverage ρice for this area. (2) Similar to work by Simpson et al. [2007b], the potential frost flower (PFF) contact time is calculated as the sum of the fractional open water coverage (1 − ρice) for trajectory points where z < z0, multiplied with the temperature-dependent probability that frost flowers are formed using the parametrization of Kaleschke et al. , and multiplied with the time the air parcel spent over this area. (3) Since solar radiation is a main driver of bromine photochemistry, the time-integrated solar flux Fsol (in units of kWh/m2) after the first contact with first-year sea ice is determined from the GDAS meteorological field. (4) We define the quantity tice as the average temperature of the air parcel while it was located over first-year sea ice if z < z0. It is important to note that these back trajectory calculations are subject to substantial uncertainties. For the Arctic, horizontal displacements of the modeled air parcels for a 5 day model run can be up to 1000 km [Kahl et al., 1989].
Figures 14 and 15 show the results of the back trajectory calculations, together with the BrO (Figures 14 and 15, first plot in both sets of profiles) and extinction (Figures 14 and 15, second plot in both sets of profiles) vertical profiles derived from MAX-DOAS. In most cases, air masses which were in contact with first-year sea ice (Figures 14 and 15, third plot in both sets of profiles) were also in contact with PFF areas (Figures 14 and 15, fourth plot in both sets of profiles). In contrast to the findings of Simpson et al. [2007b] for measurements in Barrow during 2005, it cannot be concluded from the Barrow 2009 data that first-year sea ice contact predicts BrO levels better than PFF contact (partly because both quantities are rather well correlated in our case). In general, high contact times with first-year sea ice and PFF appear to be coincident with high extinction, indicating that air masses coming from the sea contain large amounts of aerosols and/or ice particles on which heterogeneous release of reactive bromine might occur. On some occasions, such as 28 February to 1 March and 5–6 April, air masses at altitudes of up to 2 km were in contact with first-year sea ice and PFF, and very similar vertical structures were observed for the extinction signal, but not for BrO, which might be a result of the limited sensitivity of the BrO retrieval for higher altitudes when large amounts of aerosols are present (see section 3). However, in contrast to previous studies in Barrow by Simpson et al. [2007b], the correlation coefficient between sea ice and PFF contact time and BrO concentrations for the air masses arriving at the surface are very small (R < 0.2). This indicates that a simple parametrization of BrO release on the basis of ice contact time is not sufficient to explain the complex process of bromine release without considering other meteorological and microphysical parameters.
 An important prerequisite for the presence of BrO at the measurement site appears to be low temperatures tice at the time when the air mass was over the source area (Figures 14 and 15, sixth plot in both sets of profiles). Note that tice is not related to the local temperature at the time the air parcel has arrived at the measurement site (see the local temperature shown in Figures 9 and 10), which indicates that the impact of temperature on the stability of the boundary layer is of minor importance for the accumulation of BrO. Periods with tice < 250 K are coincident with elevated BrO concentrations, for example 6–7 March, 13–19 March, 29–31 March, and 10–12 April (see also the BrO surface concentrations shown in Figures 9 and 10). A correlation between high BrO and/or ozone depletion with low temperatures was already found in the past [Tarasick and Bottenheim, 2002; Bottenheim et al., 2009; Pöhler et al., 2010], and was attributed to the temperature dependence of the thermodynamical properties of the ice surfaces, such as the conditions of the quasi-liquid layer and the increase in uptake of HOBr by saline surfaces below a temperature of 252 K. Furthermore, model calculations predict that the precipitation of calcium carbonate from sea ice, which occurs almost completely at temperatures below 260 K, is an important prerequisite for the release of BrO since this process facilitates acidification [Sander et al., 2006].
 From our MAX-DOAS measurements during the Barrow 2009 campaign, for the first time simultaneous retrievals of the vertical distribution of both BrO and extinction by aerosols and snow/ice particles were continuously performed for a period of seven weeks.
 The Arctic environment is subject to extreme changes in visibility, ranging from ≈50 km (sole Rayleigh scattering at 350 nm) under clear-sky conditions down to only several meters during blizzards. Under these varying conditions, the retrieval of vertical profile information from scattered light measurements of the O4 dSCD is very challenging. Despite these adverse conditions, the extinction profiles retrieved from MAX-DOAS turned out to be in very good qualitative agreement with backscatter profiles from a colocated ceilometer instrument, after the latter were degraded to the vertical resolution of the MAX-DOAS instrument. The good agreement between the integrated extinction profiles from MAX-DOAS and AOT from a Sun photometer instrument confirms that the MAX-DOAS aerosol retrieval is valid at least for AOT values between 0.1 and 2.4 available from Sun photometer.
 BrO vertical profiles were retrieved from MAX-DOAS observations using BrO dSCDs as well as the extinction profiles from the aerosol retrieval as input. Although MAX-DOAS and LP-DOAS determine the BrO amount in very different ways, BrO mixing ratios retrieved from MAX-DOAS in the lowermost layer (0–100 m) are in very good agreement with the average concentration along a well-defined light path of several kilometers length measured directly by LP-DOAS. Good agreement between MAX-DOAS and LP-DOAS BrO measurements is achieved even if high extinction by snow and ice crystals is present at high wind speeds, although MAX-DOAS data show a larger scatter and often a significant overestimation of the BrO mixing ratio under these conditions.
 The averaging kernels of the retrievals show that, depending on the visibility, MAX-DOAS measurements are mainly sensitive for the lowermost 500–1000 m of the atmosphere. The vertical resolution of the profile retrievals near the surface under clear-sky conditions lies typically between 100 and 250 m, with the aerosol retrieval having a better resolution than the BrO retrieval. The sensitivity of the aerosol retrieval is only slightly affected by the amount of extinction, whereas the sensitivity for BrO at higher altitudes decreases as extinction increases. Depending on visibility, the number of independent pieces of information of the BrO retrieval (degrees of freedom for signal) varies between 1.0 during blizzards and 2.1 under clear-sky conditions. The information content with respect to aerosols does not depend much on visibility, with typically 2.5 pieces of information.
 Numerous ODEs were observed during the Barrow 2009 campaign. As in earlier studies, these were not always coincident with enhanced BrO. A tendency of high BrO concentrations at the onset of ozone depletion events was observed, but sometimes the air was depleted in ozone in the absence of BrO. In some cases, when ozone was only partially depleted, this lack of anticorrelation between BrO and ozone was possibly caused by the presence of aged air masses, where reactive bromine was already removed or sequestered, but ozone concentrations were still low. However, the absence of BrO throughout the boundary layer at very low ozone concentrations, below the detection limit of 1 ppb, was more likely caused by a change in partitioning between Br and BrO.
 In most cases, periods of elevated BrO were coincident with an increase in extinction at wind speeds larger than 5 m/s, when snow and ice particles become airborne. This strongly suggests that heterogeneous chemistry takes place in situ on these particles, or on aerosols produced by sublimation of the dispersed snow grains, causing the release of reactive bromine to the gas phase.
 Vertical profiles of ozone and temperature from the regular ozone soundings performed at Barrow show that ozone is depleted throughout the polar boundary layer, which has a typical vertical extent of 500 m and is capped by a temperature inversion. In view of the limited vertical resolution of MAX-DOAS, the retrieved profiles show that BrO is confined to the boundary layer. In many cases, surface BrO concentrations are smaller than at higher altitudes. This presence of an elevated BrO layer is possibly caused by a loss of reactive bromine at the surface of the snowpack. In one case, a strong surface inversion from the ground up to an altitude of only 25 m was significantly depleted in ozone, while the BrO profile from MAX-DOAS shows that BrO was only enhanced in the lowermost (0–100 m) retrieval layer.
 The history of the air masses was investigated using back trajectory calculations in combination with sea ice maps. Increased extinction is observed whenever the wind speed is high and air masses come either from areas of first-year sea ice or from areas where frost flowers are potentially present. A likely explanation for these findings is that frost flowers and saline ice particles become airborne and are transported to the measurement site, where reactive bromine is emitted locally from these particles. BrO enhancements were not correlated with the local temperature at the measurement site, but temperatures below 250 K at the potential source regions (first-year sea ice or potential frost flower areas) are in agreement with the hypothesis that changes in the thermodynamic properties of sea ice at low temperatures trigger the bromine explosion.
 This work was financially supported by the German Research Association (DFG), projects FR2497/2-1 and PL193/10-1. We would furthermore like to acknowledge the European Community Research Infrastructure Action under the FP6 Programme, EUSAAR (contract RII3-CT-2006-026140), for the support of our aerosol retrieval activities. Holger Sihler is financially supported by the International Max Planck Research School (IMPRS). Caroline Fayt and Michel van Roozendael from BIRA are gratefully acknowledged for providing the WINDOAS analysis software. Many thanks to Alexei Rozanov from IUP Bremen for providing the SCIATRAN radiative transfer model. Meteorological data, surface ozone, and ozone profiles were provided by NOAA ESRL, and ceilometer profiles were provided by ARM. Back trajectories were calculated using the HYSPLIT model from NOAA together with the GDAS data set from NCEP.