Seasonal variation of tropospheric bromine monoxide over the Rann of Kutch salt marsh seen from space

The Rann of Kutch (India/Pakistan) is one of the largest salt deserts in the world. Being a so-called ’seasonal salt marsh’, it is regularly flooded during the Indian Summer Monsoon. We present 10 years of bromine monoxide (BrO) satellite observations by the Ozone Monitoring Instrument (OMI) over the Great and Little Rann of Kutch. OMI spectra were analysed using Differential Optical Absorption Spectroscopy (DOAS) and revealed recurring high BrO VCDs up to 1.4×1014 molec/cm during April/May, but no significantly enhanced column densities during the monsoon season (June–September). 5 In the following winter months, the BrO VCDs are again slightly enhanced while the salty surface dries up. We investigate a possible correlation of enhanced reactive bromine concentrations with different meteorological parameters and find a strong relationship between incident UV radiation and the total BrO abundance. In contrast, the second Global Ozone Monitoring Instrument (GOME-2) shows about four times lower BrO VCDs over the Rann of Kutch than found by OMI and no clear seasonal cycle is observed. One reason for this finding might be the earlier local overpass time of GOME-2 compared to 10 OMI (around 9:30 vs. 13:30 LT), as the ambient conditions significantly differ for both satellite instruments at the time of the measurements. Further possible reasons are discussed and mainly attributed to instrumental issues. OMI additionally confirms the presence of enhanced BrO concentrations over the Dead Sea valley (Israel/Jordan), as suggested by former ground-based observations. The measurements indicate that the Rann of Kutch salt marsh is probably one of the strongest natural point sources of reactive bromine compounds outside the polar regions and is therefore supposed to have an significant impact on 15 local and regional ozone chemistry.

rical BrO VCDs (appropriate for an stratospheric absorber) calculated for each pixel row individually over a remote area over the Pacific (±20 • N, 105-175 • W) were subtracted from all daily measurements. Since the contribution of stratospheric BrO to the total column is hereby removed as well, the median BrO column across the track was re-added. Please note that the BrO VCDs finally used in this study are corrected for the stratospheric contribution to the total column by a specifically local separation (see Sect. 3.2).

5
The measured spectra are additionally affected by permanent and transient hot pixels, leading to increased noise at certain wavelengths for individual CCD detector rows. Such erroneous measurements can be easily identified by abnormally large fitting residuals at affected wavelength positions and again lead to strongly elevated BrO VCD stripes. Following the suggestions by Chance (2007), intensities measured at pixels showing very strong residual discrepancies (>4σ) from an initial fit were excluded for a final second fit. The vast majority of former suspiciously anomalous BrO slant column densities (SCDs) showed 10 realistic columns similar to nearby measurements after the correction.

Clouds
The observation of tropospheric trace gases is significantly affected by radiative transfer due to cloud coverage in several ways. If the trace gas is located above a thick cloud layer, satellite measurements might show an increased sensitivity. On the other hand, clouds may completely shield trace gases close to the surface. To minimize the influence of clouds, only OMI 15 measurements with an effective cloud fraction of less than 30% were considered. For this purpose two operational Level 2 OMI cloud products are provided by NASA: 1. OMCLDO2, for cloud detection using the O 2 -O 2 absorption near 477 nm (e.g. Acarreta and de Haan, 2002;Vasilkov et al., 2008) 2. OMCLDRR, using information of the so-called filling-in of solar Fraunhofer lines caused by rotational-Raman (RR) 20 scattering in the atmosphere (Ring effect) within the 346-354 nm spectral range (e.g. Joiner et al., 2002;Stammes et al., 2008) Although the OMCLDO2 algorithm is most commonly used, it turned out that the algorithm almost always mis-classifies the bright surface of the salt marsh as cloud and is thereby unsuitable for the BrO analysis presented here. In contrast, the OMCLDRR algorithm seems to better distinguish between the bright surface and clouds, probably because of an increased 25 contrast in the UV in comparison to the visible wavelength range that is used in the OMCLDO2 algorithm. It should, however, be noted that a reanalysis of the data applying a lower cloud filter threshold of only 20% showed that the OMCLDRR algorithm sometimes misinterprets the bright surface of the salt marsh as clouds as well. This is in particular the case during the first 3-4 months after the Rann is flooded (October/November-December/January). As the remaining water evaporates, a very clean and bright surface remains as indicated by increased reflectivity; this effect can be seen in MODIS true color images as in Figure   30 2. In order to provide statistically relevant monthly averaged BrO VCDs, only grid pixels that were covered at least ten times by the daily measurements were finally taken into account. It can, however, not be completely ruled out that some of the actual cloud free measurements are sorted out using the cloud filter. 5 Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-90, 2016 Manuscript under review for journal Atmos. Chem. Phys. Published: 15 February 2016 c Author(s) 2016. CC-BY 3.0 License.
For GOME-2, the only available cloud detection algorithms are the operational FRESCO (Koelemeijer and Stammes, 2001;Koelemeijer et al., 2002) and ROCINN (Loyola R., 2004;Rozanov et al., 2006) products, which both use wide parts of the oxygen A-band spectrum to determine the effective cloud fraction. Similar to the OMI OMCLDO2 algorithm, both FRESCO and ROCINN fail to distinguish between the bright surface of the salt marsh and clouds. Therefore only a qualitative comparison of the unfiltered cloudy data will be presented in Section 4.5.

Local stratospheric background correction
Because all satellite measurements include the total atmospheric BrO column, the data have to be corrected for the stratospheric fraction to retrieve the tropospheric BrO column over the salt marsh area. The stratospheric BrO distribution varies little with latitude and even less with longitude (Theys et al., 2009b). Therefore, BrO VCDs determined assuming a geometrical airmass factor were corrected by subtracting the results of a two-dimensional spatial polynomial fit of 3rd degree n applied to the daily 10 measurements as described in Hörmann et al. (2013): where V * strat,i are the fitted stratospheric BrO VCDs at the centre coordinates x and y [ • ] of satellite pixel i within a large area around the Rann of Kutch (18-30 • N, 62-78 • E). To minimize the influence of possibly enhanced BrO VCDs over the salt marsh, the actual Rann area (22.5-25.5 • N, 67.5-72.5 • W) was excluded from the polynomial fit of the stratospheric correc-15 tion approach. The resulting corrected geometrical BrO VCDs (V * trop ) were reconverted into 'tropospheric' SCDs (S * trop ) by multiplication with the geometrical AMF.

Radiative transfer
As the radiative transfer for a tropospheric absorber is not adequately represented by the geometrical approximation, the OMI observations at 342.5 nm (via the NASA OMAEROG Level 2 aerosol product, described in Torres et al., 2007). As the OMAEROG product often shows unrealistically high parameter values exclusively over the Rann area (e.g. monthly mean AODs of more than 4), the product seems to have some problems with the exceptionally bright surface of the salty desert.
Additionally, the associated parameters seem to be affected by increased cloud coverage during the monsoon season.
In order to estimate the actual aerosol parameters despite these problems, mean values for AOD, SSA and AP close to (but 5 outside of) the Rann area were chosen. An estimate of the uncertainties caused by the a priori assumptions will be given in Section 4.2.

Results
The individual daily measurements of each month from October 2004 to December 2014 were gridded on a regular lat-lon grid with a spatial resolution of 0.1 • , from which monthly mean BrO VCD maps were calculated. Figure 3 shows the monthly mean

Variation of surface reflectivity and sensitivity
The strongly enhanced albedo of the salty crust is expected to have a significant effect on the sensitivity for near-surface BrO.
This may lead to an apparently enhanced BrO VCDs over the Rann when compared to nearby areas featuring a lower albedo due to a potential enhanced tropospheric BrO background and the underestimation of the corresponding AMF. To investigate the possible influence of the bright surface on the spatial pattern of BrO VCDs, the reflectivity at 331 nm as seen by OMI (see 5 Section 3.3) was used to calculate monthly averaged maps for the same data that were used for the BrO VCD maps (i.e. using a cloud filter CF<30%). Figure   where the uncertainty represents the corresponding standard deviation σ. In contrast, the mean mass over the reference area was found to be about a factor of 13 lower (212±105)t. The deviation from zero can probably be attributed to the imperfect polynomial correction for the stratospheric BrO background. During the monsoon season (June-September) the BrO mass declines over the Rann and remains at a comparatively low level of (1066±753)t, often followed by a local maximum directly after the monsoon (October) before another minor drop leads to an annual minimum around December. While the mean 5 BrO masses over the Rann during the monsoon are still about 3x larger than over the corresponding reference area for most years (279±138)t, they are associated with a high uncertainty due to the influence of cloud coverage and associated small number of measurements. In this context it should be mentioned that the reference area is generally less affected by (convective/orographic) clouds as large parts are located over the Arabian Sea. During wintertime (November-February), the mean BrO masses remain at the same low level (1067±560)t for the area over the Rann, (187±131)t over the reference area), how-

Correlation with meteorological parameters
The influence of meteorological parameters on the 'bromine explosion' reaction cycle is still largely unclear and remains a subject of intensive discussion within the scientific community. Whereas e.g. the role of blowing snow in the activation of bromine over first-year Antarctic sea ice has recently been analysed in situ (Lieb-Lappen and Obbard, 2015), the influence of environmental conditions like temperature, relative humidity or pH value at salt lakes were almost exclusively investigated by 25 using an artificial salt pan within a smog chamber (Buxmann et al., 2012). It was found that the corresponding BrO mixing ratios were almost one magnitude higher at a relative humidity of 60% when compared to experiments at 37% or 2%. This is probably caused by quasi-liquid water layers on the salty crust that seem to support the activation of reactive bromine.
To investigate possible dependencies of the observed total BrO mass over the Rann of Kutch on humidity and other parameters,  Fig. 6g) that is well capable of describing all main features of the mean annual BrO observations (r 2 =0.95). Interestingly (but self-evident), the features that were contributed by accounting for the precipitation appear to be widely included in the CC parameter for LF4, as both parameters are closely related to each other (the second highest r 2 value for LF3, 0.86, was achieved by using CC instead of P).
The seasonal cycle of BrO formation over the Rann of Kutch can be described reasonably well as a linear function of meteo-25 rological parameters for the 10 years averaged data on a monthly basis. It should, however, be noted that this simple approach fails to adequately reproduce the seasonal cycle when monthly averages from individual years are considered. In particular, the less pronounced BrO maxima in 2007/2008/2011/2012 are clearly underestimated and also the highly variable winter months are not very well captured by the linear functions. It has to be kept in mind that a linear model assumption of independent variables constitutes a rather simplistic approach to describe the satellite observations and can therefore only be used to get a 30 rough idea about the general circumstances that are needed for the extensive formation of BrO over the salt marsh.

Seasonal wind pattern
Other possible parameters that might (at least indirectly) influence the 'bromine explosion' mechanism and the observed BrO spatial pattern are wind speed and main direction. December, which is dominated by wind coming from the north. Aerosol particles that are needed for the 'bromine explosion' mechanism might therefore have an increased probability to be swirled up from the salty surface crust in significant amounts during springtime. An examination of the actual aerosol appearance over the Rann remains, however, difficult, as available satellite aerosol products are strongly influenced by the bright surface of the salt marsh (see also Section 3.3). 20 For the Little Rann of Kutch, the wind direction has a much lower influence on the residence time of the air over the salt marsh, as this part of the salt marsh is much smaller and only slightly stretched in east-western direction (≈ 50×80 km).

Comparison of OMI results with GOME-2 observations
The GOME-2 instrument overpasses the Rann of Kutch at about 9:30 local time (and therefore 4 hours before OMI). To investigate the diurnal evolution of BrO VCDs over the Rann area, 5 years of GOME-2 data (2007)(2008)(2009)(2010)(2011) were evaluated for BrO 25 in a similar way as for OMI (see Section 3 for details). As mentioned in Section 3.1, no cloud filter could be applied without losing a significant number of probably cloud free GOME-2 observations because the operational cloud products mistakes the bright surface for clouds. Therefore, only a qualitative seasonal intercomparison of the data will be presented in the following: No cloud filter is used and consequently, a geometrical AMF is used to calculate BrO VCDs (i.e. the data are only corrected for the viewing geometry; explicit radiative transfer calculations are not involved). Please note that the geometrical AMF is this 30 time also applied to OMI data to guarantee the consistency of both data sets. Because fully cloudy cases are not filtered out, low BrO VCDs are not only due to small near-surface BrO concentrations, but also to cloud shielding (possibly even outside of the monsoon season). However, it should be noted that most days outside the monsoon months (June-September) are only effected by low cloud fractions according to MODIS and ECMWF data. In addition, no differences in cloud cover were found As it can be seen from Figure 9 (lower panel), the seasonal mean BrO VCD maps only indicate a very low enhancement over the salt marsh, which is close to background noise level. Please note that the GOME-2 BrO VCDs are presented using the same colorbar as the corresponding OMI results (Figure 9, upper panel) to emphasize the low levels of the GOME-2 BrO VCDs.
During the monsoon season (here July-September, Figure 9, lowermost left panel), neither satellite instrument detects a clear 5 enhancement, as massive cloud coverage shields the salt marsh area, but whereas OMI data clearly show a seasonal cycle with a pronounced maximum in April-June, no obvious seasonal variation can be seen in the GOME-2 data. One explanation for the very small BrO VCDs measured by GOME-2 (compared to OMI) may be differences in chemistry due to different ambient conditions at the early overpass time of GOME-2 (see below). Furthermore, the weak BrO VCD enhancement in the GOME-2 data could at least be partly caused by radiative transfer effects due to the increased surface albedo of the salt marsh. However, 10 as the surface albedo shows a typical seasonal variation in contrast to the GOME-2 BrO VCDs (see Figure 4), this effects seems to be rather small. While radiative transfer simulations suggest a possible overestimation of the BrO VCDs by a factor of 1.5 (50%) due to an inadequate consideration of the bright surface (Section 3.3), this effect can be neglected for OMI, where peak BrO VCDs over the Rann are up to a factor of 16 larger than for areas outside the Rann at low surface albedo.
In principle, there are several instrumental and chemical reasons why the BrO VCDs over the Rann of Kutch derived from 15 GOME-2 are lower than those from OMI: 1. The spatial resolution of the GOME-2 instrument is 40×80 km 2 , while the near nadir ground pixel size of OMI is 13×24 km 2 . GOME-2 spectra are therefore generally expected to be less sensitive to localised emission sources, as larger ground pixels include a larger fraction of sunlight from areas outside the investigated object (the typical extent of 20 the salt marsh area showing clearly enhanced BrO VCDs is less than that of a GOME-2 ground pixel).
2. GOME-2 is known to suffer from instrumental degradation (especially in the UV wavelength region) starting in /2009(Dikty and Richter, 2012. A significant increase in the scatter of retrieved VCDs has been observed, particularly for BrO. This, in addition to the generally low signal-to-noise-ratio of GOME-2 in the UV compared to OMI (Fioletov et al., 2013), has led to a lower sensitivity of the instrument to small BrO concentrations.

25
3. One of the main chemical reasons for much lower BrO VCDs using GOME-2 might be that the measurements take place about 4 hours earlier when compared to OMI observations (≈ 9:30 vs. 13:30 LT). At the time of the morning overpass, the 'bromine explosion' mechanism has presumably not progressed very far, as solar irradiance is approximately 50% less than during OMI's afternoon overpass (according to ECMWF data) and the process is photolytically driven. Furthermore, O 3 is needed for the rapid build-up of BrO, which might be more easily available during the morning on the one hand, 4. The boundary layer height during the GOME-2 overflight in April/May is significantly lower (≈2km) than for the OMI measurements (≈3km). As the BLH increases towards noon, BrO originating from the ground might be transported to higher altitudes where it could be more easily detected by the OMI as the instrument's sensitivity generally increases for elevated layers. Additionally, the increasing BLH might lead to an increased mixing-in of tropospheric O 3 from higher altitudes and thereby lead to further formation of BrO. 5 5. Other meteorological parameters (like relative humidity) are expected to influence the efficiency of the 'bromine explosion' (Buxmann et al., 2012). While surface temperature (and associated parameters like e.g. boundary layer height) are lower during the GOME-2 overflight, relative humidity is about 30% higher. However, the detailed role of meteorological conditions on the 'bromine explosion' mechanism remain unclear (see Section 4.3).
Although all of these effects probably contribute to the observed differences between OMI and GOME-2 observations, the 10 reasons for this discrepancies are still a matter of further research.

The Dead Sea (Israel/Jordan)
OMI data were additionally analysed over another salt lake, the Dead Sea, but the results will only be shortly discussed in the following. A more detailed analysis of the data (or for further locations) exceeds the scope of this paper.
As already mentioned in Section 1.1, BrO over the Dead Sea has been frequently observed by ground-based DOAS measure-15 ments during recent years. While the results of Hebestreit et al. (1999) suggested that the salt pans over the southern part of the Dead Sea are the main source of the observed BrO, Matveev et al. (2001) and Tas et al. (2005) concluded that BrO is produced over all parts of the sea, although the frequency and intensity of BrO production seemed to be more intense over the southern basin. Figure 10 shows the mean BrO VCDs for summer months with relatively low cloud coverage (April-October) for the complete data set from 2005-2014, including an additional cloud filter of <30%. As it can be seen from the map, the mean 20 BrO VCDs are clearly enhanced over the southern part of the Dead Sea, while only weakly enhanced VCDs can be identified over the northern part. The results seem to confirm the ground-based findings, although the maximum BrO VCDs are shifted towards the southwestern inland, probably because of northeasterly winds that predominate in this area (Matveev et al., 2001).
While regions and is therefore supposed to have an significant impact on local and regional ozone chemistry. A first attempt to describe the annual BrO cycle based on a simple linear parametrisation of different meteorological quantities indicates that, in addition to UV irradiation, the variation of the boundary layer height is an essential parameter needed to describe the annual BrO peak, which can be at least partly explained by the higher sensitivity of the satellite for elevated layers.

25
For an adequate description of the near-surface BrO variations, relative humidity and precipitation also play an important role.
The seasonal wind conditions can be additionally linked to the observed seasonal variation of the BrO VCDs: The maximum VCDs in March-May are accompanied by westerly winds, which may lead to an increased swirling up of aerosol particles due to the east-west expansion of the Rann and the thereby increased residence time of air masses over the salt crust. In contrast, low VCDs during wintertime are affected by slow winds from the northeast.
rological conditions at the time of the satellite overflight (e.g. the UV radiation at the surface is about 50% lower), and that the 'bromine explosion' still needs to evolve during the morning hours in order to allow BrO concentrations to build up to values that may be detected from space.
Additional OMI measurements of enhanced BrO VCDs over the Dead Sea demonstrate the potential of satellite instruments for the global observation of reactive halogen species over salt lakes. By the improved temporal and in particular spatial resolution the other hand strongly suggest to undertake a ground-based measurement field campaign in that area to better constrain the general release mechanisms of reactive bromine compounds from salt marshes and lakes.
Level 1 data and ECMWF for providing meteorological parameters over India/Pakistan.    Table 2. Best fits from the multilinear regression analysis in order to model the seasonal BrO mass variation over the Rann of Kutch in dependency of UV surface radiation (UV), precipitation (P), cloud coverage (CC), relative humidity (RH), boundary layer height (BLH) and surface temperature (T). # linear function r 2 1 m BrO = a 1 · UV 0.71 2 m BrO = a 1 · UV + a 2 · P 0.83 3 m BrO = a 1 · UV + a 2 · P + a 3 · CC + a 4 · RH 0.91