Atmospheric Environment 101 (2015) 82e93
Contents lists available at ScienceDirect
Atmospheric Environment journal homepage: www.elsevier.com/locate/atmosenv
Identification of surface NOx emission sources on a regional scale using OMI NO2 I. Zyrichidou a, *, M.E. Κoukouli a, D. Balis a, K. Markakis b, A. Poupkou a, E. Katragkou a, c, I. Kioutsioukis d, D. Melas a, K.F. Boersma e, f, M. van Roozendael g a
Laboratory of Atmospheric Physics, Physics Department, A.U.Th, Thessaloniki, Greece Laboratoire de Meteorologie Dynamique (LMD), IPSL Ecole Polytechnique, Palaiseau Cedex, Paris, France Department of Meteorology and Climatology, School of Geology, Aristotle University of Thessaloniki, Thessaloniki, Greece d Laboratory of Atmospheric Physics, Physics Department, University of Patras, Greece e Royal Netherlands Meteorological Service, De Bilt, The Netherlands f Eindhoven University of Technology, Fluid Dynamics Lab, Eindhoven, The Netherlands g Belgian Institute for Space Aeronomy, Brussels, Belgium b c
h i g h l i g h t s The Balkan a posteriori NOx emissions inventory resulted in 1.11 Tg N/y. Fossil fuel NOx emissions over Greece comprise of the 80% of the total emissions. Soil emissions, omitted in a priori, accounted for 20% of the total over Greece. Microbial activity emissions are important in high resolution emission inventories. Biomass burning NOx emission rate, accounted for 0.5 106 Tg N/km2 over Greece.
a r t i c l e i n f o
a b s t r a c t
Article history: Received 13 May 2014 Received in revised form 10 November 2014 Accepted 11 November 2014 Available online 12 November 2014
In this study, an inverse modeling technique is applied to obtain, at a regional scale, top-down emission estimates for nitrogen oxides utilizing tropospheric nitrogen dioxide (NO2) columns retrieved by the OMI/Aura instrument and estimated by the Comprehensive Air Quality Model with extensions (CAMx). The main idea, applied previously using models with coarse spatial resolution, is to combine the a priori information from the bottom up emission inventory used in an air quality simulation that covers the Balkan peninsula in a high resolution grid (0.1 0.1 ) with the tropospheric NO2 quantities estimated for one complete year by CAMx and the tropospheric NO2 columns retrieved by satellite observations in order to identify missing emissions sources on a regional scale. The results have identified biases between the a priori and a posteriori emission inventories due to the missing emission sources or overestimation of the spread and quantity of certain emission sources. In such a fine resolution grid we have also analyzed and considered the horizontal transport on the a posteriori NOx emissions. The deduced a posteriori NOx emissions, dominated by the fossil fuel emissions, were found to be1.11 ± 0.30 Tg N/y, compared to 0.87 ± 0.43 Tg N/y found in the a priori Balkan emission inventory. Soil emissions over the extended Greek domain, omitted in the a priori inventory, were estimated to account for almost 20% of the total emitted amount, while for the year 2009 the biomass burning NOx emission flux was also estimated and the average rate accounted for 0.5 106 Tg N/km2. © 2014 Elsevier Ltd. All rights reserved.
Keywords: Nitrogen oxides Emission inventory OMI Inverse modeling
1. Introduction
* Corresponding author. E-mail address:
[email protected] (I. Zyrichidou). http://dx.doi.org/10.1016/j.atmosenv.2014.11.023 1352-2310/© 2014 Elsevier Ltd. All rights reserved.
One of the major pollutants in the troposphere are nitrogen oxides (NOx ¼ NO þ NO2). These compounds affect tropospheric chemistry, air quality and, as a result, climate change (e.g. Shindell
I. Zyrichidou et al. / Atmospheric Environment 101 (2015) 82e93
et al., 2009). Mostly they regulate the levels of ozone in troposphere, they lead to the formation of nitric acid (OH þ NO2 þ M / HNO3 þ M) and influence hydroxyl (OH) concentrations. Nitrogen oxides are released to the atmosphere from various anthropogenic and natural sources like fossil fuel combustion, biomass burning, microbial processes in soils and lightning. Fossil fuel combustion is the main anthropogenic source, mainly caused by traffic and heavy industry. Olivier et al. (1998) estimate that 75% of the yearly global NOx emissions are coming from anthropogenic sources and roughly 50% of the total NOx sources is accounted for by fossil fuel combustion only. Gas-phase formations of HNO3 during daytime and N2O5-hydrolysis during nighttime are the dominant sinks of tropospheric NOx (e.g. Jacob, 2000). Accurate knowledge of NOx sources is required for an adequate description of both the present and future state of the atmospheric composition. Nearly two decades of satellite trace gas column measurements has led to high spatio-temporal resolution trace gas databases with complete global spatial coverage. Space-based observations provide timely information that can be used to more accurately reflect current emissions (e.g. Martin et al., 2006) compared to the measurements provided by the rather sparse ground-based instruments. The tropospheric NO2 columns derived from satellite observations have been used to evaluate chemical transport models (e.g. Zyrichidou et al., 2009; Huijnen et al., 2010 and references therein), to study long term changes in anthropogenic emissions of NOx over developing areas (e.g. Stavrakou et al., 2008; Wang et al., 2012) and to examine the spatio-temporal patterns of NOx emission on a global and regional scale (e.g. Leue et al., 2001; Castellanos and Boersma, 2012). In recent studies these measurements have already been used to provide topedown estimates of land surface NOx emissions via inverse modeling on the global scale (Martin et al., 2003; Jaegle et al., 2005; Lin et al., 2010; Miyazaki et al., 2012) and on the regional scale like in France (e.g. Deguillaume et al., 2007), China (Wang et al., 2007; Kurokawa et al., 2009), Southeastern United States (Boersma et al., 2008; Napelenok et al., 2008) and India (Ghude et al., 2013) or on continental scale like in Europe (Konovalov et al., 2006) as well as for the estimation of NOx emissions from different kind of sources such as lightning (e.g. Miyazaki et al., 2014 and references therein), soils (e.g. Bertram et al., 2005; Wang et al., 2007; Lin, 2012), biomass burning (e.g. Spichtinger et al., 2001; Leue et al., 2001), fossil fuel combustion (e.g. Jaegle et al., 2005; Zhao and Wang, 2009) or more specifically ships (e.g. Beirle et al., 2004; Richter et al., 2004). The emission updates offer an improved estimate of NOx that are crucial to our understanding of air quality, and climate change. The main objective of this study is to identify and then estimate NOx emissions at a regional scale for missing emission sources, mainly biomass burning and soil, by employing measurements observed by the Ozone Monitoring Instrument (OMI) on board the Aura satellite. In addition, it is within the aims of this work to assess disagreements between the model calculated and space-based constrained emission fields and eventually to complete and improve the existing estimates of 10 10 km2 gridded NOx emissions. These emissions are used in the chemistry transport model Comprehensive Air Quality Model with extensions, CAMx, in a run for the Balkan domain. The OMI satellite measurements with their associated high spatial resolution provide a rich dataset for inverse modeling studies. In recent literature only few papers on regional emissions have used the OMI NO2 retrievals to derive NOx emissions, and in particular the new version of KNMI (Royal Netherlands Meteorological Institute) DOMINO algorithm, version 2.0 (Boersma et al., 2011), used here. For the inversion modeling method we used the mass balance approach (Martin et al., 2003). This approach has been used widely (Boersma et al., 2008; Zhao and Wang, 2009;
83
Lamsal et al., 2010) but to the best of the authors’ knowledge, no published research has yet been conducted to estimate regional NOx emissions over the Balkan Peninsula in such fine spatial resolution. Lamsal et al. (2011) discusses that the current knowledge about NOx emissions in Eastern European countries may be inadequate. Furthermore, Schaap et al. (2013) attributes the poor model (LOTOS-EUROS, Long Term Ozone Simulation-European Ozone Simulation) performance over Southeastern Europe to the more complex terrain and the variable meteorology dominated in this area and mainly highlight the need of the estimation of more representative emission information for this part of Europe. In Section 2 we provide an account of the current a priori NOx emissions and a description of the CAMx model. The OMI, the tropospheric NO2 retrieval and also the comparison between measured and simulated tropospheric NO2 vertical column densities (VCDs) are also presented in this section. The method of the top-down and a posteriori emission inventories computation and error statistics are described in Section 3. Section 4 shows an optimization of the partitioning of the a posteriori emission inventory for each NOx source over the extended Greek domain. Summary remarks are presented in Section 5. 2. Model and measurement data description 2.1. CAMx model In the present study, the simulated NO2 tropospheric vertical column densities (VCDs) are based on the results of the air quality forecast modeling system Fifth Generation Mesoscale Model version 3.7, MM5-CAMx (http://lap.phys.auth.gr/gems/). The Comprehensive Air Quality Model with extensions (CAMx) version 4.40 is applied for four nested grids that share the same Lambert Conformal Conic projection (User's Guide CAMx e Comprehensive Air Quality Model with Extensions, 2006). The coarse grid covers Europe and has a spatial resolution equal to 30 km. In this study we use the second grid that focuses on the Balkan Peninsula with a 10-km spatial resolution and the third and the fourth domains of 2-km resolution that extend over the two largest Greek urban agglomerations, Athens (37.99 N, 23.77 E) and Thessaloniki (40.62 N, 22.97 E) respectively. The vertical profile of the domain contains 15 layers of varying thickness, extending up to about 300 hPa, due to unsatisfying results above probably associated to problems in the coupling with the Model for Ozone And Related Chemical Tracers e Integrating Forecasting System (MOZART-IFS) (Zyryanov et al., 2012). However, over the latitudes considered in this study the average tropopause is close to 200 hPa and this fact is not expected to dominate possible discrepancies in the results. Therefore, a small part of the differences in tropospheric NO2 between model and OMI might be attributed to the low top boundary of the model setup. Validation, meteorological and emission inventory information of MM5 e CAMx model has already been elaborately presented in Zyrichidou et al. (2013). 2.2. OMI tropospheric NO2 column retrievals OMI is the Dutch Finnish Ozone Monitoring Instrument on NASA's EOS Aura satellite. OMI features the highest spatial resolution (13 km [along track] 24 km [across track] at nadir) among the UVeVis spectrometers aboard polar-orbiting, sun e synchronous satellites. Its design and performance are described in detail in Levelt et al. (2006). The OMI data sets used in this study are tropospheric NO2 VCDs retrieved by KNMI within the DOMINO project and are publicly available on a day-by-day basis via ESA's Tropospheric Emission Monitoring Internet Service (TEMIS) at http://www.temis.nl/airpollution/no2.html. In this study the NO2
84
I. Zyrichidou et al. / Atmospheric Environment 101 (2015) 82e93
tropospheric VCDs from the DOMINO product, version 2.0, for a 12month period are used. The improved DOMINO retrieval algorithm v2.0 is described in Boersma et al. (2011) and in the DOMINO Product Specification Document (http://www.temis.nl/docs/OMI_ NO2_HE5_1.0.2_2011.pdf). According to Boersma et al. (2011) improvements were found in the calculation of the air mass factors, through improved radiative transfer modeling, the use of highresolution terrain height and surface albedo data, and better a priori NO2 profiles (improving the sampling of the Tracer Model Version 4 (TM4)). For providing a detailed analysis of the spatial distribution of emissions over the Balkans and for easy comparison with the CAMx model, monthly averaged NO2 VCDs retrieved from level 2 OMI orbit files were regridded onto a 0.1 lon 0.1 lat grid (attributing an OMI value to a 0.1 0.1 model grid cell by using the corner coordinates) and compared with monthly mean model results sampled at the time of the satellite overpass. Concerning the satellite retrievals we used screening criteria for some of the important parameters following the suggestions in the DOMINO Product Specification Document (http://www.temis.nl/ docs/OMI_NO2_HE5_2.0_2011.pdf) in order to retrieve valid tropospheric NO2 VCDs over the Balkans. As a criterion for cloud screening, we used only observations with cloud radiance fractions of less than 50%, which corresponds to a cloud fraction of less than about 20% (Van der A et al., 2008) and implies that the model is evaluated for (mostly) clear-sky conditions. During the analysis period several row anomalies (which appear since June 2007) occurred in OMI data. The affected rows have been removed from the dataset; see http://www.temis.nl and Boersma et al. (2009). Finally, we discarded scenes with surface albedo values greater than 0.3 and tropospheric air mass factor (AMF) of less than 0.1 (extensive description on these choices can be found in Zyrichidou et al., 2013). The Averaging Kernels (AKs) of DOMINO v2.0 is applied at each CAMx grid box and vertical profile in order to compute weighted NO2 tropospheric VCDs and to make the data comparable to the OMI observations (the AKs computation is described elaborately in Zyrichidou et al., 2013). The advantage of a comparison through the AKs is that the comparison is now independent of the a priori profile shape chosen in the retrieval (Eskes and Boersma, 2003).
2.3. Comparison between simulated and retrieved vertical column densities of tropospheric NO2 In this section we investigate the performance of the model and satellite retrievals. The 1 h averaged tropospheric NO2 predictions
(between 13:00 and 14:00 LT) correspond to the range of OMI overpass times and hereafter are referred to as Ua while the OMI retrievals as Ut in units 1015 molecules/cm2. The left panel of Fig. 1 depicts the yearly averaged difference between retrieved and simulated NO2 tropospheric VCDs without using the AKs information over the Balkan domain. The white dashed line frame defines the extended Greek domain that is included in the initial Balkan domain. Likewise, the right panel presents the corresponding difference weighted by the OMI AKs. The negative values in Fig. 1 denote the cases where CAMx simulations overestimate the magnitude of retrieved NO2 tropospheric VCDs. It appears that the distributions are rather similar in the two plots of Fig. 1, especially over the rural regions. Minor differences were found for the area e averaged model columns with (right plot) and without (left plot) applying the AKs. For the case of not using the AKs the spatial correlation coefficient of the mean annual values between OMI and CAMx is found to be slightly lower by 3.5% and the Root Mean Square Error is larger over all types of regions (urban, industrial, rural) (Zyrichidou et al., 2013). These discrepancies could be partly attributed to the differences between the TM4 model (used for the OMI retrievals) (Boersma et al., 2011) and CAMx as well as to the limitation of the OMI to resolve tropospheric NO2 near the surface (reflected in the AKs) where we have the highest molecular density and therefore greatest contribution to the column. The application of the AKs results in a positive annual Mean Bias P (MB ¼ N1 N i¼1 ðPi OiÞ, where Pi is the prediction and Oi is the observation at time and location i and N is the number of measurements) of 0.25 1015 molecules/cm2 instead of a negative one (0.19 1015 molecules/cm2) when the AK information is not used. The impact of the AKs on the CAMx simulations is decreased NO2 tropospheric VCDs reported by the model over polluted regions, urban and industrial sites, and over the south Mediterranean shipping routes. The above findings agree with the results of other studies (Huijnen et al., 2010; Miyazaki et al., 2012) that used simulated NO2 tropospheric VCDs weighted by the AKs. Since the application of the AK in CAMx generally shows better agreements with OMI retrievals, the analysis in the next sections has been done with these weighted NO2 tropospheric VCDs of CAMx. The right panel of Fig. 1 shows that over large cities (like Athens and Thessaloniki) CAMx overestimates by 3.0e6.0 1015 molecules/ cm2 the tropospheric NO2 VCDs, while secondary maximum differences (1.0e3.0 1015 molecules/cm2) can be identified over some Southeastern European capitals for e.g. Belgrade (44.80 N, 20.47 E) and Bucharest (44.44 N, 26.13 E) and industrialized areas (e.g. Ptolemaida (40.52 N, 21.70 E) and Eskisehir (39.78 N, 30.52 E)). Given that the CAMx a priori emission inventory was compiled for
Fig. 1. Annual distribution of OMI retrieved minus CAMx simulated tropospheric NO2 columns (unit: 1015 molecules/cm2). The white dashed line frame defines the extended Greek domain. The white numbers are referred to: 1 for Athens, 2 for Thessaloniki, 3 for Ptolemaida, 4 for Thessaly plain, 5 for Thrace and 6 for Peloponnese. Left plot: without applying the averaging kernels. Right plot: with applying the averaging kernels.
I. Zyrichidou et al. / Atmospheric Environment 101 (2015) 82e93
the reference year 2003, the decreases in NO2 column densities over densely populated Balkan regions for the year studied here (April 2009eMarch 2010) may be partly attributed to model biases or errors in the a priori emissions of 2003 and to the reduced anthropogenic activities. Vrekoussis et al. (2013) reported 30e40% NO2 reductions over Athens from 2008 onwards mostly as an impact of the recent economic recession. . Recently, a number of studies have used OMI-NO2 data to investigate the trends of nitrogen dioxide across Europe for the period 2005e2010 (e.g. Konovalov et al., 2010; Castellanos and Boersma, 2012; Zhou et al., 2012). Although approaches used in these studies differ, they show a consistent picture, with decreasing trends (z20% on average) in most of Europe. The larger positive differences (5.0e8.0 1015 molecules/cm2) of NO2 tropospheric VCDs are found over Istanbul (41.07 N, 29.01 E). In Markakis et al. (2012), the TNO (Netherlands Organisation for Applied Scientific Research) inventory was found to indicate a possible underestimation of NO2 of more than 57% for Istanbul. On an annual basis, the spatial correlation coefficient between modeled and retrieved VCDs reaches almost 0.6 which shows that CAMx model captures fairly well the spatial distribution of retrieved VCDs. Note that in Zyrichidou et al., 2013, the OMI retrieval used in this study was compared with ground-based measurements (available at http://www.ypeka.gr/) from an urban station in Athens in order to give more insight into how the surface layer NO2 is reflected in the tropospheric NO2 VCDs and the results showed a rather good correlation (R ¼ 0.6). This finding indicates that the elevated NOx sources are insignificant especially over urban regions.
3. Inversion method or NOx emissions estimation 3.1. Top-down NOx emission estimates from OMI Following the mass balance method (Leue et al., 2001) as described by Martin et al. (2003, 2006), the Balkan top-down land surface NOx emissions, Et, for April 2009 to March 2010 from the respective retrieved tropospheric NO2 VCDs, Ut, are computed through the following relationship for each grid box of our domain (of 0.1 lon x 0.1 lat):
. Etði;jÞ ¼ a*Utði;jÞ ¼ Eaði;jÞ Uaði;jÞ *Utði;jÞ
(1)
where the coefficient a is derived from the simulation with a priori emissions, Ea, and the indices (i,j) indicate the location of the grid cell. The coefficient a equals Ea/Ua where Ua is the tropospheric NO2 VCDs from that simulation. In the above method the horizontal transport is neglected, as was also performed in similar studies (e.g. Lin et al., 2010; Lin, 2012). However, in these studies the grid cells were coarser than the 0.1 0.1 cells of CAMx and neglecting horizontal transport and assuming a linear relationship (Equation (1)) between NOx emissions and NO2 VCDs may introduce significant uncertainties in the a posteriori NOx emissions, for instance in areas downwind of hotspots (strong emissions sources) (Mijling et al., 2012). High a posteriori emissions will probably be inferred but over the hotspots themselves the a posteriori emissions are likely to be underestimated because of the NO2 being vented away from the source, e.g. Beirle et al., 2011). In summer the tropospheric lifetime of NOx is relatively short due to the photochemistry and thus the impacts of horizontal transport in summer are less significant than in winter over polluted areas. Consequently, we can expect both minimal carryover from the previous day and minimal transport from sources to regions downwind, yielding a column that is especially representative of local surface emissions for more than half of our period. However, although the NOx lifetime in summer is shorter than in
85
winter it may be still long enough to get vented out of the grid cell. The method used in this study for top-down constraints on the NOx emissions works reasonably well for coarse grid cells (e.g. 2.0 2.5 as used in GEOS-Chem by Martin et al., 2003) because the majority of the NOx will stay within the same grid cell after emission taking into account the lifetime of NOx. Martin et al. (2003) indicated that the spatial smearing error can be neglected if the grid length is greater than 100 km. Consequently, in our study the presumed linearity might certainly be compromised because of transport. Since the presumed linear relation between surface emission and vertical column concentrations may result in a decrease in local vertical column densities compensated by an increase elsewhere, we should evaluate the uncertainty in the inversion results using the outputs of a process analysis from a CAMx run. To this purpose we calculated the local sensitivities of concentration over emission through a process analysis method and hence we evaluated the uncertainties introduced in the inverse modeling method used in this study. This analysis enabled us to examine if, for 0.1 0.1 grids and even when averaging over many days, there was smearing of the NO2 signal away from the location, where the NOx was originally emitted and to evaluate the effect of transport away from the source. 3.1.1. Results of process analysis A process analysis method was used for CAMx model in order to analyze the contributions of several processes (focusing on the horizontal transport) to the emissions in each CAMx grid cell. This method combines integrated process rate information across several cells. This is useful for analyzing the contributions of model processes to a geographic area that spans multiple cells (e.g., an urban area). As mentioned above since the horizontal transport affects mostly the areas around the hotspots we chose a multi-cell area over two urban regions (Athens and Thessaloniki) and a twocell area over an industrial region (Ptolemaida) to study this effect. Fig. 2 shows daily contributions to the final nitrogen oxides concentrations (orange line) and to the emissions (blue line) over Athens (top), Thessaloniki (middle) and Ptolemaida (bottom) in a winter month (February, left plots) and in a summer month (June, right plots). To reduce the number of processes being displayed in order to make the plots less complicated, the horizontal advection for each boundary (south, north, east and west boundary), the gas phase chemistry, the two deposition processes (dry and wet) and the diffusion are combined into a single “other processes” term (green line) and any series with zero contribution over any type of areas is excluded from the chart. The bar charts in Fig. 2 show the daily time series of contributions from all major processes. Note that the green line in the plots of Fig. 2 is dominated by horizontal advection terms, since the rest of the processes are negligible. The positive and negative terms almost cancel each other and the contribution to the final concentration is small (z0.4 1015 molecules/cm2 on average) but not negligible over all types of regions. Comparing all the processes which are dominated by the slightly negative horizontal advection (west, east, south and north boundary advection) to the area emissions plus point source emissions (referred as “emissions” term in Fig. 2) at a certain time (a 1-h time period at around 13:30 LT) in a grid cell, we estimated the uncertainties introduced in our a posteriori NOx emissions (section 3.2). On an annual basis, the percentage uncertainty [((emissions horizontal advection)/emissions) * 100] ranges between 6.0 and 10.0% over urban and 15.0% over industrial regions on average, whilst on a daily basis the effect of horizontal transport processes seems to be more significant (see Fig. 2). No substantial differences were found between the two seasons. In order to avoid this source of uncertainty in our NOx emissions estimates, we considered the approach used by Toenges-Schüller et al. (2006),
86
I. Zyrichidou et al. / Atmospheric Environment 101 (2015) 82e93
Fig. 2. Daily (JD: Julian Day) contribution to NOx (NO þ NO2) final concentrations (orange line) from emissions (blue line) and other processes (green line) in February (left plots) and in June (right plots) over Athens (top), Thessaloniki (middle) and Ptolemaida (bottom). Note the different y-axis. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Boersma et al. (2008) and Lamsal et al. (2010) to account for the emissions from adjacent grid cells. 3.1.2. Optimized top-down NOx emissions Since the smearing error cannot be neglected in our study, a smoothing kernel should be applied in developing the top-down NOx emissions inventory. The easy way to simulate this smoothing without the need of explicit assumptions about meteorological parameters is the application of a kernel (Toenges-Schüller et al., 2006; Boersma et al., 2008), K, on each of the cell of the a priori emission field and this is defined as:
0 1 1 @ 1 K¼ pþ8 1
1 p 1
1 1 1A 1
(2)
where p is a smoothing parameter. It should be noted that e.g. for areas with a strong dominating wind direction or regions with a
gradient in orography the applied kernel will have some limitations. In this work our grid horizontal resolution is 0.1 0.1 which is much finer than the one in previous studies (e.g. 2 2.5 in Boersma et al., 2008). This led us to use a wider matrix, 11 11 (121 elements) instead of 3 3 (trying to approach the K matrix's spatial extent of the previous studies), in Equation (2) This matrix is defined by using 9 free p parameters of the same value instead of only one used in Toenges-Schüller et al. (2006) and Boersma et al. (2008). We assume that the kernel elements are equal to 1 for the remaining 112 grid boxes. After many trials before the final definition of the 11 11 matrix, trying different number of p parameters, as well as p/2, p/3, etc values, we concluded that the matrix used in this study is reasonable since it smoothes the uncertainties estimated via the process analysis and which are caused by the horizontal transport effect. We have tried the Gaussian shape as in Beirle et al. (2011) [for Riyadh, the capital of Saudi Arabia] and Ialongo et al. (2014) [for Helsinki], however the correlation coefficient (R2 < 0.5) was quite low. In spite of the known
I. Zyrichidou et al. / Atmospheric Environment 101 (2015) 82e93
87
Martin et al., 2003, 2006; Wang et al., 2007). Particularly, the a posteriori inventory, Ep (Equation. (3)), and its error 3 p (Equation. (4)) are obtained by combining the top-down NOx emission inventory, Et, with a priori estimates from the bottom-up inventory, Ea, and their relative errors 3 t, 3 a (discussed below in Section 3.3). Assuming a log-normal distribution of errors, the optimized inventory is obtained as:
shortcomings with the above boxed-shape kernel it achieves a better R2 (>0.6) than a simple Gaussian smoothing and thus the selected matrix, in combination with the appropriate p parameter was considered to be the optimal choice for the case of Balkan domain. Thereby, considering the horizontal transport of the emitted NOx we proceed to the application of the above kernel, K, to Equation (1) to obtain the smoothed topedown NOx emissions inventory. The inversion is conducted on a grid box by grid box basis to derive the respective emissions. Fig. 3 (left plot) depicts the annual distribution of the a priori total NOx emissions included in the CAMx model. The NOx hotspots appear over the greater area of Athens and the power plants of Ptolemaida (known as the “Energy Valley” of Greece) and Eskisehir and then over major Balkan cities like Budapest (47.5 N, 19.05 E), Vienna (48.21 N, 16.37 E), Bucharest, Thessaloniki, Naples (40.83 N, 14.25 E) and the industrialized area of Megalopolis (37.40 N, 22.14 E). The ship tracks in the Mediterranean are also evident whereas Istanbul, as discussed above, appears to provide a low rate of emissions in direct contrast with its population and economic development due to an unrepresentative a priori emission inventory (Markakis et al., 2012). The OMI data are regridded to the CAMx resolution before the application of Equation (1). This equation is applied over the total Balkan domain (land and sea). The resulting annual top e down OMI inventory for the Balkans is displayed in the right plot of Fig. 3. Both top-down and a priori datasets suggest significant surface NOx sources over the greater area of Athens and Ptolemaida. However, the a priori emissions seem to be overestimated over Athens resulting in large concentrations of tropospheric NO2 (Fig. 1). As it is shown in Fig. 3 (right plot), OMI allows the identification of characteristic patterns of major ship tracks not only over the Mediterranean (west of Crete and southeast of Italy) and Adriatic Sea (not evident in the a priori), but also over the Aegean Sea and the Dardanelles (a long narrow strait dividing the Balkans along the Gallipoli peninsula from Asia Minor) where the ship routes are more dense. Our topedown results over the Mediterranean Sea agree with those of Marmer et al. (2009) to the point that ship signals are recognizable from the satellite NO2 retrievals over the Western Mediterranean. Finally, the top-down emissions are higher over Istanbul as well as over the Asia Minor coastline. Overall the OMI emission inventory (1.1 Tg N/y) is about 25% higher compared to the a priori (0.87 Tg N/y).
Over areas where OMI observations are not available or top down emissions are equal to zero (less than 0.2% of the whole domain) the a posteriori directly reflects the a priori inventory. The left plot of Fig. 4 gives the spatial distribution of NOx emissions for the a posteriori inventory computed from Equation (3). The annual surface NOx emissions in the a posteriori inventory are 1.11 ± 0.30 Tg N/y, not significantly different from the CAMx a priori (0.87 ± 0.43 Tg N/y) values. However, there are some regional differences shown in the right plot of Fig. 4. Overall, the annual mean optimized NOx Balkan source is up to 15% higher than the a priori emission. Differences between the a posteriori and a priori inventories in Southeastern Europe could reflect an underestimation of soil NOx emissions in the a priori and a possible overestimation (more than 50%) of fossil fuel emissions in the a priori over Athens and Thessaloniki also shown below. It is noticeable that whilst soil emissions are lacking from the a priori inventory of CAMx, many studies have suggested substantial soil NOx emissions for the Balkans (e.g. Steinkamp and Lawrence, 2011) and in Europe (Jaegle et al., 2005). As already discussed, there are many studies that assess NOx emission inventories through satellite observations on a global scale and some recent ones that use satellite constraints to improve model performance on a regional scale. The Balkan Peninsula is a region that has not been studied regionally by inverse modeling methods so far. Therefore, we cannot directly compare our results with previous estimates.
3.2. A posteriori emission inventory
3.3. Estimation of uncertainties
The a posteriori emissions are estimated as error weighted averages of the smoothed a priori and top-down emissions (e.g.,
Topedown approaches aim to reduce the discrepancies between the model and observation by taking into account the error
ln Ep ¼
ln Ea ðln 3 t Þ2 þ ln Et ðln 3 a Þ2 ðln 3 t Þ2 þ ðln 3 a Þ2
(3)
with its relative error expressed as:
1 1 1 þ 2 ¼ ðln 3 t Þ2 ðln 3 a Þ2 ln 3 p
Fig. 3. Annual CAMx a priori (left plot) and top-down (right plot) surface NOx emissions in 103 kg.
(4)
88
I. Zyrichidou et al. / Atmospheric Environment 101 (2015) 82e93
Fig. 4. Left plot: Annual a posteriori surface NOx emissions in 103 kg. Right plot: Annual spatial distribution of the percentage difference between a posteriori and a priori inventories ((Ep Ea/Ea) * 100).
in both model and retrievals. A priori emission inventories contain large uncertainties related to statistics and emission factors, while air mass factor, clouds, surface albedo and profile shape, are some of the error sources of the satellite retrievals (Boersma et al., 2011). According to Lin (2012) potential sources of model errors include emissions of NOx emissions of other pollutants affecting the chemistry of NOx, the chemical mechanism for NOx, the scheme for mixing in the boundary layer, and the meteorological fields. The estimation of the errors in input and output results is a crucial issue in inversion modeling procedures (e.g. Konovalov et al., 2006). Errors in model simulations are estimated often from ground-based or aircraft measurements of nitrogen (e.g., Hudman et al., 2007; Lamsal et al., 2010). The total model error from factors other than emissions of NOx is estimated to be about 30e50% (Martin et al., 2003; Jaegle et al., 2005; Wang et al., 2007). Therefore, a number of assumptions and crude estimations have to be used in our statistical analysis. The mean monthly a priori relative error (3 a) is taken into account under the assumption that it amounts to the 50% of the mean monthly a priori (Ea) value for each grid box for the inversion purposes of this study. The error in the a priori emissions over Europe is more than 30% (5.5 ± 1.7 Tg N/y and 5.0 ± 2.0 Tg N/y in Jaegle et al., 2005; in Martin et al., 2003 respectively) on average. Specifically, in Jaegle et al. (2005) the a priori emissions error is smaller than 50% over regions dominated by fuel combustion and increases over the rest of the world. Moreover, in Miyazaki et al., 2012 the initial error is set to 40% of the a priori emission (CHASER model). The emission inventory that was used in the MM5eCAMx simulations is a compilation process of several emission inventories (Zyrichidou et al., 2013). The spatial resolution of the OMI data (¼13 km 24 km) is much finer than that for e.g. of the EMEP grid (¼50 km) from where the emissions for the cargo ships were extracted. Thus, there are might be large representativeness errors in the model because of unresolved small-scale variation. Consequently, since there is no specific error estimate in the CAMx a priori emissions inventory, considering also the bibliography mentioned above, we postulate that 50% error is a reasonable percentage in our case. The annual 3 a for each grid box is computed by the geometric standard deviation. The annual 3 a for the whole domain is estimated applying the propagation error analysis function. Errors in the retrieval of tropospheric NO2 VCDs (Ut) derive from the calculation of SCDs and its tropospheric portion over cleaner regions, and mainly arise from the calculation of AMFs for polluted regions (Boersma et al., 2007). Estimated errors in the tropospheric NO2 VCDs are 20e30%, about ±1.0 1015 molecules/cm2 (an
absolute error), under clear-sky conditions, likely with a magnitude larger in winter than in summer but can reach up to 100% under certain conditions (Boersma et al., 2011; Bucsela et al., 2013). Similar to the annual 3 a, the annual 3 t for each grid box is calculated using the geometric standard deviation of the monthly values of 3 t. A propagation error analysis (Equation a) is applied in order to provide an accurate computation of the top-down monthly relative error, 3 t. Finally, the annual 3 a and 3 t (Equation b, where n is the total number of grid boxes) (3 t value: around 25% of the topedown NOx emissions) for the whole Balkan domain are used in Equation (4) in order to estimate the annual a posteriori relative error (3 p).
vEt 3a vEa
2
vEt 3Ut vUt
2
2
¼
2
¼ ð3 t1 Þ2 þ ð3 t1 Þ2 þ ð3 t2 Þ2 þ …… þ ð3 tm Þ2
3t
3t
þ
(a)
(b)
From all the above statistical analysis we tentatively result to a 30% on average annual error in the a posteriori emission inventory. The a posteriori inventory includes the missing emissions of the a priori inventory. The next step is to partition first the top down and then the a posteriori NOx emissions among fossil fuel, soil and biomass burning. 4. Apportionment of OMI derived emissions 4.1. Methodology Since the case of producing NOx emissions over a Balkan area especially when there is no a priori information has not been explored yet, we focused this part of our study, as a test case scenario, on the extended domain of Greece. This part of study is focalized on the portioning of surface NOx emissions among fossil fuel emissions (that include the sectors: Energy, Heating, Distribution, Industry, Off road land, Off road sea, Road, Solvents) and soil and biomass burning emissions that are not determined in the MM5-CAMx run following the approach described by Jaegle et al. (2005). The latter did not attempt to constrain lightning emissions as in most inverse estimates (Wang et al., 2007; Zhao and Wang, 2009; Lin et al., 2010) considering that the effect of lightning NOx on inversion is small. However, Schumann and Huntrieser (2007) have provided a best estimate of 5 ± 3 Tg N for the annual global lightning NOx emissions. There are quite a few studies that showed that taking into account lightning NOx is important, especially in summer and mostly in the tropics (Napelenok et al.,
I. Zyrichidou et al. / Atmospheric Environment 101 (2015) 82e93
2008; Miyazaki et al., 2014). In particular, according to Miyazaki et al. (2014) the total lightning NOx production is globally scaled to 6.3 TgN yr1 (CHASER model) a number that is mostly attributed to enhanced lightning sources over the tropics and the U.S. NorthEast. Unfortunately, to the best of the authors’ knowledge, there is not yet any reference about the lightning estimation over the Balkan Peninsula. Hence, in our study we consider that the magnitude of lightning emissions over the Balkans is very low. Similarly to Jaegle et al. (2005) a block diagram (Fig. 5) is applied to each month and each 0.1 0.1 grid box using information from the spatiotemporal distribution of a priori total (Ea) and fossil fuel NOx emissions (Eaff), top-down total NOx emissions (Et) and space-based fire observations available here from the World Fire Atlas, WFA, (http://wfaa-dat.esrin.esa.int/). Particularly, this flow chart assumes the a posteriori non-lightning emissions to be solely anthropogenic if the a priori anthropogenic emissions exceed 90% of the a priori total emissions or if they exceed the a posteriori non-lightning emissions. Otherwise, differences between the a posteriori and a priori emissions are attributed to soil or biomass burning sources using fire spots. Since biomass burning emissions are missing in CAMx, we explore if this method can help their computation over grid boxes where fire has been detected (Section 4.4). For each grid box where fires exist (Fire > 0) the estimation of background soil emissions is achieved by calculating the median soil emissions in grid boxes within a Dlat ¼ 0.3 in latitude and a Dlon ¼ 0.4 in longitude where there are no fires. The median soil emissions are calculated according to the Equation (1) in Jaegle et al. (2005). Once background soil emissions are determined, the residual is assigned to biomass burning. Accordingly, replacing the total and per sector a priori information with the corresponding total and per sector top down emissions in the left flowchart of Fig. 5 (Jaegle et al., 2005), we construct the right flowchart of the same Figure. This new block diagram allows the portioning of the total a posteriori emission inventory (Ep) into a posteriori fossil fuel (Epff), soil (Es) and biomass burning (Ebb) emissions. The annual distribution of a posteriori NOx emissions per sector is given in Table 1. Individual optimization for each NOx source of the two inventories (a priori and a posteriori) is shown in the next sections. The CAMx modeling system currently runs without biomass burning and soil emissions. On the global scale, soils represent
89
Table 1 Annual surface NOx emissions per sector over the extended Greek domain (units Tg N y1). NOx sources
A Priori
A Posteriori
Fossil fuel Soil Biomass burning
0.84 0.03 e
0.80 0.17 0.0008
approximately 15% of global NOx emissions, of which 70% are estimated to originate from the tropics. However, there is a considerable uncertainty in the number of global soil emissions, since estimates over the last decade indicate soil NOx emissions to be between 4 and 20 Tg N/yr (e.g. Steinkamp and Lawrence, 2011; Hudman et al., 2012; Stavrakou et al., 2013). In some cases, soil emissions may contribute a lot more to the local NOx budget on a regional scale, such as over NortheCentral Montana where the local NOx budget is controlled exclusively by soil emissions (Bertram et al., 2005). In addition, fertilizer application during the cultivation period and precipitation events results in soil NOx emission fluxes enhancements (e.g. Akiyama et al., 2003). On the other hand, fires, as one of the largest sources of biomass burning emissions, have a major impact on air quality and climate (Langmann et al., 2009). At the same time, biomass burning emissions account for less than 15% of global surface NOx emissions and also fires over Africa account for half of the global biomass burning emissions (Jaegle et al., 2005). According to the results of Jaegle et al. (2005) the percentage with which the biomass burning emissions contribute to the European surface NOx emissions is almost 7.0%. The largest contribution to this percentage comes from countries in Northeastern Europe (Baltic countries, western Russia, Belarus, and the Ukraine) during late summer (end of July and August) (Amiridis et al., 2010). Forest fires in this area are a major source of pollution in the Northern Hemisphere and especially Europe (Korontzi et al., 2006). However, forest fires are also a real threat to natural environments in the countries of Mediterranean Europe and in August 2007, after a severe drought, forest fires broke out in the area, resulting in a major disaster that received global attention (Gitas et al., 2008).
Fig. 5. Block diagram of algorithm (adapted from Jaegle et al., 2005) applied to partition top-down (left flowchart) OMI NOx emissions Et into: fossil fuel (Etff), soil (Ets) and biomass burning (Etbb) emissions and a posteriori (right flowchart) NOx emissions Ep into: fossil fuel (Epff), soil (Eps) and biomass burning (Epbb) emissions.
90
I. Zyrichidou et al. / Atmospheric Environment 101 (2015) 82e93
urban areas, they are not detected by OMI through the inverse modeling method over Athens and Thessaloniki.
4.2. Estimated fossil fuel emissions In Fig. 6 the a priori and a posteriori emission inventories are illustrated respectively over the extended Greek domain. The annual a priori and a posteriori fossil fuel emissions are in quantitative agreement (0.40 Tg N/y and 0.42 Tg N/y respectively). However, the a priori and a posteriori inventories are moderately spatio-temporally correlated (R ¼ 0.73, n ¼ 23,560). Large differences are observed over the Minor Asia coastline and the greater area of Istanbul (45%) where the a priori emission inventory is underestimated, in the Aegean and the Dardanelles (55%) where the ship tracks are more distinct in the a posteriori emission inventory and over the urban regions of Athens (80%) and Thessaloniki (60%) where the a priori emissions seem to be overestimated compared to the posteriori emission inventory. These differences are shown in Fig. 1 (right plot) as well. The differences are slighter over the rest rural areas of the domain under study, ranging between 10 and 25%. Finally, it is noteworthy that the ship lines among the Aegean islands in the a posteriori emission inventory can be seen somehow as distinct ship tracks because of the various routes between numerous islands for the biggest part of the year. 4.3. Estimated soil emissions As discussed in 4.1, for the grid boxes where fires exist the a posteriori soil emissions (Ets) are calculated according to Equation (5) that gives the median soil emissions in grid boxes within a Dlat ¼ 0.3 in latitude and a Dlon ¼ 0.4 in longitude where there are no fires.
Ets ¼ Median½Ets ðDlat; DlonÞFire¼0
(5)
For the grid boxes where there is no fire the a posteriori soil emissions are calculated according to the Equation (6).
Ets ¼ Et Etff
(6)
In Fig. 7 (left plot), despite the absence of soil emissions in the a priori emission inventory, the a posteriori soil emissions are shown explicitly and are equal to 0.12 Tg N/y over the extended Greek domain. The largest soil emissions detected by OMI are found over the Thessaly plain (central Greece) as well as over Thrace (northeast Greece) and in some places over Peloponnese, areas known for their major agricultural activities and the extensive use of fertilizers. These areas also appeared as agricultural areas and forests (light yellow and green colors) in the right plot of Fig. 7 (http:// www.eea.europa.eu). Microbial activity also leads to substantial soil NOx emissions. Since soil emissions are not large over great
4.4. Estimated biomass burning emissions Because of the large spatial and temporal variability of biomass burning, emissions monitoring and forecasting must be based on satellite observations of the currently active fires (Kaiser et al., 2006). As mentioned above, biomass burning activities are not considered in the MM5-CAMx run. Thus, we examine whether the implementation of the above inverse modeling method can provide biomass burning NOx emissions constrained by OMI. In the last step of the block diagram of Fig. 5 we use fire statistic data available in the World Fire Atlas (http://wfaa-dat.esrin.esa.int/ ) focusing on Greece. The left plot of Fig. 8 presents the spatial distribution of the fire hotspots over Greece from April 2009 to March 2010. The red spots in the left plot of Fig. 8 correspond to fire episodes over these specific grid boxes as extracted from the WFA database. The right plot depicts the monthly fire count distribution. These spots are used as the grid cells over which biomass burning emissions are calculated through the block diagram of Fig. 5. We have to note that in the case of biomass burning emissions the range of values is extended up to 150 tn per grid box, referred to as hotspot in Fig. 8 (left plot). The annual NOx biomass burning emissions, considering the fire episodes over Greece during the period, are estimated to be about 0.8 Gg N/year from the a posteriori inventory. For the time period of our analysis the a posteriori amount is distributed as follows: 93% of the biomass burning emissions on an annual basis are observed in August, i.e. 0.7 Gg N, 3% in June, 2% in July and 1% for each of the months of September and November. These amounts are expected to be much greater for years such as 2007, when a series of massive forest blazes were observed in several areas across Greece throughout the summer. In the framework of the Monitoring Atmospheric Composition and Climate (MACC) project, the estimation of NOx biomass burning emissions according to Global Fire Emission Database (GFEDv.3) data resulted in 2.1 Gg N(4.84 kilotons of NOx) in August 2007 in the Peloponnese (Poupkou et al., 2014). In late August 2007, large parts of the Peloponnese suffered from forest fires. It is estimated that almost 3138 km2 of forests in Greece, according to the provisional data from the European Commission Joint Research Center (JRC) were burnt in 2007 most of them in August (EFFIS Forest Fires in Europe, (2007) report, 2008). Considering that our annual estimation for fire episodes in Greece covered about 1700 km2 in this study, the results seem to be quite consistent. The average rate that is inferred from both studies is about 0.5 106 Tg N/km2. Although the total amounts of NOx from biomass burning emissions are not
Fig. 6. Fossil fuel a priori (left) and a posteriori NOx emissions (right) in 103 kg.
I. Zyrichidou et al. / Atmospheric Environment 101 (2015) 82e93
91
Fig. 7. Soil a posteriori NOx emissions (103 kg) over the greater area of Greece (left) and a map of Europe illustrating the land use (courtesy of: http://www.eea.europa.eu) (right).
so large as in the tropics and subtropics, where fire is also a humandriven process (Van Leeuwen and Van der Werf, 2011), the accurate embedment of this NOx source in the regional inventory is essential, especially in cases of large fire episodes.
5. Concluding remarks This paper presents the first implementation of an inverse modeling methodology on a regional scale with a fine spatial resolution model aiming to study the perspectives of improvements to model performance over the Balkan region. We used OMI/Aura satellite observations of tropospheric NO2 VCDs in order to examine the possibility of improving the existing estimates for the spatial and temporal distribution of NOx emissions that are employed in a regional air quality model. The model evaluation with OMI data and ground-based measurements has already been done in Zyrichidou et al., 2013 for the Balkan region and for the year 2009e2010. The in situ measurements showed moderate correlation with OMI tropospheric NO2 column retrievals and the timeseries showed a similar seasonality between the two data sets following a double peak during the winterespring period. CAMx and OMI DOMINO v2.0 are better correlated (R z 0.6) over the urban sites and the seasonality results were in good agreement. Overall, during the summer, the TNO database underestimates slightly NO2 columns over the whole domain. The discrepancies during summer may be
attributed to missing biomass emission sources in the CAMx model that are combined with fire episodes usually observed during this season. Thus the new satellite-derived NOx emission inventory is expected to partly improve the model performance. In particular, we applied a fast inverse modeling scheme to identify missing CAMx emissions over the extended Greek domain and finally to incrementally increase the quality of the a priori inventory. The main outcomes of our study are summarized in the following: - Applying an inverse modeling method in the Balkan domain the a posteriori NOx emissions resulted in1.11 ± 0.30 Tg N/y (compared to 0.87 ± 0.43 Tg N/y in the a priori), which is about 12% higher than a priori estimates. The deduced Balkan a posteriori inventory in our study rises to almost the 21% on average of the total NOx emissions over Europe found in previous recent studies. - The uncertainties introduced by the horizontal transport effect are estimated through a process analysis and range on average between 6% and 15% on annual basis, depending on the type of region. To avoid the effects of these smearing errors, the turbulent diffusion simulated using a smoothing kernel, reducing the total uncertainty of our computations to about 30%. - The a posteriori fossil fuel NOx emissions over the extended Greek domain comprise of the 80% of the a posteriori total NOx
Fig. 8. Spatial distribution of fire hotspots (left plot) and monthly distribution of fire counts (right plot) in Greece. Courtesy of the World Fire Atlas (http://wfaa-dat.esrin.esa.int/) database.
92
I. Zyrichidou et al. / Atmospheric Environment 101 (2015) 82e93
emissions which now include soil and biomass burning emissions. - The a priori soil emissions are absent in the model run over the Balkan domain and should include registered (fertilizer-related) soil NOx emissions for controlled agricultural practices. In addition, microbial activity also leads to substantial soil NOx emissions and this is typically not taken into account by regional CTMs (see e.g. Huijnen et al., 2010). In this study the application of an inversion method resulted in the derivation of 0.12 Tg N/y (almost 20% of the total NOx emissions) over the extended Greek domain. From this work it is inferred that these emissions are important and need to be included in future high resolution emission inventories or in any case, to be accounted for in regional models. - The about 1.5% residual of the total a posteriori NOx emissions (considering that we disregard the impact of lightning over the Balkan Peninsula) belongs to biomass burning emissions coming from fires over Greece. 93% of the biomass burning emissions on an annual basis are observed in August (i.e. 0.0007 Tg N/year), 3% in June, 2% in July and 1% for each of the months September and November. - For the year 2009, the biomass burning NOx emissions were estimated and account for 0.4 106 Tg N/km2, which is quite consistent with the those used in MACC project (0.6 106 Tg N/ km2). The results presented above indicate that space-based NO2 measurements and novel inverse modeling methods are essential for improving the quality of regional emission inventories by identifying and quantifying the different NOx sources and further decrease the existing model uncertainties.
Acknowledgments The authors acknowledge the free use of tropospheric NO2 column data from the OMI/Aura sensor from www.temis.nl. Part of this work was supported by the FP7 EU project Monitoring Atmospheric Composition and Climate Interim Implementation (MACC ІІ) (Grant Agreement no. 283576). Part of this work has been supported by the European FP7 project PASODOBLE (Space-based applications at the service of European Society e Preoperational validation of GMES services and products e Stimulating the development of downstream services) Grant Agreement No 241557. K.F. Boersma acknowledges funding from the Netherlands Organisation for Scientific Research, NWO Vidi grant 864.09.001.
References Akiyama, H., et al., 2003. Nitrous oxide, nitric oxide, and nitrogen dioxide fluxes from soils after manure and urea application. J. Environ. Qual. 32, 423e431. Amiridis, V., Giannakaki, E., Balis, D.S., et al., 2010. Smoke injection heights from agricultural burning in Eastern Europe as seen by CALIPSO. Atmos. Chem. Phys. 10, 11567e11576. Beirle, S., Platt, U., von Glasow, R., et al., 2004. Estimate of nitrogen oxide emissions from shipping by satellite remote sensing. Geophys. Res. Lett. 31, L18102. http:// dx.doi.org/10.1029/2004GL020312. Beirle, S., Boersma, K.F., Platt, U., et al., 2011. Megacity emissions and lifetimes of nitrogen oxides probed from space. Science 333, 1737e1739. http://dx.doi.org/ 10.1126/science.1207824, 2011. 28697, 28702, 28706. Bertram, T.H., Heckel, A., Richter, A., et al., 2005. Satellite measurements of daily variations in soil NOx emissions. Geophys. Res. Lett. 32 (24), L24812. http:// dx.doi.org/10.1029/2005GL024640. Boersma, K.F., et al., 2007. Near-real time retrieval of tropospheric NO2 from OMI. Atmos. Chem. Phys. 7, 2103e2118. Boersma, K.F., et al., 2008. Validation of OMI tropospheric NO2 observations during INTEX-B and application to constrain NOx emissions over the eastern United States and Mexico. Atmos. Environ. 42, 4480e4497.
Boersma, K.F., Dirksen, R.J., Veefkind, J.P., et al., 2009. Dutch OMI NO2 (DOMINO) Data Product, HE5 Data File User Manual. Tech. rep., KNMI, TEMIS website, available at: http://www.temis.nl/airpollution/no2.html. Boersma, K.F., Eskes, H.J., Dirksen, R.J., et al., 2011. An improved retrieval of tropospheric NO2 columns from the Ozone Monitoring Instrument. Atmos. Meas. Tech. 4, 1905e1928. Bucsela, E.J., Krotkov, N.A., Celarier, E.A., et al., 2013. A new stratospheric and tropospheric NO2 retrieval algorithm for nadir-viewing satellite instruments: applications to OMI. Atmos. Meas. Tech. Discuss. 6, 1361e1407. http:// dx.doi.org/10.5194/amtd-6-1361-2013. Castellanos, P., Boersma, K.F., 2012. Reductions in nitrogen oxides over Europe driven by environmental policy and economic recession. Sci. Rep. 2 (265), 1e7. http://dx.doi.org/10.1038/srep00265. Deguillaume, L., Beekmann, M., Menut, L., 2007. Bayesian Monte Carlo analysis applied to regional-scale inverse emission modeling for reactive trace gases. J. Geophys. Res. 112, D02307. http://dx.doi.org/10.1029/2006JD007518. EFFIS, 2008. Forest Fires in Europe 2007 Report, Annual Fire Report. European Forest Fire Information System, JRC. http://effis.jrc.ec.europa.eu/reports/firereports. ENVIRON, September 2006. User's Guide CAMx e Comprehensive Air Quality Model with Extensions, Version 4.40. ENVIRON International Corporation. Eskes, H.J., Boersma, K.F., 2003. Averaging kernels for DOAS total-column satellite retrievals. Atmos. Chem. Phys. 3, 1285e1291. http://dx.doi.org/10.5194/acp-31285-2003. Ghude, S.D., Pfister, G.G., Jena, C., et al., 2013. Satellite constraints of nitrogen oxide (Nox) emissions from India based on OMI observations and WRF-Chem simulations. Geophys. Res. Lett. 40, 423e428. http://dx.doi.org/10.1029/ 2012GL053926. Gitas, I.Z., Polychronaki, A., Katagis, T., Mallinis, G., 2008. Contribution of remote sensing to disaster management activities: a case study of the large fires in the Peloponnese, Greece. Int. J. Remote Sens. 29 (6), 1847e1853. http://dx.doi.org/ 10.1080/01431160701874553. Hudman, R.C., et al., 2007. Surface and lightning sources of nitrogen oxides over the United States: magnitudes, chemical evolution, and outflow. J. Geophys. Res. 112, D12S05. http://dx.doi.org/10.1029/2006JD007912. Hudman, R.C., Moore, N.E., Martin, R.V., et al., 2012. A mechanistic model of global soil nitric oxide emissions: implementation and space based-constraints. Atmos. Chem. Phys. 12, 7779e7795. http://dx.doi.org/10.5194/acp-12-77792012. Huijnen, V., Eskes, H.J., Poupkou, A., et al., 2010. Comparison of OMI NO2 tropospheric columns with an ensemble of global and European regional air quality models. Atmos. Chem. Phys. 10, 3273e3296. Ialongo, I., Hakkarainen, J., Hyttinen, N., et al., 2014. Characterization of OMI tropospheric NO2 over the Baltic Sea region. Atmos. Chem. Phys. Discuss 14, 2021e2042. http://dx.doi.org/10.5194/acpd-14-2021-2014. Jacob, D.J., 2000. Heterogeneous chemistry and tropospheric ozone. Atmos. Environ. 34, 2131e2159. Jaegle, L., Steinberger, L., Martin, R.V., Chance, K., 2005. Global partitioning of NOx sources using satellite observations: relative roles of fossil fuel combustion, biomass burning and soil emissions. Faraday Discuss. 130, 407e423. http:// dx.doi.org/10.1039/b502128. Kaiser, J.W., Schultz, M.G., Gregoire, J.M., et al., 2006. Observation Requirements for global biomass burning emission monitoring. In: Proc. 2006 EUMETSAT Met. Sat. Conf. Konovalov, I.B., Beekmann, M., Richter, A., Burrows, J.P., 2006. Inverse modelling of the spatial distribution of NOx emissions on a continental scale using satellite data. Atmos. Chem. Phys. 6, 1747e1770. http://www.atmos-chemphys.net/6/ 1747/2006/. Konovalov, I.B., Beekmann, M., Richter, A., et al., 2010. Multi-annual changes of NOx emissions in megacity regions: nonlinear trend analysis of satellite measurement based estimates. Atmos. Chem. Phys. 10, 8481e8498. http://dx.doi.org/ 10.5194/acp-10-8481-2010. Kurokawa, J.-i., Yumimoto, K., Uno, I., Ohara, T., 2009. Adjoint inverse modeling of NOx emissions over eastern China using satellite observations of NO2 vertical column densities. Atmos. Environ. 43, 1878e1887. http://dx.doi.org/10.1016/ j.atmosenv.2008.12.030. Korontzi, S., McCarty, J., Loboda, T., et al., 2006. Global distribution of agricultural fires in croplands from 3 years of Moderate Resolution Imaging Spectroradiometer (MODIS) data. Glob. Biogeochem. Cycles 20 (2), GB2021. http:// dx.doi.org/10.1029/2005GB002529. Lamsal, L.N., Martin, R.V., van Donkelaar, A., et al., 2010. Indirect validation of tropospheric nitrogen dioxide retrieved from the OMI satellite instrument: insight into the seasonal variation of nitrogen oxides at northern midlatitudes. J. Geophys. Res. 115, D05302. http://dx.doi.org/10.1029/2009JD013351. Lamsal, L.N., Martin, R.V., Padmanabhan, A., et al., 2011. Application of satellite observations for timely updates to global 2086 anthropogenic NOx emission inventories. Geophys. Res. Lett. 28, L05810. Langmann, B., Duncan, B., Textor, C., et al., 2009. Vegetation fire emissions and their impact on air pollution and climate. Atmos. Environ. 43, 107e116. Leue, C., Wenig, M., Wagner, T., et al., 2001. Quantitative analysis of NOx emissions from GOME satellite image sequences. J. Geophys. Res. 106, 5493e5505. Levelt, P.F., et al., 2006. The ozone monitoring instrument. IEEE Trans. Geosci. Rem. Sens. 44 (5), 1093e1101. Lin, J.-T., McElroy, M.B., Boersma, K.F., 2010. Constraint of anthropogenic NOx emissions in China from different sectors: a new methodology using multiple
I. Zyrichidou et al. / Atmospheric Environment 101 (2015) 82e93 satellite retrievals. Atmos. Chem. Phys. 10, 63e78. http://dx.doi.org/10.5194/ acp-10-63-2010. Lin, J.-T., 2012. Satellite constraint for emissions of nitrogen oxides from anthropogenic, lightning and soil sources over East China on a high-resolution grid. Atmos. Chem. Phys. 12, 2881e2898. http://dx.doi.org/10.5194/acp-12-28812012. Markakis, K., Im, U., Unal, A., et al., 2012. Compilation of a GIS based high spatially and temporally resolved emission inventory for the greater Istanbul area. Atmos. Pollut. Res. 3, 112e125. Marmer, E., Dentener, F., Aardenne, J.V., et al., 2009. What can we learn about ship emission inventories from measurements of air pollutants over the Mediterranean Sea? Atmos. Chem. Phys. 9, 6815e6831. Martin, R.V., et al., 2003. Global inventory of nitrogen oxide emissions constrained by space-based observations of NO2 columns. J. Geophys. Res. 108 (D17), 4537. http://dx.doi.org/10.1029/2003JD003453. Martin, R.V., Sioris, C.E., Chance, K., et al., 2006. Evaluation of space-based constraints on global nitrogen oxide emissions with regional aircraft measurements over and downwind of eastern North America. J. Geophys. Res. 111, D15308. http://dx.doi.org/10.1029/2005JD006680. Mijling, B., Van derA, R.J., 2012. Using daily satellite observations to estimate emissions of short-lived air pollutants on a mesoscopic scale. J. Geophys. Res. 117 http://dx.doi.org/10.1029/2012JD017817. Miyazaki, K., Eskes, H.J., Sudo, K., 2012. Global NOx emission estimates derived from an assimilation of OMI tropospheric NO2 columns. Atmos. Chem. Phys. 12, 2263e2288. Miyazaki, K., Eskes, H.J., Sudo, K., Zhang, C., 2014. Global lightning NOx production estimated by an assimilation of multiple satellite dataset. Atmos. Chem. Phys. 14, 3277e3305. http://dx.doi.org/10.5194/acp-14-3277-2014. Napelenok, S.L., Pinder, R.W., Gilliland, A.B., Martin, R.V., 2008. A method for evaluating spatially-resolved NOx emissions using Kalman filter inversion, direct sensitivities, and space-based NO2 observations. Atmos. Chem. Phys. 8, 5603e5614. Olivier, J.G.J., Bouwman, A.F., van der Hoek, K.W., Berdowski, J.J.M., 1998. Global air emission inventories for anthropogenic sources of NOx, NH3 and N2O in 1990. Environ. Pollut. 102, 135e148. Poupkou, A., Markakis, K., Liora, N., et al., 2014. A modeling study of the impact of the 2007 Greek forest fires on the gaseous pollutant levels in the Eastern Mediterranean. Atmos. Res. 149, 1e17. Richter, A., Eyring, V., Burrows, J.P., et al., 2004. Satellite measurements of NO2 from international shipping emissions. Geophys. Res. Lett. 31, L23110. http:// dx.doi.org/10.1029/2004GL020822. Schaap, M., Kranenburg, R., Curier, L., et al., 2013. Assessing the Sensitivity of the OMI-no2 product to emission changes across europe. Remote Sens. 5, 4187e4208. http://dx.doi.org/10.3390/rs5094187. Schumann, U., Huntrieser, H., 2007. The global lightning-induced nitrogen oxides source. Atmos. Chem. Phys. 7, 3823e3907. http://dx.doi.org/10.5194/acp-73823-2007. Shindell, D.T., Faluvegi, G., Koch, D.M., et al., 2009. Improved attribution of climate forcing to emissions. Science 326, 716e718. http://dx.doi.org/10.1126/ science.1174760.
93
Spichtinger, N., Wenig, M., James, P., et al., 2001. Satellite detection of a continentalscale plume of nitrogen oxides from boreal forest fires. Geophys. Res. Lett. 28, 4579e4582. Stavrakou, T., et al., 2008. Assessing the distribution and growth rates of NOx emission sources by inverting a 10-year record of NO2 satellite columns. Geophys. Res. Lett. 35 http://dx.doi.org/10.1029/2008GL033521. Stavrakou, T., Müller, J.-F., Boersma, K.F., et al., 2013. Key chemical NOx sink uncertainties and how they influence top-down emissions of nitrogen oxides. Atmos. Chem. Phys. 13, 9057e9082. Steinkamp, J., Lawrence, M.G., 2011. Improvement and evaluation of simulated global biogenic soil NO emissions in an AC-GCM. Atmos. Chem. Phys. 11, 6063e6082. http://dx.doi.org/10.5194/acp-11-6063-2011. Toenges-Schüller, N., Stein, O., Rohrer, F., et al., 2006. Global distribution pattern of anthropogenic nitrogen oxide emissions: correlation analysis of satellite measurements and model calculations. J. Geophys. Res. 111, D05312. Van der A, R.J., Eskes, H.J., Boersma, K.F., et al., 2008. Trends, seasonal variability and dominant NOx source derived from a ten year record of NO2 measured from space. J. Geophys. Res. 113, D04302. http://dx.doi.org/10.1029/2007JD009021. Van Leeuwen, T.T., Van der Werf, G.R., 2011. Spatial and temporal variability in the ratio of trace gases emitted from biomass burning. Atmos. Chem. Phys. 11, 3611e3629. Vrekoussis, M., Richter, A., Hilboll, A., et al., 2013. Economic crisis detected from space: air quality observations over Athens/Greece. J. Geophys. Res. Lett. 40, 1e6. http://dx.doi.org/10.1002/grl.50118. Wang, Y.X., McElroy, M.B., Martin, R.V., et al., 2007. Seasonal variability of NOx emissions over east China constrained by satellite observations: implications for combustion and microbial sources. J. Geophys. Res. 112, D06301. http:// dx.doi.org/10.1029/2006JD007538. Wang, S.W., Zhang, Q., Streets, D.G., et al., 2012. Growth in NOx emissions from power plants in China: bottom-up estimates and satellite observations. Atmos. Chem. Phys. 12, 4429e4447. http://dx.doi.org/10.5194/acp-12-4429-2012. Zhao, C., Wang, Y.H., 2009. Assimilated inversion of NOx emissions over east Asia using OMI NO2 column measurements. Geophys. Res. Lett. 36, L06805. http:// dx.doi.org/10.1029/2008GL037123. Zhou, Y., Brunner, D., Hueglin, C., et al., 2012. Changes in OMI tropospheric NO2 columns over Europe from 2004 to 2009 and the influence of meteorological variability. Atmos. Environ. 46, 482e495. Zyryanov, D., Foret, G., Eremenko, M., et al., 2012. 3-D evaluation of tropospheric ozone simulations by an ensemble of regional Chemistry Transport Model. Atmos. Chem. Phys. 12, 3219e3240. http://dx.doi.org/10.5194/acp-12-32192012. Zyrichidou, I., Koukouli, M.E., Balis, D.S., et al., 2009. Satellite NO2 observations and model simulations of tropospheric columns over Southeastern Europe. Atmos. Chem. Phys. 9, 6119e6134. Zyrichidou, I., Koukouli, M.E., Balis, D.S., et al., 2013. Evaluation of high resolution simulated and OMI retrieved tropospheric NO2 column densities over the Balkan region. Atmos. Res. 122, 55e66. http://dx.doi.org/10.1016/ j.atmosres.2012.10.028.