Quantifying NO x emissions in Egypt using TROPOMI observations

. Urban areas and industrial facilities, which concentrate the majority of human activity and industrial


31
Economic growth in developing countries has led to a strong increase of urban air pollution (Baklanov et al., 2016 [1]).

32
Among the different pollutants, nitrogen oxides are key species. They are generally the products of fuel combustion, 33 such as the combustion of hydrocarbons in the air at high temperature. The main sources of these compounds are 34 therefore vehicle engines, but also heavy industrial facilities such as power plants, metal smelters and cement plants. 35 Their accumulation in the lowest layers of the troposphere contributes to the formation of smog and acid rain (Singh 36 et al., 2007 [2]). They also have a significant effect on human health, as they can cause various respiratory diseases 37 (EPA, US., 2016 [3]). To deal with these phenomena, national and regional governments generally enact a series of air 38 pollution control strategies, which typically take the form of bans on certain polluting technologies, with the aim of 39 reducing the concentration of pollutants at the local level to targets that must be achieved within a given timeframe.

40
[6]). However, most developing countries, such as Egypt, lack the local infrastructure to access detailed information 48 on technical factors such as energy consumption, fuel type and technology, leading to discrepancies in inventories (Xue 49 et al., 2012 [7]). As a consequence, the monitoring of emissions, which is important to evaluate the effects of the air 50 pollution control policies, is of limited reliability.

51
To overcome the uncertainties in the emission inventories, the use of independent observations systems is becoming 52 increasingly prevalent. In this study, we investigate the use of satellite remote sensing of atmospheric concentrations to 53 improve the quantification of NO x emissions in Egypt. Spectrally resolved satellite measurements of solar backscattered sulphur dioxide, carbon monoxide and ozone, as well as aerosols and cloud physical properties. The very high spatial 82 resolution offered by TROPOMI (originally 3.5 × 7 km 2 at nadir, improved to 3.5 × 5.5 km 2 since 6 August 2019) 83 provides unprecedented information on tropospheric NO 2 . Its large swath width (∼ 2600 km) enables to construct 84 NO 2 images on large spatial scales. Those images greatly improve the potential for detecting highly localised pollu-85 tion plumes above the ground, identifying small-scale emission sources but also estimating emissions from megacities, 86 industrial facilities and biomass burning. We use TROPOMI NO 2 retrievals from November 2018 to November 2020 87 over the Middle East and Eastern Mediterranean region, and more specifically over Egypt and the city of Riyadh, 88 Saudi Arabia. The arid climate of this region offers a large number of clear-sky days throughout the year, but also 89 the presence of large plumes of pollutants due to a large human concentration along rivers and around megacities, 90 which allows us to observe high NO 2 concentrations in the region with a high signal-to-noise ratio. These datasets are used to remove the non-anthropogenic part of the NO x emissions signal. We conduct this removal 142 by subtracting the mean emissions over desert and rural areas from the mean emissions over urban and industrial 143 areas. In order to perform this distinction between two types of areas, our study is carried out using a grid with 144 a resolution of 0.1 • × 0.1 • characterised by two types of cells. A cell is considered "urban" if it has a population 145 density higher than a threshold of 100 hab.km −2 , or if its centre is close to an industrial facility. Otherwise, the cell 146 is considered "rural". In order to avoid any smearing that would correspond to abnormally high emissions outside 147 urban areas (which can happen if the wind is poorly estimated), transition cells (in the immediate vicinity of the 148 mentioned urban cells) are also considered to be urban. Figure 1 shows the distinction between urban and rural cells grid is conveniently used as a mask for the urban enhancement, whose calculation is explained in Section 3.4. As a first step, we use TROPOMI's tropospheric NO 2 columns Ω NO2 to derive top-down NO 2 production maps. Using 172 the horizontal wind w, the NO 2 flux is given as Ω NO2 w. The divergence of this flux is added to the local time derivative 173 ∂Ω NO2 ∂t to balance NO 2 sources e NO2 and sinks s NO2 according to the continuity equation: In steady state, the time derivative disappears and the mass balance is reduced to three terms. Rather than calculating the air mass factor, we take this factor into account in the final uncertainty estimates. The . We calculate the value of k mean using a temperature-dependent analytical formula for different 193 pressure ranges (Burkholder et al., 2020 [28]). Note that although this reaction rate accounts for both reactions with 194 OH, the second channel is minor and cannot be considered as a true NO x sink, HOONO being rapidly decomposed 195 back to NO 2 and OH in the lower troposphere (Sander et al., 2011 [39]). The value of k mean therefore represents the 196 total loss of NO 2 due to OH and cannot be used to infer HNO 3 and HOONO production. Thus, the NO 2 production 197 can be calculated as the sum of a transport term and a sink term: is almost at its daily maximum. Similarly, the production of PAN, which peaks in late afternoon and early evening 216 (Seinfeld, 1989), is neglected at 13:30. Finally, the uptake of NO 2 onto black carbon is of limited amount in the 217 Mediterranean region (Friedrich et al., 2021 [48]). All these processes being neglected, the reaction between NO 2 and 218 OH is the only sink that is considered in our calculations to provide a reliable indication of NO x emissions.  domain (of about 450,000 km 2 ) thus reflects chemical losses of the sink term. In this term, the NO 2 lifetime calculation 232 involves the reaction rate k mean , whose annual variability is low due to small changes in Egyptian midday temperatures 233 throughout the year, and OH concentration, whose annual variability is highly marked. In Egypt, tropospheric OH Here, x is the distance in the downwind-upwind direction, B is the NO 2 background, A is the total number of 248 NO 2 molecules observed in the vicinity of the point source, x 0 is the e-folding distance downwind, representing the 249 exponential length scale of NO 2 decay, µ is the location of the apparent source relative to the centre of the point 250 source, and σ is the standard deviation of the Gaussian function, representing the length scale of Gaussian smoothing.

251
Using a non-linear least squares fit, we estimate the five unknown parameters: A, B, x 0 , µ and σ. Using the mean 252 zonal wind module √ w 2 of the NO 2 line density domain, the mean effective NO 2 lifetime τ fit can be estimated from 253 the fitted parameters: The geography of Egypt does not suit the method described here. The Egyptian population is contiguously concen-255 trated along the Nile, which makes it difficult to define point sources isolated from human activity. Furthermore, 256 large isolated cities such as Alexandria or Suez are too close to the coast for the wind to be considered homogeneous. leading to retrievals with a high signal-to-noise ratio. Third, Riyadh is far from the coast, and its flat terrain makes 262 the surrounding wind fields rather homogeneous during most of the year.

263
As the fitting algorithm is very sensitive to any disturbance that might be induced by NO 2 production from other 264 point sources, it is necessary to identify heavy industrial facilities in the area. As Riyadh is also an industrial area, 265 many power plants are located close to the city centre. Figure

304
We calculate NO x emissions on the entire domain from NO 2 production by using CAMS NO and NO 2 concentrations.

305
These quantities are not intended to replace TROPOMI observations; they are used to apply the concentration ratio For robust statistics, these derived emissions can be averaged monthly, enabling a month-by-month comparison with Here we compare the results of the TROPOMI line densities fits for Riyadh to the lifetime calculated by Equation

337
(2) using CAMS data. The two years of TROPOMI observations (from November 2018 to November 2020) provide a     the TROPOMI slant column densities, also highlights these cities. However, unlike the transport term, which has 380 a similar spatial pattern from month to month, the sink term is clearly stronger in summer than in winter. This is  Table 1 compares the monthly values for the sink term and the absolute value of the transport term above five major 391 cities of the country, with populations ranging from 193,000 to 19 million inhabitants, as well as Ain-Sokhna's area.

392
The mean values for TROPOMI column densities are also provided. According to the results, the capital city of Cairo why Aswan, which has a population that is comparable to Beni Suef or Asyut, but which does not have any major 403 industrial site, has slightly lower emissions than the two other cities. Finally, the Gulf of Suez displays relatively large 404 emissions, which might be attributed to the shipping sector, the region being a major gateway for international trade.

405
Because it also hosts several flaring sites, these emissions might also be due to the oil and gas extraction activity.  Table 1: Comparison between the transport term and the sink term above different cities among the 20 most populous cities in Egypt, as well as the industrial region of Ain Sokhna located 45 km southwest of Suez for January and July 2019. Numbers correspond to average values within 18 km from the city centre. The value for the mean TROPOMI NO2 column density is also given. The population density of the corresponding governorate (2020 value) is noted as a comparison. Unit M stands for a quantity of 10 15 molecules (NO2 or NOx).
Although these cities and areas can be described as high-emission sites, the term responsible for these emissions differ  Here we investigate the influence of the choice of the vertical level in the representation of the model parameters.

426
This influence is of considerable importance, as NO x sources in urban areas can be located at different altitudes. For 427 instance, emissions from the road sector from tailpipes are located at ground level, whereas NO x from power plants 428 and industrial facilities can be emitted from stacks, which are usually located between 50 and 300 m above the ground.

429
The results of Section 4.1 showed that level B was more appropriate for the representation of the NO 2 lifetime. This given for the months of January, April, July and October 2019 in Table 2. As the wind field is only involved in the 434 transport term whose spatial integration nearly leads to zero, the influence of this parameter is not studied.   The transition to the level A generally results in a decrease in temperature, leading to an increase in the reaction rate 436 k mean and thus an increase in the emissions from the sink term. This transition has only a small influence on the 437 total NO x emission estimates, with the total increasing by 4 to 6%. The effect is slightly more pronounced in urban 438 areas, due to a steeper vertical temperature profile in these areas. The influence of OH goes in the opposite direction: 439 its concentration decreases strongly with altitude, weakening the sink term. The share of emissions due to the sink 440 term being proportional to this parameter, the effect of the vertical is very pronounced. Thus, the transition to level 441 A weakens the sink term by 4 to 9% in summer (with an average of -6.03% for the months June/July/August) and 442 by 9 to 26% in winter (with an average of -15.70% for the months December/January/February). This weakening 443 seems more pronounced over urban areas than over rural areas from March to October, and more pronounced over  during the other days of the week. We therefore try to characterise this feature, by evaluating the weekly cycle of 453 NO x emissions. We use the TROPOMI-inferred emissions to obtain averages per day of the week. We use the quality 454 assurance q a of TROPOMI retrievals to ignore the days for which more than 20% of the domain has low-quality  Here, we attempt to compare our TROPOMI-derived NO x emissions to emissions from CAMS and EDGAR inventories.

503
The total calculated for each month therefore corresponds to the NO x production by human activities in Egyptian 504 urban and industrial areas. After aggregating the different sectors of activity, CAMS and EDGAR inventories directly 505 provide the anthropogenic NO x emissions over the same areas. All NO x emissions are expressed in mass terms as NO. 506 We note that the EDGAR inventory does not cover the period 2018-2020 (the inventory ends in 2015). On Figure 9, this decrease is mainly due to a relatively low value of the OH concentration which reaches 5.86 ×10 6 molecules.cm −3 512 on average for these two months, with 4.95 ×10 6 molecules.cm −3 above urban areas and 6.09 ×10 6 molecules.cm −3 513 over rural areas. They were respectively 6.96, 6.94 and 6.97 ×10 6 molecules.cm −3 for the previous year (December 514 2018-January 2019) and 7.24, 6.94 and 7.31 ×10 6 molecules.cm −3 for the year before (December 2017-January 2018).

515
A decrease in tropospheric columns (-14.3% for urban areas and -4.6% for rural areas) also contributes to this drop.  shown in Figures 5 and 9 are thus calculated from uncertainty statistics whose references are presented in this section.

545
Since these references do not specify the exact nature of these statistics, we assume they correspond to standard 546 deviations. The uncertainty of tropospheric NO 2 columns under polluted conditions is dominated by the sensitivity 547 of satellite observations to lower tropospheric air masses, expressed by the tropospheric air-mass factor (AMF small. Therefore, we neglect the impact of temperature on final uncertainty. As a consequence, the propagation of 564 these different uncertainties on the monthly estimates of NO x emissions in Egypt leads to an expanded uncertainty 565 between 40 and 43%. For lifetimes calculated with the EMG function fitting, the corresponding expanded uncertainty 566 ranges between 18% and 79%. 567 We acknowledge the fact that our treatment of uncertainties is simplified there. Many minor sinks highlighted in which leads to absolute uncertainties that are roughly proportional to monthly emissions. As a result, the confidence 571 interval displayed on Figure 9 is larger in summer than in winter (with a length of 6.0 kt in January 2020 and of 17.4 572 kt for July 2020), and the drop in emissions for winter 2019-2020 appears as a persistent feature of the model outputs.

573
If this drop is realistic, then our top-down model provides a method for improving national inventories. If it is not, 574 then several assumptions of our model can be questioned. For instance, because this drop is largely due to variations 575 in OH concentrations provided by CAMS, whose reliability has been evaluated for Riyadh, then the transposability 576 hypothesis between Riyadh and Egypt may be subject to wider discussion. A better understanding of OH levels in 577 Egypt, supported by in-situ measurements, might answer these questions and allow to improve our model.