Aura OMI observations of regional SO2 and NO2 pollution changes from 2005 to 2015

. The Ozone Monitoring Instrument (OMI)


Introduction
Sulfur dioxide (SO 2 ) and nitrogen dioxide (NO 2 ) are reactive, short-lived atmospheric trace gases with both anthropogenic and natural sources. Major sources of NO x (NO x = NO + NO 2 ) include fossil fuel combustion, biomass burning, soil emissions (Vinken et al., 2014b), and lightning (Schumann and Huntrieser, 2007). NO 2 participates in the nitrogen cascade of air, water, and soil (EPA, 2011;Galloway et al., 2013), affects atmospheric oxidation rates (Valin et al., 2013), and contributes to surface ozone production (Duncan et al., 2010;Seinfeld and Pandis, 2006). The principal sources of SO 2 are volcanic and anthropogenic emissions from burning sulfur-contaminated fossil fuels and the refinement of sulfide ores. Volcanic SO 2 is often injected into the atmosphere at high altitudes above the planetary boundary layer (PBL), while anthropogenic SO 2 emissions are predominantly in or slightly above the PBL. Chemical reactions in the PBL involving SO 2 and NO 2 lead to the production of sulfate and nitrate aerosols, and tropospheric ozone (Seinfeld and Pandis, 2006). Volatile organic compounds (VOCs) oxidize in the presence of NO x and sunlight to form ozone (O 3 ), a major tropospheric pollutant and greenhouse gas (EPA, 2013), and the oxidation product of NO 2 , nitric acid (HNO 3 ), reacts with ammonia (NH 3 ) to form ammonium nitrate aerosols. SO 2 is oxidized in gas-phase reactions with the hydroxyl radical (OH) or in aqueous-phase reactions with O 3 or hydrogen peroxide (H 2 O 2 ) to form sulfate aerosols. Sulfate and nitrate aerosols contribute to fine particulate matter pollution with aerodynamic diameters less than 2.5 µm (PM 2.5 ). PM 2.5 poses serious health concerns Liu et al., 2015), degrades visibility, causes acidification of water and the biosphere with adverse effects on plants and soil, and impacts weather and climate through direct radiative forcing and indirectly modifying cloud formation and optical properties (IPCC Working Group 1 et al., 2013;Twohy, 2005). SO 2 , NO 2 , and their oxidation products, O 3 and PM 2.5 , are designated "criteria pollutants" (European Commission, 2015;US EPA, 2016). Space-based characterization of these pollutants enables global, consistent monitoring, which is independent from ground-based measuring networks.
The first space-based quantitative data on SO 2 mass in volcanic clouds after major eruptions of the El Chichón volcano in March-April 1982 were obtained from NASA's Nimbus-7 Total Ozone Mapping Spectrometer (TOMS) (Krueger, 1983). The TOMS SO 2 detection sensitivity was limited by the instrument's six narrow wavelength bands. In practice, only exceptionally strong anthropogenic SO 2 signals could be detected, such as those produced by Norilsk smelting plants in Russia or from an accidental combustion of elemental sulfur (S) at the Al-Mishraq state sulfur mine plant in Iraq (Carn et al., 2004;US Department of Veterans Affairs, 2015). Greatly improved sensitivity was demonstrated through detection of SO 2 emissions from coal-fired power plants using ESA's Global Ozone Monitoring Experiment (GOME, 1995(GOME, -2005 Eisinger and Burrows, 1998) and SCanning Imaging Absorption spectrometer for Atmospheric CHartographY, (SCIAMACHY, 2002(SCIAMACHY, -2012 (Bovensmann et al., 1999) hyperspectral UV spectrometers. The first tropospheric NO 2 quantification was demonstrated using GOME and SCIAMACHY visible data (Leue et al., 2001;Martin et al., 2002;Richter and Burrows, 2002;Richter et al., 2005). These sensors needed several days to acquire a contiguous global map. The Ozone Monitoring Instrument (OMI) is the first satellite hyperspectral UV/Visible spectrometer with a push broom CCD detector and a 2600 km wide swath (Levelt et al., 2006b), enabling daily, global contiguous mapping of ozone and other trace gases, including SO 2 and NO 2 (Levelt et al., 2006a). OMI was launched in July 2004 on NASA's Aura sun-synchronous afternoon equator-crossing polar satellite (Schoeberl et al., 2006) and continues measurements through its 12th year, providing the longest data record currently available. NO 2 and SO 2 observations are also made by two GOME-2 instruments on EUMETSAT's MetOp-A (2006) and B (2012) operational polar satellites (Callies et al., 2000;Richter et al., 2011;Rix et al., 2012;Valks et al., 2011) and the Ozone Mapping and Profiler Suite (OMPS) onboard the NASA-NOAA Suomi NPP satellite (Dittman et al., 2002;Flynn et al., 2014;Seftor et al., 2014), which have coarser spatial resolutions and higher detection thresholds for emissions from point sources (Fioletov et al., 2013). ESA's next-generation Sentinel series will provide higher spatial resolution and greater sensitivity to SO 2 and NO 2 sources Veefkind et al., 2012).

OMI standard SO and NO products
OMI is the result of a partnership between NASA and the Dutch and Finnish meteorological institutes and space agencies (Levelt et al., 2006b) and flies on the NASA EOS-Aura satellite (Schoeberl et al., 2006). It measures sunlight backscattered from the Earth over a wide range of Ultraviolet (UV) and visible (Vis) wavelengths to derive abundances of ozone and other trace gases important for air quality and climate. The measurements of SO 2 and NO 2 are both explicit objectives of the Aura OMI mission (Levelt et al., 2006a) that are aimed at advancing our understanding of the sources and transformation processes of these pollutants and enabling the application of OMI data to inform public policy . Compared with other satellite UV-Vis instruments, OMI has the highest spatial resolution, least degradation and the longest record, allowing improved space-borne estimation of NO 2 and SO 2 emissions and the study of their temporal behavior (Carn et al., 2007;Castellanos and Boersma, 2012;Fioletov et al., 2011Fioletov et al., , 2013de Foy et al., 2009;Lamsal et al., 2015;Lu et al., 2013;McLinden et al., 2012;Zhou et al., 2012).
Aura has a local equator-crossing time of approximately 13:45 in the ascending node and provides nearly global coverage each day. The OMI detector is a 2-D charge-coupled device (CCD) array. The instrument optics are designed such that the spatial dimension of the detector is oriented across the orbit track, with an 115 • field of view, while the other dimension records spectral information. Three separate detectors (Dobber et al., 2006;Levelt et al., 2006b), designated UV-1, UV-2, and Vis, have spectral coverage (full performance) in the ranges of 270-310 nm (spectral resolution, full width at half maximum (FWHM), of 0.63 nm), 310-365 nm (0.45 nm), and 365-500 nm (0.63 nm), respectively. The OMI SO 2 product uses spectral measurements between 310.5 and 340 nm in the UV-2 (Li et al., 2013) and the NO 2 product uses spectral measurements between 405 and 465 nm in the Vis region Bucsela et al., 2013). The spatial dimension of both detectors is divided into 60 cross-track fields of view (FOV) corresponding to the specific binned CCD detector rows, such that rows 1 and 60 correspond to the western and eastern edges of the swath, respectively. Spectral measurements are made over 2-second exposure intervals. This results in along-track coverage of 13 km and cross-track coverage of 24 km for the near-nadir FOVs (CCD rows about 30). During each orbit, a total of about 1640 exposures are recorded on the sunlit side of the Earth. The width of the swath (2600 km) is such that 14-15 orbits per day are required to observe the entire surface of the Earth, although with increased FOV size at the swath edges. Beginning in 2007, some cross-track positions of the OMI swath were affected by FOV blockage and scattered light, also known as the "row anomaly" (KNMI, 2012). Here we use only unaffected OMI cross track FOVs throughout the entire mission, also excluding large FOVs at the edge of the swath, thus considering only the values for CCD rows 6-23.

Retrieval of PBL SO 2
The original OMI PBL SO 2 product employed the band residual difference (BRD) algorithm, which used only 4 discrete wavelengths (Krotkov et al., 2006). The BRD product is sensitive to the large SO 2 point sources, but has a high noise level  and systematic artifacts that required empirical corrections (Fioletov et al., 2011;Lee et al., 2009). In 2014, a new PBL SO 2 product was released, in which SO 2 is retrieved with a new algorithm that employs a principal component analysis (PCA) technique applied to OMI radiances (Li et al., 2013). Using a clear-sky air mass factor (AMF) similar to the previous SO 2 product, but with the full spectral content between 310.5 and 340 nm, the PCA algorithm reduces retrieval noise by a factor of 2 (Li et al., 2013). Recently, the Differential Optical Absorption Spectroscopy (DOAS) SO 2 algorithm developed for the Sentinel 5 Precursor (TROPOMI) has been applied to the OMI radiances and compared with the operational PCA product (Theys et al., 2015). The two products compare well, which lends confidence in the OMI SO 2 data. The estimated SO 2 noise is similar between PCA and DOAS algorithms, when using similar assumptions for AMF calculation for pollution SO 2 . However, the DOAS SO 2 algorithm requires empirical corrections to remove background bias.
In this study we will use the OMI operational PCA PBL SO 2 product, which contains the vertical column density (VCD) in Dobson units (1 DU = 2.69 × 10 16 molecules cm −2 ). The product (OMSO2 v1.2.0) is publicly available from the NASA Goddard Earth Sciences (GES) Data and Information Services Center (DISC) (http://disc.sci.gsfc.nasa.gov/Aura/data-holdings/ OMI/omso2_v003.shtml). For background areas the estimated 1σ noise is ∼ 0.5 DU over tropical oceanic areas (Li et al., 2013). If we assume that the noise is random and that there are about 100 cloud-free samples per year, the detection limit over low latitudes is estimated to be 4 times the mean error: ∼ 0.2 DU for the annual mean. For a single retrieval over polluted areas, random error due to instrument noise is typically on the order of 50-100 %. The systematic uncertainties due to our use of fixed Jacobians are 50-100 % for cloud-free scenes. The total error for a single retrieval is 70-150 %. For an annual average the uncertainties due to the retrieval noise are reduced to the level of 10-15 % of the actual signal, and become insignificant relative to the systematic errors. The systematic errors could be further reduced to the level of 20 % applying improved local Jacobians (McLinden et al., 2014). An important advantage of the PCA algorithm is that the bias over background regions (where SO 2 columns are below the OMI detection limit) is small enough (< 0.1 DU) to require no empirical background correction, as applied in other satellite SO 2 algorithms (e.g., Fioletov et al., 2013;Theys et al., 2015). The improved data quality, combined with the pixel averaging and oversampling techniques (e.g., de Foy et al., 2009;Fioletov et al., 2011Fioletov et al., , 2013Fioletov et al., , 2015Lu et al., 2013;McLinden et al., 2014McLinden et al., , 2016, provides greatly enhanced sensitivity to anthropogenic SO 2 sources near the surface McLinden et al., 2014). It has been demonstrated that US SO 2 point sources (e.g., power plants, smelters) with emissions rates as low as ∼ 30-40 kt yr −1 can be detected and analyzed using the PCA OMI SO 2 product (Fioletov et al., 2015). This limit is substantially lower than that reported (70 kt yr −1 ) for the previous version OMI SO 2 data (Fioletov et al., 2011).

Retrieval of tropospheric NO 2
There are two algorithms used operationally to determine tropospheric NO 2 VCDs: the NASA standard product (SP, version 2.1, http://avdc.gsfc.nasa.gov/pub/tmp/OMNO2D_HR/) (Bucsela et al., 2013;Lamsal et al., 2015)   . Both products share a common DOAS spectral fitting of the OMI-measured, sun-normalized backscattered radiances to laboratory-measured absorption spectra of NO 2 , H 2 O, and O 3 , and a calculated ring pseudo absorption spectrum (Chance and Spurr, 1997), to give NO 2 slant column densities (SCDs). The estimated 1σ noise is ∼ 10 15 molecules cm −2 or ∼ 10 % of the measured SCD over polluted regions . The SCDs, after subtraction of the stratospheric contribution are converted to tropospheric VCDs by applying AMFs interpolated from the look-up tables (LUTs) with OMI-measured input parameters, such as viewing geometry, climatological surface reflectivity, cloud pressure and cloud radiance fraction, assuming a priori NO 2 vertical profile shapes. The NASA and KNMI algorithms differ in how they remove the stratospheric contribution and use different a priori tropospheric NO 2 profile shapes in the AMF calculation. DOMINO subtracts stratospheric SCD as determined in a data assimilation system, in which the measured SCDs are assimilated with the TM4 chemical transport model . The SP estimates stratospheric NO 2 from OMI data without using stratospheric chemical transport models directly. The AMFs are calculated with a priori NO 2 monthly mean vertical profile shapes from the Global Modeling Initiative (GMI) model (Bucsela et al., 2013). Despite the differences, both algorithms produce statistically similar regional trends (see Supplement Fig. S1). Here we use the SP tropospheric NO 2 VCD product version 2.1 publicly available from NASA GES DISC at http://disc.sci.gsfc.nasa.gov/Aura/ data-holdings/OMI/omno2_v003.shtml. Over polluted areas the total errors in OMI tropospheric NO 2 VCDs are typically less than 20 % for cloud-free FOVs, as confirmed by validation studies employing in situ and remotely sensed data (Bucsela et al., 2013;Irie et al., 2012;Lamsal et al., 2015;Oetjen et al., 2013).

Postprocessing of NO 2 and SO 2 data
For this study, level 2 (L2) tropospheric NO 2 and PBL SO 2 VCDs are gridded at different ground resolutions after excluding FOVs possibly affected by the (1) row anomaly; (2) snow; (3) transient volcanic SO 2 clouds (Appendix A); (4) cloudy scenes with cloud radiance fraction, CRF > 0.5 for NO 2 or CRF > 0.2 for SO 2 . We note that the CRF is approximately twice as large as the effective cloud fraction derived assuming a mixed Lambert-equivalent reflectivity (MLER) cloud model Bucsela et al., 2013;Stammes et al., 2008). Given the very small CRF thresholds, the remaining cloud related errors were estimated to be less than 20 % (Lee et al., 2009;McLinden et al., 2014). However, by selecting mostly clear-sky conditions, our sampling of the OMI data may introduce a bias relative to all-sky conditions (Geddes et al., 2012;McLinden et al., 2014). Clouds are also associated with certain weather conditions, which in turn may affect the level of pollution. These factors may introduce biases in our derived trends in SO 2 and/or NO 2 , but only if there is a significant, long-term shift in weather regimes. However, for polluted regions in Fig. 1 satellite derived regional trends in cloud reflectivity (less than ±2 % per decade; Herman et al., 2013) are much smaller than those caused by changes in emissions (see Sect. 3). The standard gridded (0.25 • × 0.25 • ) level 3 (L3), filtered, monthly regional mean values are used in time series analyses following Lamsal et al. (2015) (Appendix B). The L3 data are publicly available from NASA GES DISC archive at http://disc.sci.gsfc.nasa.gov/Aura/data-holdings/OMI. We also use L2 (pixel level) data oversampled at higher resolutions (0.01 • × 0.01 • for NO 2 and 0.02 • × 0.02 • for SO 2 ) to create global and regional maps that highlight point pollution sources. The regional maps are created directly from pre-filtered L2 data by averaging all OMI pixels within a 20 km smoothing radius (30 km for SO 2 ) for 3 year time periods. Unlike previous studies (Lee et al., 2009;Fioletov et al., 2011Fioletov et al., , 2013Lu et al., 2013;McLinden et al., 2014), no empirical background correction was applied to the PBL SO 2 data. Figure 1 shows SO 2 and NO 2 multi-year average maps at the beginning of the OMI mission (2005)(2006)(2007) over the northern hemisphere. Regionally, population density , type of power generation and fuel used, economic activity, and regulatory policies determine average levels of air pollution. The SO 2 map ( Fig. 1a) shows hotspots associated with major coal-fired power plants and industrial activities, such as oil and gas refining and metal smelting. The highest SO 2 is found over industrialized and populated regions in eastern China, as the world's second-largest economy relies on sulfur (S)-rich coal for ∼ 70 % of its energy consumption (Klimont et al., 2009;Zhang and Cheng, 2009;Wang et al., 2015). Based on bottom-up emission inventories, SO 2 emissions from China were the world's largest, at ∼ 33 Tg SO 2 in 2005 (Lu et al., 2010(Lu et al., , 2011. High S coal-fired power plants are the major contributors to the SO 2 over the eastern US (SO 2 emissions 14.5 Tg SO 2 in 2005, US EPA, 2015), eastern Europe and India (∼ 6.7 Tg SO 2 , Lu et al., 2011). SO 2 is undetectable over the western US and western Europe, where emissions of SO 2 have been relatively small due to a smaller proportion of coal-fired power plants, the low S content of coal, and installation of effective flue gas de-sulfurization devices (FDG) capable of capturing more than 95 % of SO 2 emissions (US EIA, 2010).

Regional pollution changes and interpretation
Large SO 2 column amounts are also observed over the Persian Gulf, due to emissions from the oil and gas industry, gas flaring and shipping in the region. Based on a bottom-up SO 2 emission inventory, the total SO 2 emissions from the Middle East in 2005 were ∼ 6 Tg SO 2 (Smith et al., 2011), less than those from India and the US. However, OMI-observed SO 2 columns over the Persian Gulf region are significantly larger than over these two regions. This implies that real SO 2 emissions from the Middle East (particularly in the Persian Gulf) are significantly underestimated in current bottom-up emission estimates.
In addition to anthropogenic SO 2 , volcanic SO 2 is frequently observed over Kamchatka (Russian Federation), Japan, the South Pacific (e.g., Anatahan volcano, Mariana Islands, Mauna Loa, Hawaii), Sicily (Etna), Mexico (Popocatepetl volcano, south of Mexico City), Central America, and Montserrat, West Indies. Although transient volcanic signals were filtered from the PBL SO 2 data (Table A1), the signals from frequently erupting (e.g., Mt. Etna, Popocatepetl) or degassing volcanos remain. Except for Mt. Etna, Iceland volcanoes (Ialongo et al., 2015;Schmidt et al., 2015), and Mt. Popocatepetl (de Foy et al., 2009), most volcanic sources are located in remote locations and do not contribute to the SO 2 in industrial regions considered here (see OMI daily SO 2 maps for the world's volcanic regions at http://so2.gsfc.nasa.gov).
The average OMI NO 2 map (Fig. 1b) is correlated with the nighttime lights map (Fig. 1c), used here as a proxy for population density and energy production . For example, the highest NO 2 levels are observed over the world's most populated and industrialized regions, including eastern China, western Europe, and the eastern US, where local NO 2 "hot spots" coincide with large urban agglomerations (Schneider et al., 2015), power plants de Foy et al., 2015), and industrial complexes. NO 2 tropospheric columns over India and the Middle East are significantly less than those over China, western Europe, and the US. This can be explained by low NO x emissions, especially from mobile sources, and, partly, by year-round high temperatures, leading to shorter NO 2 lifetimes (Beirle et al., 2011). For example, Indian NO x emissions were relatively low, at 5.7 Tg NO x in 2005 (Lu and Streets, 2012), whereas those from China and the US were 16.9 Tg NO x (Klimont et al., 2009) and 20.4 Tg NO x (US EPA, 2015), respectively. Relatively small, but significant, areal NO 2 enhancements over west African forest are caused by seasonal biomass burning NO x emissions (Mebust and Cohen, 2014).
The differences between the spatial distributions of NO 2 and SO 2 over the large regions indicated as boxes in Fig. 1a and b are related to economic activity, fuel types, combustion technology, and different regulatory policies. The most abundant source of SO 2 is pyrite (FeS 2 ) and organic S in lowergrade coal as well as liquid fuel, mostly contained in heterocyclic aromatic compounds in oil, which largely accounts for high SO 2 levels over the Persian Gulf from gas flaring and oil refining. Many developed countries have regulated the S content of fuels and also required catalytic exhaust gas processing, resulting in decreased mobile-source NO x and SO 2 emissions in exhaust. Regulations are also focused on stack emissions of NO x and SO x (SO x = SO 2 + SO 3 ) at point sources, such as power plants and smelters. This, in turn, has driven technological changes upstream to meet regulatory requirements. For example, fluidized-bed combustion technology permits burning at lower temperature, producing less NO x , and condensed phase chemical capture of S, producing less gaseous SO x . Chemical loop combustion technology uses catalytic oxygenation to oxidize the fuel largely in the absence of N 2 , again resulting in greatly reduced NO x leaving the combustion chamber. Stack scrubbers (i.e., flue gas de-sulfurization devices, FDG) have been widely deployed in Europe and the US, in particular, for existing plants, to remove SO 2 and other chemicals -notably mercury -from the flue gases, in order to meet regulatory standards. However, these changes have yet to be widely implemented in developing countries.
In addition to emissions, meteorology also plays an important role in regional air pollution, particularly on relatively short time scales (days to months). For midlatitude areas discussed in this study (the eastern US, eastern China, and eastern Europe), the concentrations of SO 2 and NO 2 often exhibit large day-to-day changes. They tend to increase under the relatively stagnant conditions ahead of a cold front and decrease dramatically after the cold front brings precipitation and strong winds into the area . On the interannual time scale, the frequency of cold front passages may be influenced by large-scale circulation patterns such as the position of the Siberian high for eastern China (Jia et al., 2015), leading to interannual changes in SO 2 and NO 2 . But meteorology probably plays a lesser role in the longer-term trends that we discuss in this study. For example, given the general trend of weakening surface winds in the northern hemisphere (Vautard et al., 2010), one would expect both SO 2 and NO 2 to increase over time in China, with constant emissions. While OMI did initially observe growths in both SO 2 and NO 2 over China (Sect. 3.3), the different trends between the two gases after 2007 imply that different emission control measures may play a more significant role in OMI-observed trends. Similarly, the decreasing pollution levels observed over the eastern US (Sect. 3.1) and eastern Europe (Sect. 3.2) can only be explained by a reduction in emissions. As for tropical areas such as India, the impact of year-to-year fluctuations in OMI SO 2 and NO 2 data caused by meteorological variations is small relative to the observed fast growth in emissions that occurred over areas with newly built power plants and many cities (Sect. 3.4).
Another factor that can potentially affect derived longterm trends is long-term changes in the vertical profile shape, because our a priori profiles are constant for the entire mission. We believe that the impacts are relatively minor for OMI measurements, as the boundary layer is often thick and quite well mixed during OMI overpass time (in local afternoon). Our previous aircraft measurements over northeastern China and the eastern US show that the difference in AMF due to different SO 2 profile shapes over the two regions are Atmos. Chem. Phys., 16, 4605-4629, 2016 www.atmos-chem-phys.net/16/4605/2016/ very small (within a few percent, see Krotkov et al. (2008) for more detailed discussion). With this understanding of the influence of different factors on anthropogenic NO 2 and SO 2 columns, we turn, in the remainder of this section, to examining regional decadal trends as seen by OMI measurements. We examine five regions indicated in Fig. 1: the eastern US, eastern Europe and Turkey, eastern China, India, and the Middle East, which all have SO 2 and NO 2 sources detectable by OMI. The regions are in different phases of economic development and environmental regulations. We can therefore compare and contrast the trends in SO 2 and NO 2 that have different sources depending on the types of fuels burned, industrial activity, and regulations.

Eastern US
Over the eastern US the highest levels of SO 2 were observed in areas of intense high-S coal combustion for industrial processes and electricity generation, including the Ohio River valley and SW Pennsylvania (ORV, blue box in Fig. 2). Concentrations are undetectable over the western US where the local coal is intrinsically lower in S and emissions of SO 2 have been relatively small (US EIA, 2010). Prior investigations involving OMI have reported a 40 % SO 2 reduction near power plants in the eastern US between 2005 and 2010 (Fioletov et al., 2011). More recent OMI observations (Fig. 2) show that the SO 2 levels continued to drop after 2010 due to both national (e.g., Clean Air Interstate Rule, CAIR, CAIR, 2009) and state regulations, such as 2005 Maryland Healthy Air Act (HAA) (He et al., 2016). Currently, US regional SO 2 levels are at or below the OMI SO 2 detection limit of ∼ 0.2 DU. The dramatic decrease over the course of the first 11 years of the OMI mission (Fig. 2) closely matches trends in reported SO 2 emissions (US EPA, 2015) and sulfate deposition (−5 % yr −1 decrease over the eastern US from -2010, Hand et al., 2012Solomon et al., 2014) and has also been observed from surface and aircraft measurements (He et al., 2016). This striking improvement in SO 2 coincides with implementation of control technology, such as flue gas de-sulfurization (FGD), closure of some of the oldest coal power plants and fuel switching from coal to natural gas. Reductions in SO 2 emissions are required by the 1990 Clean Air Act Amendments (CAAA, 1990) and other regulations. Substantial success has been achieved through market-based cap and trade programs such as the Acid Rain Program (ARP, 2010) and The Clean Air Interstate Rule (CAIR, 2009). These allow electricity producers to pick the most economical emission control methods. The conversion to natural gas with much less fuel S than coal has also contributed to the reduction in SO 2 pollution.
Unlike SO 2 , which originates primarily from fuel-bound S, all high-temperature combustion, including internal combustion engines, can generate NO x . As expected, OMI NO 2 columns peak over major cities and highways, as well as over clusters of power plants. Chicago, Atlanta, and the megalopolis from Washington, DC to New York, also called the I-95 corridor (red box in Fig. 2), stand out. At the beginning of the OMI mission in 2005, a broad background of elevated NO 2 was detected over rural areas of the eastern US underlying the hot spots over large metropolitan areas (Fig. 2). Since that time, NO 2 has significantly decreased as a result of emission regulations on power plants and cars Lamsal et al., 2015;Lu et al., 2015;Russell et al., 2012). Decreases in NO 2 are evident in OMI NO 2 data over all major cities Tong et al., 2015), especially over the I-95 corridor (red box in Fig. 2 and Supplement Fig. S1). NO 2 from clusters of power plants has also decreased (e.g., ORV, blue box in Fig. 2). In general, downward trends in OMI NO 2 data near US power plants correlate well with trends in NO x emissions from the Continuous Emissions Monitoring System (CEMS)  and with surface NO 2 concentrations reported by EPA Air Quality Systems (AQS) Lu et al., 2015;Tong et al., 2015). The NO 2 reductions are due to selective catalytic reduction (SCR) on point sources and threeway catalytic converters on vehicles (Russell et al., 2012). Figure 3 (upper row) compares year-to-year changes in the OMI SO 2 and NO 2 annual columns and bottom-up emissions from power plants over the ORV region (blue box in Fig. 2) with other heavily polluted regions discussed later. Overall, between 2005 and 2015 the SO 2 drop over ORV was close to 80 %, while NO 2 dropped by 40 %, the largest reductions seen in this study. Previous studies demonstrate a linear ∼ 1:1 relationship between the percent change in NO x or SO 2 emissions from isolated power plants and the corresponding changes in OMI columns (Fioletov et al., 2011(Fioletov et al., , 2015de Foy et al., 2015). However,  show that most power plants, such as in the eastern US, are co-located with mobile NO x sources, so that this relationship is not always obvious. Indeed, OMI observed smaller drop in NO 2 columns (∼ 40 %) than would have been expected from ∼ 60 % reduction in NO x emissions from the power plants in the region (Fig. 3).
The magnitude of the relative reduction in NO 2 over the I-95 corridor is similar to that over the ORV (Supplement Fig. S1), suggesting similar reduction in NO x emissions from cities and mobile sources. An independent analysis of OMI NO 2 data confirmed that NO x emissions of 35 major US urban areas decreased by ∼ 50 % from 2006 to 2013 (Lu et al., 2015). We also note the faster decline in NO 2 levels before 2009 because of the installation of NO x emission control devices (ECDs) on power plants and the impact of the economic recession in 2007-2009. Power plants that were already operating ECDs during the ozone season began operating them year-round . The annual reduction rate in NO 2 has slowed since 2009 as the US economy has recovered from the recession and the implementation of further pollution controls has slowed. Although both SO 2 and NO 2 are criteria pollutants, and there remain jurisdictions in the US in violation of the National Ambient Air Quality Standards (NAAQS) for these primary pollutants, just as important is their role as precursors of key secondary air pollutants such as fine particles (PM 2.5 ) and ozone. The greatest numbers of Americans at risk for harmful effects of air pollution are subject to exposure to these secondary pollutants . By 2015, total US SO 2 emissions fell to about 1/6 of their 1970 peak, but NO x emissions only fell substantially after 2000 and are now about 1/2 of their peak in 2000 (https://www.epa.gov/air-emissions-inventories/ air-pollutant-emissions-trends-data). Because of these NO x reductions, photochemical smog over the eastern US has improved significantly over the same time period (Castellanos et al., 2011;Hogrefe et al., 2011;Simon et al., 2015). The total deposition of oxidized N (the combination of wet and dry deposition of species such as NO 2 and NO − 3 ) has improved as well (Nowlan et al., 2014) indicating that the efforts to control NO x emissions have been successful. As a result of larger SO 2 reductions, the SO 2 / NO 2 column ratio dropped over the ORV region from its maximal values of ∼ 4-5 in 2005 to less than 2 in 2012 (Supplement Fig. S2). We expect a similar change in PM speciation with increasing relative contribution of nitrate aerosols.

Eastern Europe
Europe experienced an ∼ 80 % reduction in SO 2 emissions between 1990 and 2011 (EEA, 2013). Particularly, in western Europe, after significant reduction of SO 2 emissions in the 1980s-1990s, the SO 2 levels have dropped below the OMI detection limit of ∼ 0.2 DU. There are, however, detectable SO 2 sources in eastern Europe (Fig. 4). The spatial distribution of the observed SO 2 columns at the beginning of OMI mission is consistent with the spatial pattern of SO 2 concentrations derived from the surface monitoring stations for 2005 (Denby et al., 2010). Notable anthropogenic SO 2 sources include, for example, the mining and industrial districts in Donbass region in eastern Ukraine, large coalfired thermal power plants around the Sea of Marmara and those near Kahramanmaras in southern Turkey, as well as those near Galabovo in Bulgaria, Gorj County in southwestern Romania, Belgrade in Serbia, and Megalopolis in south- Atmos. Chem. Phys., 16, 4605-4629, 2016 www.atmos-chem-phys.net/16/4605/2016/ ern Greece . Most SO 2 hot spots are due to use of local high-S lignite (brown) coal for power generation and incomplete SO 2 removal from the flue gas. Figure 3 (second row) shows interannual variations in SO 2 and NO 2 columns over the Maritsa Iztok power complex in Stara Zagora, Bulgaria (see blue box in Fig. 4) Figure 4 (bottom row) gives the spatial distribution of OMI tropospheric NO 2 over eastern Europe, which shows enhanced columns in densely populated and industrial areas. By far the largest NO 2 was observed over Moscow, Russian Federation, confirmed by in situ measurements at different heights within PBL (Chubarova et al., 2009(Chubarova et al., , 2016Elansky et al., 2007;Gorchakov et al., 2011). In Moscow maximal surface concentrations exceed 100 ppb for NO 2 , but are less than 2 ppb for SO 2 (Elansky et al., 2007). OMI also observed large NO 2 over industrial regions near Katowice in south Poland, eastern Germany, and the northwestern Czech Republic. Elevated NO 2 columns are evident over large cities, such as Istanbul, Prague, Warsaw, Vienna, Rome, Athens, and Budapest. These enhancements correlate well with emissions source distribution (Janssens-Maenhout et al., 2015). While road traffic is in general the most important NO x source in Europe (EEA, 2013;Vestreng et al., 2009), in some eastern European countries the power sector is the major contributor (Zyrichidou et al., 2013). New construction and upgrades in capacity of older power plants, as well as emission control measures affect NO 2 columns (Castellanos and Boersma, 2012;Zhou et al., 2012). Several studies based on bottom-up emissions and satellite observations have reported substantial decreases in NO x emissions and NO 2 columns in most western European countries due to stricter emission regulations (Castellanos and Boersma, 2012;Curier et al., 2014;EEA, 2013;Lamsal et al., 2011;Schneider et al., 2015;Vestreng et al., 2009;Zhou et al., 2012). In contrast, changes in emissions are rather small in eastern Europe (Zyrichidou et al., 2013). An increase in NO x emissions is reported for those countries where implementation of the European Union (EU) air quality standards is less effective (AQ_Environment_EC, 2015;Vestreng et al., 2009). OMI measurements are consistent with previous studies, suggesting small or insignificant NO 2 column trends on a regional level. Changes appear to be country-specific and likely depend on the socioeconomic and political situation and legislative abatement measures of the country. The EU air quality standards hold for all EUcountries (including Poland, Hungary, Bulgaria, Croatia, the Baltic States, Slovenia, Slovakia), but not for Serbia, Russia, Ukraine, Belarus, and Turkey. Some countries have asked for a time extension to meet certain standards because several member states have particular difficulties achieving compliance with the criteria for PM and NO 2 .

Eastern China
The growth of the Chinese economy over the past two decades has been mainly driven by rapid industrialization and urbanization  and has been accompanied by large increases in both electricity generation (mainly coal-fired power plants) and the number of vehicles on Chinese roads. As evident in Figs. 1a and 5, China has the world's highest SO 2 emissions, particularly over the high-S coal-rich, heavily industrial areas in Hebei, Henan, and Shandong provinces in the North China Plain (NCP, blue box in Fig. 5), Inner Mongolia , the highly populated Sichuan Basin (SB, red box in Fig. 5), as well as the megacity clusters around Shanghai (the Yangtze River Delta, YRD -black box in Fig. 5) and Guangzhou-Hong Kong (the Pearl River Delta, PRD). Similarly, OMI retrievals also reveal much greater NO 2 over eastern China than other regions of the world (Fig. 1b), especially over NCP, YRD, and PRD (Fig. 5). The NO 2 levels are relatively low over SB, but higher over YRD and PRD. The SO 2 / NO 2 column ratios were 8-10 over SB, 3-5 over NCP and less than 2 over YRD and PRD in 2005 (Supplement Fig . S4). The ratios reflect to some extent the level of modernization in the regions. The PRD and YRD have relatively less coal-fired power plants but higher population and car density, therefore greater NO 2 relative to SO 2 .
The overall SO 2 loading, although still at a relatively high level, has decreased over the recent years (Fig. 5). This is more clearly shown in the SO 2 time series in Fig. 3e, which suggests that the SO 2 loading over the NCP peaked in 2007, and has since shown an overall decreasing trend despite relatively large year-to-year variations. The reduction in SO 2 during 2008-2010 may be attributed to both the economic recession and emission control measures before the 2008 Beijing Olympic Games Lu et al., 2011;Mijling et al., 2009;Witte et al., 2009). The temporary rebound in 2011 may reflect a resurgence in the economy due to stimulation by the government. This is followed by a dramatic   Fig. S3). In relative terms, the SO 2 reduction in 2005-2015 was larger over YRD and SB regions compared to NCP (Supplement Fig. S3). NO 2 over NCP, on the other hand, peaked in 2011 after dramatic ∼ 50 % increase since 2009 (Fig. 3) and decreased slightly in 2012 and 2013 (Fig. 3). Temporary drop in 2008 can be attributed to strict pollution reduction measures implemented before 2008 Olympic games and economic recession. The reductions were strongest in Beijing, Tianjin, and Shijiazhuang regions (Mijling et al., 2009;Witte et al., 2009). The dramatic ∼ 40 % drop in NO 2 in 2014-2015 is likely a result of the slowest economic growth rate for China in nearly 25 years. According to the National Bureau of Statistics, the electricity generation by thermal power plants decreased by several percent in the second half of 2014 as compared with 2013. Similarly there is also a slowdown in coal-intensive industrial sectors (Guay, 2015) and stricter emission control policies (MEP, 2013). Independent satellite NO 2 retrievals with GOME-2A, GOME-2B, and OMI also confirm a large reduction in NO 2 over eastern China between 2013 and 2014 (Richter et al., 2015). Over SB and YRD, NO 2 columns peaked in 2010 and remained relatively constant afterwards (Supplement Fig. S1). As a result of the different trends between SO 2 and NO 2 , the SO 2 to NO 2 ratios dropped to their lowest values of ∼ 2-3, ∼ 1-2 and less than 1 over SB, NCP, and YRD regions, respectively (Supplement Fig. S4). Figure 6 shows 3-year mean OMI SO 2 and NO 2 maps over India. A number of SO 2 and NO 2 hot spots are observed, and they match the locations of large coal-fired power plants and major cities (Ghude et al., 2011(Ghude et al., , 2013. This is because coal-fired power plants are the dominant SO 2 and NO x emission sources in India, and they are often built near large cities where other anthropogenic emissions are also high. Figure 6 also shows that from 2005 to 2015, there was an increase in the OMI-observed SO 2 and NO 2 columns over India, mainly reflecting the fast expansion of the power sector driven by rapid economic growth. Based on an updated unitbased coal-fired power sector database (Lu and Streets, 2012;Lu et al., 2013), the total installed capacity, power generation, and fuel consumption of Indian coal-fired power plants increased dramatically by 126, 91, and 93 %, respectively, during 2005-2014. The SO 2 emissions from power plants are high, because S in local coal is mostly in organic form and cannot be removed by physical cleaning methods (Lookman and Rubin, 1998).

India
Unlike the US, Europe, and China, SO 2 and NO x emitted from coal-fired power plants are not regulated in India and the installation and operation rates of SO 2 and NO x emis-sion control devices are very low. FGD devices for SO 2 were reported to be operating in only three power plants at the beginning of OMI mission (Chikkatur et al., 2007). NO x emissions by coal-fired power plants are also not regulated in India. Although some new plants were reported to be equipped with low-NO x burners (LNBs), the actual installation rate and performance of these LNB devices are not known. Based on bottom-up emission inventories, we estimate that SO 2 and NO x emissions from Indian coal-fired power plants increased by 103 and 94 %, respectively, during 2005-2014(Lu and Streets, 2012Lu et al., 2013).
As shown in Fig. 3, the growth rates in OMI-observed SO 2 (200 % ± 50 %) and NO 2 (50 % ± 20 %) columns during 2005-2015 were particularly large over the industrial regions in Chhattisgarh and Odisha (blue box in Fig. 6), one of India's most active areas in terms of building new power plants. By the end of 2014, the total installed capacity of coal-fired power plants in this region was 28 GW, 85 % of which (∼ 24 GW) was installed after 2005, accounting for ∼ 26 % of the total newly installed capacity in India. As a result, SO 2 and NO x emissions from coal-fired power plants in this region were both estimated to increase by ∼ 190 % from 2005 to 2014 (Lu and Streets, 2012;Lu et al., 2013), largely in line with OMI SO 2 observations (Fig. 3g). India's total annual SO 2 emissions almost doubled from 6.7 Tg in 2005 to estimated 12 Tg in 2014. In 2014, India has not only surpassed the US to be the world's second largest SO 2 emitting country, but also has reached more than 40 % of the SO 2 emissions of the world's largest emitter, China.
During the last decade OMI observed much smaller NO 2 increases (∼ 50 %) than one would have expected from the increase in NO x emissions from the coal-fired power plants (Fig. 3h). One possible explanation for the discrepancy might be relatively high NO 2 background from other NO x emission sources. While coal-fired power plants may be the single largest contributor to SO 2 in this region, transportation is a larger contributor to NO x , and the slower increase in transportation emissions could have masked the sharp increase in coal-fired power plants NO x emissions. In India, the prevalence of motorcycles with small, two-stroke engines lead to high transportation emission factors for CO, VOC, and PM, but produce only modest amounts of NO x (Dickerson et al., 2002). Also, with a 3-fold increase in NO x emissions from the power plants, there could be some non-linear effects in NO x chemistry, changing the lifetime of NO 2 . Heavy loadings of soot may also remove NO 2 (Dickerson et al., 2002). The discrepancies will be addressed in future studies.

Middle East
In the Middle East, abundant oil and gas deposits supply cheap and relatively clean fuels for electricity generation, water desalination, and industry. OMI detects the largest SO 2 emissions over the Persian Gulf. The sources for these emis- inventories, such as the EDGAR-HTAP data set (Janssens-Maenhout et al., 2015). Based on the most recent SO 2 emission inventory, the total SO 2 emissions from the Middle East in 2005 were ∼ 6 Tg (Klimont et al., 2013;Smith et al., 2011), less than those from India and the US. However, OMI observed SO 2 columns over the Gulf region are significantly larger than those over India and the US. That suggests that the real SO 2 emissions from the Middle East (particularly in the Persian Gulf) may be several times higher than current bottom-up emission estimates. This is consistent with independent OMI SO 2 retrievals (Theys et al., 2015). Inverse modeling using OMI and SCIAMACHY retrievals also suggests an underestimate of SO 2 emissions from the Persian Gulf (Lee et al., 2011).
In situ measurements of SO 2 and other pollutants are rarely reported for the region, but available data generally indicate significant SO 2 loading over the Persian Gulf. For example, an aircraft campaign conducted north of the United Arab Emirates during winter 2001 measured SO 2 concentrations of up to 40 ppb (see https://www.rap.ucar.edu/asr2002/ i-precip_physics/precip_physics.htm), greater than what has been previously observed over eastern China He et al., 2012). The largest hotspot observed by the aircraft, near Zirku Island, also appears to be co-located with a hotspot in OMI retrievals. In another study, passive sampling of SO 2 at various locations on Khark Island near the north end of the Gulf during 2003-2004 reported that the SO 2 loading was above the air quality standard (sometimes by several-fold) most of the time (Pourzamani et al., 2012). These high SO 2 columns over the Persian Gulf are likely the result of gas flaring activities from offshore oil and natural gas facilities, although shipping emissions and other sources may also contribute to them. Gas flaring is used on offshore oil rigs to dispose of gases such as hydrogen sulfide (H 2 S) for safety, operational, and economic reasons and can have significant impacts on the local and regional environment in the Middle East.
Middle East cities also show SO 2 emissions due to both mobile and stationary sources. Oil-burning boilers may constitute another important source of SO 2 in cities or population centers, as implied by the relatively high sulfate (∼ 10 µg m −3 ) that is closely associated with oil combustion tracers (e.g., vanadium), according to an aerosol source apportionment study for Kuwait City (Alolayan et al., 2013). The S content in gasoline and diesel is much higher in this region as compared with others such as Europe, which enforces stricter emission control measures (see http://www.unep.org/transport/pcfv/PDF/ JordanWrkshp-MiddleEastFuelQuality.pdf). Some of the largest point SO 2 sources in the region coincide with smelters or oil refineries, such as the Sarcheshmeh Copper Complex in Kerman Province, Iran, which is the largest copper smelter in the Middle East. Figure 3 (bottom row) shows interannual variations in observed SO 2 and NO 2 columns over the Persian Gulf (blue box in Fig. 7). Since 2010 SO 2 columns have steadily dropped by ∼ 20 % but increased again in 2014-2015 to 2005 levels. A recent study (Lelieveld et al., 2015) reported that OMI SO 2 over the Persian Gulf increased between 2005 and 2010 and then decreased between 2010 and 2014. Their results are based on retrievals using a different algorithm but are qualitatively consistent with this study.
OMI-retrieved regional NO 2 levels over the Middle East are much smaller than over China (Fig. 5) and the US (Fig. 2). This may also be the results of the short lifetime of NO 2 in this hot and photochemically active region (Beirle et al., 2011). NO x emissions in the region are associated with power generation and mobile sources. Local NO 2 enhancements coincide with heavily populated cities that have high car densities, such as Jerusalem (Israel) and Cairo (Egypt) , Tehran (Iran), Kuwait City (Kuwait), Dubai (UAE), Riyadh and Jeddah (Saudi Arabia). In terms of the regional trend over the Persian Gulf (blue box in Fig. 7), NO 2 columns increased by ∼ 20 % between 2005 and 2008 but remained approximately constant afterwards (Fig. 3). For major metropolitan areas in the region, Lelieveld et al. (2015) focused on the reversal of OMI NO 2 trends due to recent air quality regulations and domestic and international conflicts in the region. Their results are, for the most part, qualitatively consistent with Fig. 7. For example, their reported decrease of NO 2 over Damascus, Syria since 2011 (due to civil war) and increase over Baghdad, Iraq since 2007 are also visible in Fig. 7.

Conclusions
The first decade of OMI observations have yielded profound insights into the spatial distribution and temporal trends in SO 2 and NO 2 pollution around the world. For regions with detailed bottom-up emissions estimates or continuous emissions monitoring, OMI shows generally good agreement with these independent data sources. OMI-derived trends also agree well with those from available in situ measurements and deposition data. This adds confidence to the use of OMI to track locations, changes, and transport patterns of SO 2 and NO 2 over areas of the planet lacking local observations. In many regions pollution levels have changed dramatically reflecting underlying changes in SO 2 and NO x emissions (Fig. 8 5. Over the Middle East, OMI detected several SO 2 hot spots with a broad maximum over the Persian Gulf region. These hotspots are probably related to oil and gas operations but are mostly absent in bottom-up emission inventories, such as EDGAR. High concentrations of NO 2 are observed over major cities but less so over oil and gas operations. SO 2 shows no discernable trend over the Gulf while NO 2 rose from 2005 to 2008 and has since remained largely unchanged.
In summary, this study demonstrates that satellite remote sensing from advanced instruments such as OMI can provide long-term, nearly continuous global monitoring of SO 2 and NO 2 . Where in situ concentration measurements, emission inventories, and deposition monitoring are available, OMI provides complementary measurements to supplement and verify those other data sources. OMI can also find unreported or underreported major emissions such as over the Persian Gulf. OMI SO 2 and NO 2 data can also help to further our understanding of the production and the impact of secondary pollutants such as tropospheric ozone and particulate matter. Better understanding of these secondary pollutants will help refine satellite SO 2 and NO 2 retrievals.
Space-based monitoring plays an increasingly important role in the science of tropospheric chemistry and air quality applications to help mitigate anthropogenic and natural impacts on climate, sensitive ecosystems, and human health. It is essential to continue and maintain overlapping longterm satellite data records. The baseline established during the first 11 years of OMI is invaluable for the interpretation of measurements from future atmospheric chemistry satellite missions. The OMI NO 2 and SO 2 data sets used in this study will be refined and continued by the TROPOspheric Monitoring Instrument (TROPOMI) , which is planned for launch on ESA's Sentinel 5 Precursor (S5P) satellite in 2016. TROPOMI will have a significantly higher signal-to-noise and spatial resolution (7 km × 7 km at nadir) than OMI; both features are very important for monitoring point pollution sources and trends. S5P is part of the European Sentinel series that will continue the global pollution data record for another 20 years . The space-based capabilities for air quality applications will be further enhanced by the addition of higher-ground resolution hourly observations from the three geostationary satellites over North America (Tropospheric emissions: monitoring of pollution (TEMPO), http://tempo.si.edu) (Chance et al., 2013), over Europe (Sentinel 4 UVN, Ingmann et al., 2012) and eastern Asia (Geostationary Environment Monitoring Spectrometer (GEMS) onboard the GeoKOMPSAT satellite) (Kim, 2012). This constellation will allow for unprecedented observations of the key pollutants in the atmosphere.

Appendix A: Filtering transient volcanic clouds
Days affected by transient volcanic SO 2 signals were excluded as follows. Every day the region-wide 99.9 percentile of SO 2 VCDs was computed. If it was found to exceed a threshold value (Table A1) then all data from that day were excluded. This was found to perform better than a simple maximum SO 2 or NO 2 column cut-off as it tended to remove volcanic signals that, while elevated, would not exceed the maximum. A disadvantage of this method is that, while the volcanic contamination would generally only impact a small portion of the region, all data from that day were removed. The SO 2 threshold employed varied from 5 to 10 DU (Table A1) and was chosen by examining the 99.9-percentile time series during known periods of minimal volcanic activity. Different regions were found to be affected differently, a result of their proximity to significant eruptions. For regions that span the northern mid-latitudes such as US, Europe, and China, many of the days occurred in 2008 and 2009 and can be attributed to the eruptions of Kasatochi (Aleutian Islands, Alaska, August 2008, 52 • N) and Sarychev (Kuril Islands, eastern Russia, June 2009, 48 • N). By contrast, the Nabro eruption (northern Africa, June 2011, 13.37 • N) removed the most days over India and Africa, whereas the Middle East appeared to be largely unaffected by volcanic emissions.

Appendix B: Time series analysis
We use standard level 3 monthly regional mean SO 2 and NO 2 columns and a regression model discussed in Lamsal et al. (2015) to compare inter-annual time series for different regions. The time series of monthly average values ( ) are assumed to be comprised of three additive subcomponents: a seasonal component (α), a linear trend component (β), and a residual or noise (R) component: where t represents time (month). The time-dependent seasonal regression coefficient (α) is given by a constant plus intra-annual sine and cosine harmonic series (Randel and Cobb, 1994): where c 0 , c 1j , and c 2j are constant coefficients. The major portion of the annual cycle is explained by the seasonal variation of the NO x and SO 2 lifetime. Other factors, such as monthly variation in source strength, could also affect the annual variation of NO 2 and SO 2 columns, but these contributions, especially for NO 2 , to the seasonal cycle are typically smaller, especially for polluted areas. The seasonal pattern can evolve with time. We identify and extract seasonal and trend components by exploiting changes in the seasonal pattern (amplitude and phase) for individual years. For each year we fit a regression line using monthly observations from that year itself and six observations each from adjacent years. This provides a series of local regression lines that incorporate explicit time dependence. Comparison of local regression lines with high-and low-amplitude regression lines allows identification and isolation of two seasonal terms (α 1 , α 2 , where α = α 1 + α 2 in Eq. 1) and the linear trend (β) and residuals. Since we are interested in interannual changes, we do not explicitly derive the linear trend but rather calculate changes from 2005 from de-seasonalized NO 2 and SO 2 columns (Fig. 3).