columns over south-eastern Europe

Abstract. Satellite observations of nitrogen dioxide (NO2) tropospheric columns over south-eastern Europe are analyzed to study the characteristics of the spatial and temporal variability of pollution in the area. The interannual variability of the tropospheric NO2 columns is presented over urban, rural and industrial locations based on measurements from four satellite instruments, GOME/ERS-2, SCIAMACHY/Envisat, OMI/Aura and GOME-2/MetOp spanning a period of over twelve years. The consistency between the different datasets over the area is investigated. Two operational algorithms for the retrieval of tropospheric NO2 are considered, the one developed jointly by the Royal Netherlands Meteorological Institute and Belgian Institute for Space Astronomy and the one developed by the University of Bremen. The tropospheric NO2 columns for the area under study have been simulated for the period 1996–2001 with the Comprehensive Air Quality Model (CAMx) and are compared with GOME measurements. Over urban and industrial locations the mean tropospheric NO2 columns range between 3 and 7.0×1015 molecules/cm2, showing a seasonal variability with a peak to peak amplitude of about 6.0×1015 molecules/cm2, while the background values over rural sites are close to 1.1×1015 molecules/cm2. Differences in the overpass time and spatial resolution of the different satellites, as well as differences in the algorithms, introduce significant differences in the estimated columns however the correlation between the different estimates is higher than 0.8. It is found that the model simulations reveal similar spatial patterns as the GOME observations, a result which is consistent with both algorithms. Although the model simulations show a mean bias of −0.1×1015 molecules/cm2 under clean conditions, the modeled temporal correlation of 0.5 is poor in absence of biogenic and biomass burning emissions.


Introduction
Nitrogen dioxide plays a key role in tropospheric chemistry with important implications for air quality and climate change.Measurements of nitrogen dioxide (NO 2 ) are important for the understanding of tropospheric and stratospheric chemistry, particularly in relation to ozone production and loss (e.g.Crutzen et al., 1979;Murphy et al., 1993;Finlayson-Pitts and Pitts, 2000).On the one hand, tropospheric NO 2 is essential for maintaining the oxidizing capacity of the atmosphere.Photolysis of NO 2 during daytime is the major source of ozone (O 3 ) in the troposphere and photolysis of O 3 in turn initializes the production of the hydroxyl radical (OH), the main cleansing agent of the atmosphere (van Noije et al., 2006).On the other hand, NO 2 as well as O 3 are toxic to the biosphere and may cause respiratory problems for humans.Moreover, NO 2 may react with OH to form nitric acid (HNO 3 ), one of the main components of acid rain.As a greenhouse gas, NO 2 contributes significantly to radiative forcing over industrial regions, especially in urban areas (e.g.Solomon et al., 1999), due to its short lifetime, and hence has a local and not global effect.Although the direct contribution of tropospheric NO 2 Published by Copernicus Publications on behalf of the European Geosciences Union.
I. Zyrichidou et al.: Tropospheric NO 2 columns over south-eastern Europe to global warming is relatively small, emissions of nitrogen oxides (NOx≡NO+NO 2 ) affect the global climate indirectly by perturbing O 3 and methane (CH 4 ) concentrations.More details on the chemistry of tropospheric NO 2 are given, for example, by Seinfeld and Pandis (1998) and Finlayson-Pitts and Pitts (2000).The abundance of NO 2 in the troposphere is highly variable and influenced by both anthropogenic and natural emissions (e.g.Bradshaw et al., 2000).On a global scale, the main sources of nitrogen oxides are fossil fuel combustion, biomass burning, lightning and microbiological processes in soil (e.g. Lee et al., 1997).
In the past, NO 2 fluxes could be assessed by modeling, aircraft and ground-based measurements or using a combination of the above (Dickerson, 1984;Lelieveld et al., 1989;Lefohn and Shadwick, 1991).However, following the advances in satellite technology and the development of new instruments and algorithms the observation of NO 2 columns from space has become a reality.A global picture of the spatial distribution of tropospheric NO 2 is now obtainable since satellite measurements provide a global coverage in a very short time (between 1 and 6 days depending on instrument and cloud cover).
Tropospheric NO 2 columns retrieved from the Global Ozone Monitoring Experiment (GOME), the Scanning Imaging Absorption spectroMeter for Atmospheric Cartogra-pHY (SCIAMACHY) and the Ozone Monitoring Instrument (OMI) span more than ten years and have been used for air quality studies and satellite instrument validations (Richter et al., 2000(Richter et al., , 2005;;van der A et al., 2006, 2008).Most of these studies focus on Asia, as this part of the planet currently shows the largest variability and increasing trends in species relevant for air quality.As was shown in van der A et al. (2008) there is an interesting outflow of anthropogenic NO 2 over the oceans at the East coast of North America and China.A large positive trend is clearly visible in East China as has been reported in Richter et al. (2005) where it is shown that a strong increase in NOx emissions in China due to an increase in industry and traffic has been detected from 1996 to 2005 using a combination of GOME and SCIA-MACHY observations.A yearly growth was determined in terms of percentage with respect to the initial NO 2 concentrations from 1996 which was 10±4% over Beijing, 2.2±2% over Sao Paulo and 1.7±1% over Mexico City (van der A et al., 2006).There are also clear spots of increasing NO 2 in Mid-USA, South Africa, Delhi (India), Tehran (Iran) and surrounding areas, and several cities in mid-Russia.Furthermore, outflow of biomass burning NO 2 is visible west of Africa and Australia (van der A et al., 2008).
Many studies have also recently used tropospheric NO 2 satellite measurements in order to validate air quality models.In van Noije et al. (2006) different NO 2 retrievals have been inter-compared and also compared with results from 17 atmospheric chemistry models on a global scale.They found that on average the models underestimate the retrievals in industrial regions and overestimate the retrievals in regions dominated by biomass burning.In Uno et al. (2007) systematic analyses of inter-annual and seasonal variability of tropospheric NO 2 vertical column densities based on GOME satellite observations and the regional scale CTM CMAQ (Community Multi-scale Air Quality) were presented over Asia.CMAQ results underestimated GOME retrievals by factors of 2-4 over polluted industrial regions.
To date no clear assessment of the behavior of nitrogen dioxide over the Balkan Peninsula exists.Few studies have localized their results over this part of Europe.For instance, in Ladstätter-Weißenmayer et al. (2007) the synergistic use of GOME tropospheric column data (version 1, developed at the University of Bremen) with back-trajectory analysis and box model calculations enabled the detection of significant changes in pollutant tropospheric columns related to general air circulation patterns.It was found that when the Mediterranean is influenced by polluted air masses from Central Europe, the Balkans and the Black Sea, pollution leads to an increase in NO 2 .Furthermore, the observed mean NO 2 tropospheric column densities (in 10 15 molecules/cm 2 ) were determined to be: for Crete 1.1, for Athens 2.0, for Thessaloniki 2.3 and for Istanbul 2.4 for the month of May as a mean value of the years 1996 to 2002.A detailed analysis for Western Europe was presented by Blond et al. (2007), who compared tropospheric NO 2 from a vertically extended version (up to 200 hPa) of CHIMERE with high-resolution column observations from SCIAMACHY as retrieved by BIRA/KNMI.Konovalov et al. (2005) used GOME-based data products (version 2, developed at the University of Bremen), to evaluate the CHIMERE CTM over Western (10 • W, 18 • E, 35 • N and 60 • N) and Eastern (18 • E, 65 • E, 40 • N and 65 • N) Europe.Their study indicated much lower levels of NO 2 in Eastern Europe (which includes the Balkan Peninsula) compared to Western Europe and no clear evidence could be found that either the performance of CHIMERE or the quality of NO 2 columns derived from GOME measurements performs poorer for Eastern than for Western Europe.
The Eastern Mediterranean is a known cross road of air masses where anthropogenic pollution emissions converge with natural ones (e.g.Lelieveld et al., 2002;Mihalopoulos, 2007).The ground-based stations that exist in the region are managed by the various local authorities, such as municipalities, prefectures, and so on.As a result, such stations exist either only in the capitals of states and in largely populated cities, or near airports.As an example, according to the European Air Quality data base (Airbase) (http://www.eea.europa.eu/)for Greece only 24 ground stations (16 urban, 7 suburban and 1 rural) provide NO 2 measurements, most of which ( 14) located in the Greater Athens area and 4 in the Greater Thessaloniki (the second biggest Greek city).The same pattern is found for the rest of the Balkan states considered and therefore from the airquality/environmental ground-based network one cannot deduce the NO 2 variability over the Balkan Peninsula.Since the presently available ground-based stations do not have  a proper spatial distribution in south-eastern Europe (http: //www.eea.europa.eu/)and there are not many scientific studies that focus on NO 2 variability over the Balkan Peninsula, the present study aims at providing more details on the temporal and spatial distribution of tropospheric NO 2 columns over the area through satellite observations and model simulations and to examine the satellites' consistency over areas with moderate loading of tropospheric NO 2 .
In Sect. 2 we give a description of the instruments, the algorithms and the photochemical model we used in our analysis.In Sect. 3 we investigate the long-term variability of tropospheric NO 2 over several Balkan geolocations as derived from GOME, SCIAMACHY, OMI and GOME-2 retrievals in order to assess the ability of satellite sensors to detect pollution and to investigate if significant trends can be derived on a regional scale.In addition, we compare concurrent (on the same day and on the same geolocation) satellite measurements.This comparison is used to infer whether or not the satellite data are consistent over the region under study.In the last part of Sect. 3 we evaluate the CAMx model using satellite retrievals from both the KNMI/BIRA and the Institute of Environmental Physics (IUP), University of Bremen DOAS algorithms.Finally, in Sect. 4 we present the conclusions derived from this study.

Methodology and data
Thirty-two geolocations around the Balkan Peninsula were chosen as focal point for this study (Fig. 1) and are listed in Table 2.They were selected according to the following criteria: their spatial distribution around the region, polluted sites such as industrial and commercial centers or capitals, unpolluted sites that can provide background values and sites that may help the detection of potential transboundary transport of NO 2 inside the Balkan Peninsula.For all these locations overpass files for the tropospheric NO 2 columns were generated from level-2 data of GOME, SCIAMACHY, OMI and GOME-2.The extraction criteria and the main characteristics of the instruments and the algorithms are discussed in the following paragraphs of this section.

Instruments
The GOME instrument is a nadir-viewing spectrometer that measures upwelling radiance from the atmosphere and solar irradiance, covering the spectral range of 240 nm to 790 nm at a spectral resolution of 0.2-0.4nm.Global coverage is achieved within three days at the Equator and within one day at 65 • latitude.The GOME instrument principles are described by Burrows et al. (1999).
SCIAMACHY is a passive remote sensing spectrometer observing backscattered, reflected, transmitted and emitted radiation from the atmosphere and the Earth's surface, in the wavelength range between 240 nm and 2380 nm and with a spectral resolution of 0.25 nm in the UV and 0.4 nm in the visible.SCIAMACHY alternately makes limb and nadir measurements.Global coverage at the Equator is achieved in six days and more frequently at higher latitudes.The SCIA-MACHY measurement principles are described in Bovensmann et al. (1999).
The Dutch -Finnish OMI is the first of a new generation of space borne spectrometers that combine a high spatial resolution with daily global coverage because of the wide swath of the measurement.OMI is a nadir viewing imaging spectrograph measuring direct and backscattered sunlight in the ultraviolet -visible (UV/VIS) range from 270 nm to 500 nm with a spectral resolution of about 0.5 nm and is described in detail in Levelt et al. (2006).
The GOME-2, an improved version of ESA's GOME instrument, is a nadir -looking UV -visible spectrometer.GOME-2 covers the spectral range between 240 nm and 790 nm and has a spectral resolution between 0.25 nm and 0.5 nm and provides global coverage within 1.5 days.The GOME-2 instrument principles are described in Callies et al. (2000).
The main features of the four instruments, satellite platforms and data versions used in this study are summarized in Table 1 for quick reference.

GOME, SCIAMACHY, OMI and GOME-2 tropospheric NO 2 retrievals
The main data set used in this study are NO 2 vertical tropospheric column densities retrieved by KNMI (Royal Netherlands Meteorological Institute) and BIRA/IASB (Belgian Institute for Space Astronomy) which are publicly available on a day-by-day basis via ESA's TEMIS project (http: //www.temis.nl).In this study we also considered GOME NO 2 data from IUP Bremen which are available via http: //www.iup.uni-bremen.de.This paper however does not aim to provide a detailed intercomparison between the two algorithms and the two satellite data sets.The two algorithms are independently compared with model simulations for the period when GOME measurements are available.They are referred to as GOMEtemis for the KNMI/BIRA algorithm and GOMEbremen for the IUP Bremen algorithm.Also, OMI measurements are referred to as OMIT3 in plots.

KNMI/BIRA algorithm
The TEMIS NO 2 vertical tropospheric column for GOME, SCIAMACHY and GOME-2 tropospheric NO 2 columns are all products of the same retrieval algorithm.The retrieval of tropospheric NO 2 is performed in three steps: first the total slant NO 2 column density is retrieved by BIRA/IASB using a Differential Optical Absorption Spectroscopy (DOAS) technique (e.g.Platt, 1994), then the stratospheric contribution is deduced by assimilating the total slant column data in the TM4 chemistry model driven by meteorological analysis from the European Center for Medium-range Weather Forecasts (ECMWF) and subsequently the vertical tropospheric column is derived, applying a tropospheric air mass factor correction.More details on the retrieval can be found in Boersma et al. (2004) and in Blond et al. (2007).The KNMI/BIRA tropospheric NO 2 retrievals have been also validated in several studies (e.g.Schaub et al., 2006;Blond et al., 2007;Lambert et al., 2007).
The OMI retrievals were developed at KNMI within the DOMINO (Dutch OMI NO 2 ) project.The DOMINO product is available from www.temis.nl.The DOMINO retrieval algorithm is described elaborately in Boersma et al. (2007) and recent updates can be found in the DOMINO Product Specification Document (http://www.temis.nl/docs/OMINO2 HE5 1.0.2.pdf).The DOMINO tropospheric NO 2 columns have been validated versus independent measurements during various campaigns (Boersma et al., 2008a(Boersma et al., , 2009)).
For the purpose of our study we note that the cloud fraction for GOME, SCIAMACHY and GOME-2 is taken from the Fast Retrieval Scheme for Clouds from the oxygen A band (FRESCO) algorithm (Koelemeijer et al., 2001).OMI's cloud fraction is provided by a cloud retrieval algorithm that uses the absorption of the O 2 -O 2 collision complex near 477 nm (Acarreta et al., 2004).The FRESCO and O 2 -O 2 algorithms are based on the same set of assumptions, i.e. they both retrieve an effective cloud fraction (clouds are modeled as Lambertian reflectors) that holds for a cloud albedo of 0.8.(Boersma et al., 2007).The similarities and the significant differences between the cloud parameter retrievals from SCIAMACHY and OMI are described quite extensively in Boersma et al. (2007).In that paper, since temporal variation in global cloud fraction and cloud pressure between 10:00 and 13:45 local time is small (Bergman and Salby, 1996), an evaluation of the consistency between the two cloud parameters was made as well.The comparison of the FRESCO and O 2 -O 2 cloud algorithms showed that on average SCIA-MACHY cloud fractions are higher by 0.011 than OMI cloud fractions and OMI cloud pressures are about 60 hPa higher than FRESCO cloud pressures (for cloud fractions >0.05).
For our spatial and temporal variability analysis only observations with a radiance reflectance of less than 50% from clouds were used which corresponds to a cloud fraction of less than about 20% (van der A et al., 2008).In addition, only completely un-flagged retrievals were accepted.For GOME, SCIAMACHY and GOME2 we used only the measurements with flag=0 and for OMI those measurements with even flag values, that correspond to meaningful tropospheric retrievals.For all the satellite instruments the distance between the satellite's center field of view and the ground locus was set to 50 km, in order to obtain spatially comparable measurements.This distance was the minimum we could use in our analysis to get adequate or coincident measurements from each satellite for the intercomparisons.Finally, we used only the forward scans for GOME, SCIAMACHY, and GOME-2 and from OMI pixels with CTP (Cross Track Position) 10 to 50 that corresponds to the OMI pixels which are closest to the near-nadir viewing position.

BREMEN algorithm
As mentioned above, apart from the KNMI/BIRA product, in this paper we also use the GOME tropospheric NO 2 columns from the Institute of Environmental Physics of the University of Bremen.The GOME tropospheric NO 2 retrieval method is performed in a series of similar steps as those described above: the total NO 2 slant column is extracted using the DOAS method, the stratospheric contribution is subtracted and then, via an air mass factor calculation, the remaining tropospheric NO 2 slant column is converted to a geometry independent tropospheric NO 2 vertical column.The retrieval algorithm is described in details in Richter and Burrows (2002) and Richter et al. (2005).
The differences, and similarities, of the two algorithms are mentioned briefly below: 1.The DOAS method is applied for the 405-465 nm region for OMI and for the 420-450 nm region for SCIA-MACHY and GOME(-2) in the KNMI/BIRA algorithm and for the 425-450 nm region for GOME in the IUP/Bremen technique.
2. KNMI/BIRA has developed an assimilation approach in which the GOME slant columns force the stratospheric component of NO 2 of the Tracer Model Version 4 (TM4) to be consistent with the observations (Boersma et al., 2004).In IUP/Bremen stratospheric NO 2 fields from the SLIMCAT model are used (Chipperfield et al., 1999) (Horowitz et al., 2003).
7. The BIRA/KNMI algorithm does not explicitly account for aerosol effects, assuming that they are at least partially accounted for by the cloud retrieval algorithm, whereas the IUP/Bremen retrieval accounts for three different aerosol scenarios (maritime, rural, and urban) taken from the Low Resolution Transmission (LOW-TRAN) database.
Almost eight years of continuous tropospheric NO 2 measurements from GOME/ERS-2 from the IUP/Bremen algorithm are used in this paper.The same extraction criteria as for the KNMI/BIRA retrievals have been used.
Organic biogenic emissions were calculated with the use of the RegCM3-CAMx interface, which extracts meteorological parameters from RegCM3 (temperature and radiation) and uses the available land use categories to calculate emission potentials and foliar biomass densities (Guenther et al., 1993).Anthropogenic emissions were calculated with data from the UNECE/EMEP data base (http://webdab.emep.int/)for European emissions (Vestreng et al., 2005) for the year 2000.These data comprise the annual sums of the emissions of NOx, CO, non-methane hydrocarbons, SO 2 , NH3, fine particles (<2.5 µm) and coarse particles (2.5 µm to 10 µm) on a 50 km×50 km grid.Eleven sectors of anthropogenic activity are distinguished in accordance to SNAP97.For every sector different distributions for the month, the day of the week and the hour of the day were applied for the temporal disaggregation.The disaggregation factors are taken from the inventory by Winiwarter and Zueger (1996).Emissions from lightning and biomass burning activities were not considered in the model runs.The model run used in this study was performed in the frame of CECILIA EU project and at the moment the model is not yet optimized to include operationally these two mechanisms in the simulations.This is an ongoing effort.The boundary conditions were set to 1 ppb.
NO 2 tropospheric vertical column densities were extracted for the altitudes between 0-7 km and the time period from 01/01/1996 to 31/12/2001 for which the model run has been performed.We compared average 2-h NO 2 predictions from CAMx with the GOME measurements, using an appropriate number of CAMx grid-cells in a way that they fit in with GOME's pixel spatial resolution.

Temporal variability
For all the locations listed in Table 2 we compiled time series of the monthly mean tropospheric NO 2 columns for each satellite instrument and algorithm using the criteria mentioned in the previous section, in order to examine to what extend we can determine certain characteristics of the temporal variability of tropospheric NO 2 over urban, rural and industrial areas in the Balkan Peninsuala.In this paragraph we present time series of the monthly mean tropospheric NO 2 column densities, derived from the satellite measurements, for all the under study geolocations in order to investigate the mean levels and the seasonal evolution of the observations from the four satellites.These time series are shown in Fig. 2a and are grouped according in small, large and mega cities, industrial and rural sites.We focus our discussion on some of the geolocations which, as it can bee seen in Fig. 2a are representative of of urban, industrial, megacities and rural geolocations (Thessaloniki, Maritsa, Istanbul and Finokalia respectively).The satellite algorithm results shown in this figure and further on, unless stated otherwise, are the KNMI/BIRA algorithm retrievals.Figure 2a demonstrates that SCIAMACHY, GOME-2 and OMI can reveal the characteristics of urban scale and industrial regions due to their finer horizontal resolutions and are quite consistent, while GOME is representative of a much larger area which smoothes out the effect of these local sources.Over the polluted sites there is a relative offset between SCIAMACHY and OMI measurements as shown here, as typical examples, for Maritsa and Thessaloniki The estimated average values for Thessaloniki and Maritsa are 3.8±1.7×10 15molecules/cm 2 and 3.2±1.0×10 15molecules/cm 2 respectively as derived from OMI monthly mean measurements, while the corresponding values estimated from SCIA-MACHY are 4.0±2.36×10 15molecules/cm 2 and 2.7±0.9×10 15molecules/cm 2 .This discrepancy between SCIAMACHY and OMI can be possibly attributed to the different local crossing time of each satellite (around three hours difference) and to the fact that the local NO 2 diurnal variability may quite possibly be different for each location.This offset is discussed in more detail in Sect.3.2.Both over urban and industrial regions OMI and SCIAMACHY measurements show a seasonal variability with maximum values during the winter months and minimum values during summer.The peak-to-peak amplitude of this variability over urban and industrial regions is consistent, and ranges from 4.0 to 6.0×10 15 molecules/cm 2 .The collocation criteria used (see Sect. 2.2 and 2.3) do not provide a continuous time series for GOME due to its orbit and pixel characteristics.However indications for the seasonal variability, consistent with SCIAMACHY and OMI, are also observed in GOME data.Without extensively homogenizing the four different data sources and the subsequent loss of spatial resolution, an analysis which is beyond the scope of this paper, possible trends over the region in question cannot be extracted in a statistical significant way.Recent satellite studies based on combined gridded 1 • ×1 • GOME-SCIAMACHY tropospheric NO 2 data for the period 1996-2006 (van der A et al., 2008) show negative trends of the order of 7% per year for whole Europe.When focusing however over certain cities and areas they show spatial differences of the trends (e.g.London versus Cologne) which in most cases become insignificant.There is no trend information available for the area under study based on satellite data.However there are various publications and reports that study long homogeneous time series of surface observations of pollutants in the area, which show a decreasing tendency in the late 90s and early 00s and a leveling to lower values during the last years.A detailed review of such studies is beyond the scope of this paper.Even over rural areas such as Finokalia, the differences in overpass time and spatial resolution are sufficient to preclude a statistically significant result.In Fig. 2a the case of Istanbul is also presented as it is the most polluted region examined in this work.Over the largest city of Turkey, the highest values of tropospheric NO 2 were observed for the entire period considered and mean values of 7.9±3.2×10 15molecules/cm 2 and 6.9±2.0×10 15molecules/cm 2 as derived from OMI and SCIAMACHY observations respectively are found with a seasonal variability similar to the other urban sites examined.Onkal-Engin et al. ( 2004) have shown that these levels of pollution are mainly due to emissions of intense land transportation in Istanbul.
In a rural area like Finokalia, with no large sources of NO 2 in the surroundings, the mean tropospheric NO 2 amounts are close to 1.0×10 15 molecules/cm 2 and they show a small seasonal variability with amplitude around 0.5×10 15 molecules/cm 2 with higher values observed during summer.There are no significant discrepancies among the different satellite instruments.In addition when combining the time series from the different satellites (GOME-SCIAMACHY-OMI) there is no sign for a long-term change in tropospheric NO 2 columns over the rural areas studied here, indicating almost constant background conditions in the greater area.
In order to examine how representative are the satellite estimates to local surface emissions we compared them with all available in situ surface NO 2 observation in the area.Figure 2b presents a scatter plot of concurrent monthly mean satellite measurements (SCIAMACHY and OMI) versus ground based in situ measurements.The ground-based data are available through the European Air Quality data base (Airbase) (http://www.eea.europa.eu/)for the time period 2003-2008.The ground stations that provide NO 2 surface measurements and are included in our domain are only six (Thessaloniki, Athens, Aliartos, Bucurest, Skopje and Sarajevo).The correlation coefficient is R=0.6 that indicates that there is a relatively good agreement between the two kinds of measurements.
The discrepancies between the different satellite products shown in the analysis discussed in this section are investigated further in the following section.

Satellite intercomparison
In this section, different satellite products are inter-compared using SCIAMACHY measurements as reference, as they provide a concurrent dataset with the other three instruments.Figure 3 compares monthly mean tropospheric NO 2 derived from the four satellites for all geolocations at the same dates.The comparison between GOME and SCIAMACHY which have a common dataset of 12 months show very good agreement with a correlation coefficient of 0.9 mainly due to their similar overpass times.The correlation coefficient between OMI and SCIAMACHY is 0.86.The differences between    OMI and SCIAMACHY are mainly credited to different overpass times and can be attributed to a moderate diurnal cycle in emissions (see Fig. 4, left panel) in combination with a strong diurnal cycle in photochemistry with maximum NO 2 loss around noon.A larger scatter between the OMI and SCIAMACHY data (middle plot) should also be expected due to their different horizontal resolution.OMI is expected to detect higher NO 2 values from more localized sources, for e.g.industries, biomass burning, and soil emissions.SCIAMACHY and GOME-2 measurements (bottom plot) are also well correlated due to their similar overpass times and horizontal coverage and so their spatial distributions have a correlation coefficient r of 0.85.For tropospheric columns less than 4.0×10 15 molecules/cm 2 we conclude that SCIAMACHY is in good agreement with GOME, OMI and GOME-2.For tropospheric columns larger than 4.0×10 15 molecules/cm 2 there are more discrepancies between the satellite measurements.
In order to investigate the systematic differences between SCIAMACHY and OMI, we used the CAMx model predictions to examine if the 13:30 to 10:00 UT differences in tropospheric NO 2 columns observed by satellites in this region are consistent with the diurnal variations predicted by a photochemical model.The left plot of Fig. 4 depicts the mean diurnal variation of tropospheric NO 2 columns (solid line) for the entire domain area as simulated by the model and coincident and averaged measurements for the years 2004-2007 by SCIAMACHY and OMI.In Fig. 4 (left plot) it is apparent that CAMx NO 2 columns increase in the late afternoon, reflecting the diurnal cycle of NO 2 emissions considered in the model (see crosses in left panel of Fig. 4), that have a broad daytime maximum.The most geolocations used in this study are urban and so CAMx emissions show a daytime maximum that mainly reflects intense vehicle use in the early morning and mostly in the late afternoon.A statistical estimation about the diurnal variability of biomass burning, mostly over Africa and South America, where biomass burning is most frequent, is provided in Ellicott et al. (2009).It shows that fire radiative energy (FRE), whose rate is proportional to the biomass consumed, records maximum values during midday.The satellite results show that the mean NO 2 columns observed from OMI are 3.03% smaller than those shown by SCIAMACHY, and this finding is consistent with the expected average diurnal variability estimated by the model simulations (1.58 %).This relative 13:30 to 10:00 UT ratio decrease is consistent with Boersma et al. (2008b) who found that there is a relative decrease in NO 2 column from 10:00 to 13:30 (6% for the satellite observations and 13% for the GEOS-Chem simulations for Europe and even larger for Northeastern United States and China).This decrease can be explained by the broad daytime maximum of the emissions and the stronger photochemical loss in the hours before the OMI overpass compared to the hours before the SCIA-MACHY overpass (Boersma et al., 2008b).The chemical loss of NO x to HNO 3 (through the gas phase NO 2 +OH reaction and by hydrolysis of N 2 O 5 in aerosols) occurs throughout the diurnal cycle but is strongest at midday, when OH concentrations are highest.The 13:30/10:00 LT ratio is not constant throughout the year and the relative difference of OMI estimates versus SCIAMACHY shows a clear seasonal behavior.This seasonality is demonstrated in the right panel of Fig. 4 where we present the seasonal variability of the ratio of tropospheric NO 2 columns from 10:00 LT to 13:30 LT (OMI vs SCIAMACHY) and 10:00 LT to 14:00 LT (CAMx model).The satellite measurements used here are concurrent in order to avoid introducing a sampling bias.As we can see from the satellite measurements and model simulations the observed and predicted 13:30/10:00 LT ratio of tropospheric NO 2 both show a consistent seasonality only for the winter months (December-April).The 13:30 to 10:00 ratios smaller than one are on average larger between February and April and smaller between January and May.During the summer months (May-October) there is a discrepancy in the seasonality between measurements-based and simulation-based ratios, i.e.OMI data of the same day are higher than SCIA-MACHY data for the summer months, which is not reproduced by the model.Similar behavior has been also observed by Boersma et al. (2008b) for areas with large biomass burning events.The emissions in the model do not consider biomass burning, which, according to fire spots available in the World Fire Atlas (http://wfaa-dat.esrin.esa.int/), has a maximum during the warm season (Fig. 4, right panel).This discrepancy was indeed larger for years with intense fire activity in and around the area (2005 and 2007) relative to a year with moderate fire activity (2006).Since biomass burning emissions are not included in CAMx, this inversion of the ratio cannot be verified by the model and will be examined in a future study.

Evaluation of CAMx for the time period 1996-2001
In this section we investigate the consistency of the spatial and temporal variability of NO 2 tropospheric columns measured by GOME and predicted by CAMx.Although GOMEtemis and GOME Bremen look very similar at first sight our analysis, involving the CAMx model, reveals subtle differences.
Figure 5 presents scatter plots of the monthly tropospheric NO 2 columns from the Bremen (left) and KNMI/BIRA (right) algorithms for concurrent measurements and CAMx predictions.The correlation between both GOME algorithm measurements and CAMx predictions is not very high (R≈0.5).As discussed earlier this is probably due to missing CAMx emissions (biomass burning), to CAMx background conditions (no long range transport is considered from sources outside the modeling domain) and also to the fact that the model resolution is 50×50 km, much finer than GOME's.GOME-Bremen data show a better correlation (0.7) especially below 0.3×10 15 molecules/cm 2 (Boersma et al., 2004) results in a slightly different number of observations between the two data sets.There is a published report in the frame of NATAIR research EU project (2007) which reports that NO emissions from lightning for Europe contributes 0.3% to the total emissions from other sources (Simpson et al., 1999).This proportion is quite small in order to consider lightning as important NO 2 source and to take it into account for the explanation of the discrepancies between the satellite measurements and the model predictions.The root mean square error (RMSE) between CAMx and GOME data is less than 2.0×10 15 molecules/cm 2 in all cases and the mean bias (MB) is less than 0.2×10 15 molecules/cm 2 when considering GOME-TEMIS data and less than 0.5×10 15 molecules/cm 2 when considering GOME-Bremen data.These results are consistent with the result shown in the TEMIS Algorithm Document for Tropospheric NO 2 where the tropospheric NO 2 columns from SCIAMACHY are compared to outputs from the CHIMERE model, as shown in Table 3.The normalized mean bias (NMB), considering the monthly mean values, is −4.02%, 3.65% and −8%, whereas considering the Bremen algorithm is 30.27%,5.36% and −1.19% over industrial, rural and urban regions respectively.The normalized mean bias is defined as: Oi and is reported as %.(Pi: the prediction at time and location i, Oi: the observation at time and location i and N: the total i).The differences between the TEMIS and BREMEN retrievals especially over industrial regions (4% versus 30%) is probably due to the fact that although the consistency between the two algorithms is satisfactory over the Balkan region (see Fig. 6) with a correlation coefficient R=0.81, the mean NO 2 GOMEbremen retrievals over industrial areas are lower than CAMx (CAMx: 2.72±1.22×10 15molecules/cm 2 vs GOMEbremen: 2.09±1.40×10 15molecules/cm 2 ), whereas GOMEtemis retrievals are a little bit higher than CAMx ones (CAMx: 2.72±1.26×10 15molecules/cm 2 vs GOMEtemis: 2.84±1.62×10 15molecules/cm 2 ).Considering however the large scatter these differences are hardly statistically significant.Moreover in addition to possible algorithm issues that require further investigation, these differences can also be partly attributed to the different sampling, since the two algorithms do not always provide simultaneous estimates for the same pixel.
Figure 6 shows the scatter between GOME monthly mean tropospheric NO 2 columns for 1996-2001 from both algorithm retrievals for all geolocations (only values greater than 0.3×10 15 molecules/cm 2 , Boersma et al., 2004, were used).The spatial distributions of the two GOME retrievals have a correlation coefficient R=0.81, that points to the high consistency of the two algorithms.GOMEtemis results slightly overestimate GOMEbremen results over the urban and industrial areas, whereas the agreement is much better over rural regions with low tropospheric NO 2 values.
The comparison between the average columns of predicted (CAMx) and observed (GOMEtemis) mean tropospheric NO 2 columns for thirty one of the chosen geolocations for the years 1996-2001 can be seen in Fig. 7 (left).Figure 7 depicts the spatial distribution between predicted and observed tropospheric NO 2 columns as a tool to verify the spatial distribution of the emissions used in the model simulations.Each diamond symbol corresponds to the average tropospheric NO 2 column, derived from concurrent daily

Summary and conclusions
We have intercompared satellite retrievals of tropospheric NO 2 columns from four different instruments, namely GOME/ERS-2, SCIAMACHY/Envisat, OMI/Aura and GOME-2/MetOp, using similar retrieval methods for a time period of more than ten years and we have examined the temporal variability of tropospheric NO 2 vertical columns over Balkan regions deduced by KNMI/BIRA satellite retrievals.We have further evaluated modeling (CAMx) simulations by comparing them with GOME observations from two different algorithms (KNMI/BIRA and IUP Bremen).
The main findings of this work may be summarized as follows: The maximum mean value of tropospheric NO 2 over the Balkan region, as derived from the satellite measurements, is observed over Istanbul, with values of 6.0±4.5, 8.4±7.0,6.5±6.1 and 7.1±6.0×10 15molecules/cm 2 from GOME, SCIAMACHY, OMI and GOME-2 respectively.Over other large urban areas the mean NO 2 tropospheric column densities range, depending on the observing satellite, between 3.0 and 5.0×10 15 molecules/cm 2 with indications for a seasonal variability with and amplitude of 6.0×10 15 molecules/cm 2 .Over industrial complexes the corresponding range is 2.5 to 3.5×10 15 molecules/cm 2 with indications for a seasonal variability with and amplitude of 4.0×10 15 molecules/cm 2 .Over rural areas the mean NO 2 tropospheric column densities range between 1.7 and 2.0×10 15 molecules/cm 2 .Over large cities the observed levels are lower than the ones observed  over the most polluted areas of Southeast Asia and Central Europe which often exceed 11.0×10 15 molecules/cm 2 .
A twelve-month common dataset of GOME and SCIA-MACHY shows a very good agreement (R=0.89) over the Balkan region.Long term common datasets of OMI and GOME-2 are also consistent with SCIAMACHY (R ≈0.86).SCIAMACHY depicts higher NO 2 values than GOME due its finer horizontal resolution, whereas SCIAMACHY and GOME-2 show consistent values due to the similar overpass times and horizontal resolution.OMI observes lower NO 2 tropospheric columns than SCIAMACHY over mostly industrial and urban regions.This is mostly attributed to different local overpass times of each satellite.The differences are affected by the local NO 2 diurnal variability due to a broad daily maximum in emission, combined with large photochemical loss of NO 2 around noon, which, however, can have different characteristics for each location.There is a relative decrease in the NO 2 column (around 3.03%) over the Balkan Peninsula deduced both by satellite measurements and CAMx predictions between 10:00 LT to 13:30 LT.However the ratio of the NO 2 tropospheric columns between 13:30 and 10:00 derived from OMI and SCIAMACHY is not constant a fact that is not reflected in the CAMx simulations.There is a consistent seasonality between the satellite and model determined ratios for the winter months (December-April) and there is a discrepancy in the seasonality during the summer months possibly due to the intense fire activity over the study area.Biomass burning emissions are not included in CAMx.Thus, there is an evident need for better understanding the 13:30/10:00 LT ratio differences between measurements and predictions.
NO 2 data from both GOME algorithms (TEMIS and Bremen) and CAMx predictions show moderate correlation (between 0.5 and 0.7), mostly because of missing CAMx biomass burning emissions and the absence of long range transport of pollution in the model estimates due to the clean boundary conditions chosen for the CAMx model (Katragkou et al., 2009).However the RMSE and MB estimates provide a consistent and slightly better result compared to previous studies, which indicates that for the purpose of air pollution management on a regional scale, highly resolved space-borne data are of great value.Particularly, UV/visible satellite measurements of tropospheric species provide valuable long-term data sets which can be used to evaluate current emission inventories used by various groups, to provide estimates for average conditions over areas with limited groundbased data availability and in addition they have a great potential to detect longterm trends in regional scale pollution.

Figure 1 :
Figure 1: Spatial distribution of the geolocations considered.

Figure 2 Fig. 2 .
Figure 2: a) Time series analysis of satellite-measured NO2 tropospheric columns for the 32 geolocations.The diamond, triangle, square and X symbols indicate the GOME, SCIAMACHY, OMI and GOME2 measurements respectively.b) Scatter plot of concurrent monthly mean satellite measurements (SCIAMACHY and OMI) versus ground measurements that are available through the European Air Quality data base.R is the correlation coefficient.

Fig. 3 .
Fig. 3. Scatter plots of GOME (upper), OMI (middle) and GOME-2 (bottom) versus SCIAMACHY tropospheric NO 2 columns (in 10 15 molecules/cm 2 ) for the entire region.Each point represents a monthly mean value for each geolocation.Nobs are the number of data.The solid line indicates y=x.

Figure 4 :
Figure 4: Left panel: Average diurnal variation of tropospheric NO 2 columns modeled (from 1996 to 2001) by CAMx (asterisk symbols) and observed by SCIAMACHY (from 2004 to 2008) at 10:00LT (diamonds symbol) and OMI (from 2004 to 2008) at 13:30LT (triangle symbol) for the Balkan geolocations.The crosses indicate the average diurnal variation of CAMx NO

Fig. 4 .
Fig. 4. Scatter plots of GOME (upper), OMI (middle) and GOME-2 (bottom) versus SCIAMACHY tropospheric NO 2 columns (in 10 15 molecules/cm 2 ) for the entire region.Each point represents a monthly mean value for each geolocation.Nobs are the number of data.The solid line indicates y=x.

Fig. 5 .
Fig. 5. Scatter plots of GOMEbremen (left plots) and GOMEtemis (right plots) monthly NO 2 tropospheric columns (y axis) versus NO 2 CAMx tropospheric columns (x axis) for urban (top), rural (middle) and industrial (bottom) regions.Columns are in 10 15 molecules/cm 2 .R is the correlation coefficient.Nobs is the number of observations.The solid line represents the y=x

Fig. 7 .
Fig. 7. Tropospheric NO 2 VCDs (10 15 molecules/cm 2 ) from GOMEtemis (left) and GOMEbremen (right) versus NO 2 model tropospheric predictions at the ensemble of 31 geolocations.Each diamond symbol corresponds to the average tropospheric NO 2 column, derived from concurrent daily values for the time period 1996-2001, over a geolocation.R is the correlation coefficient.Nobs is the number of geolocations.

Table 1 .
Sources and characteristics of satellite tropospheric NO 2 data used in this study.

Table 2 .
Mean values and standard deviations of tropospheric NO 2 columns in 10 15 molecules/cm 2 estimated from GOME, SCIAMACHY, OMI and GOME-2 for the urban (U), rural (R) and industrial (I) geolocations considered in this study.

Table 3 .
The statistical values of the comparison between GOME and CAMx tropospheric NO 2 monthly averages.The unit is 10 15 molecules/cm 2 .Scatter plot of GOMEbremen versus GOMEtemis tropospheric NO 2 columns l Balkan geolocations.Nobs is the number of observations.