Quantifying burning efficiency in megacities using the NO2/CO ratio from the Tropospheric Monitoring Instrument (TROPOMI)

This study investigates the use of co-located nitrogen dioxide (NO2) and carbon monoxide (CO) retrievals from the TROPOMI satellite to improve the quantification of burning efficiency and emission factors (EFs) over the megacities of Tehran, Mexico City, Cairo, Riyadh, Lahore, and Los Angeles. Efficient combustion is characterized by high NOx (NO+NO2) and low CO emissions, making the NO2/CO ratio a useful proxy for combustion efficiency (CE). The local enhancement of CO and NO2 above megacities is well captured by TROPOMI at short averaging times compared with previous satellite missions. In this study, the upwind background and plume rotation methods are used to investigate the accuracy of satellitederived 1NO2/1CO ratios. The column enhancement ratios derived using these two methods vary by 5 % to 20 % across the selected megacities. TROPOMI-derived column enhancement ratios are compared with emission ratios from the EDGAR v4.3.2 (Emission Database for Global Atmospheric Research v4.3.2) and the MACCity (Monitoring Atmospheric Chemistry and Climate and CityZen) 2018 emission inventories. TROPOMI correlates strongly (r = 0.85 and 0.7) with EDGAR and MACCity, showing the highest emission ratio for Riyadh and lowest emission ratio for Lahore. However, inventory-derived emission ratios are 60 % to 85 % higher than TROPOMI column enhancement ratios across the six megacities. The short lifetime of NO2 and the different vertical sensitivity of TROPOMI NO2 and CO explain most of this difference. We present a method to translate TROPOMI-retrieved column enhancement ratios into corresponding emission ratios, thereby accounting for these influences. Except for Los Angeles and Lahore, TROPOMI-derived emission ratios are close (within 10 % to 25 %) to MACCity values. For EDGAR, however, emission ratios are ∼ 65 % higher for Cairo and 35 % higher for Riyadh. For Los Angeles, EDGAR and MACCity are a factor of 2 and 3 higher than TROPOMI respectively. The air quality monitoring networks in Los Angeles and Mexico City are used to validate the use of TROPOMI. For Mexico City and Los Angeles, these measurements are consistent with TROPOMI-derived emission ratios, demonstrating the potential of TROPOMI with respect to monitoring burning efficiency.


Introduction
Rapid urbanization and economic growth in developing countries has led to a strong increase in urban air pollution (Pommier et al., 2013;United Nations, 2018).In the South Asian cities of Kabul and Dhaka, for instance, nitrogen dioxide (NO 2 ) increases in the order of 10 % yr −1 have been reported (Schneider et al., 2015).In New Delhi, emis-sions of carbon monoxide (CO) increased by 22.4 % from 2000 to 2008 (Jiang et al., 2017).In European countries, in contrast, the use of modern technology and other air pollution abatement measures decreased NO 2 concentrations by 10 % to 50 % between 2004 and 2010 (Castellanos and Boersma, 2012) and decreased CO concentrations by 35 % between 200235 % between and 201135 % between (Guerreiro et al., 2014)).Thus, to develop effective air pollution control strategies, accurate information on local emission sources and combustion processes is important (Borsdorff et al., 2018a;Ma and van Aardenne, 2004).However, developing countries and remote areas lack the local infrastructure needed to obtain detailed information about factors such as energy consumption, fuel type, and technology.Limited process information contributes greatly to the uncertainty in emission inventories (Silva and Arellano, 2017).For example, the range of uncertainty in the Chinese NO x and CO emissions between 2005 and 2008 has been estimated to range from −20 % to +45 % due to inadequate information about fuel consumption and uncertain emission factors (Zhao et al., 2011(Zhao et al., , 2012)).In the global emission inventory EDGAR v4.3.2, uncertainties in regional emissions have been estimated to range from 17 % to 69 % for NO x and from 25 % to 64 % for CO (Crippa et al., 2016).In this study, we investigate the use of satellite remote sensing to improve the emission quantification for these important air pollutants.
In global emission inventories, combustion-related emissions are computed as the product of the amount of fuel burned (activity data) and the composition of the emissions as represented by the emission factor (EF; Vallero, 2007).EFs depend strongly on the burning conditions (Sinha et al., 2003;Ward et al., 1996;Yokelson et al., 2003), in particular on the combustion efficiency (CE).The CE is defined as the fraction of reduced carbon in the fuel that is directly converted into CO 2 (Yokelson et al., 1996).Usually, EFs are measured in laboratories under controlled burning conditions.However, in the ambient environment, combustion conditions are highly variable (Andreae and Merlet, 2001;Korontzi et al., 2003) and, therefore, introduce large uncertainties into global emission inventories through the impact of the CE on the EF.A case study (Frey and Zheng, 2002) on NO x emission estimates from coal-fired power plants with dry-bottom wall-fired boilers using low NO x burners showed that the EF for NO x can vary by a factor of 4 or more within a same technology.Thus, the application of mean EFs introduces uncertainties in the range of −29 % to +35 % with respect to mean emission estimates (Frey and Zheng, 2002).Fuel type, fuel composition, combustion practices, and technology are the main factors influencing the CE in the ambient environment (Silva and Arellano, 2017;Tang et al., 2019).To improve the accuracy of global inventories, a better quantification of the CE and EFs is needed.
In recent years, the availability of atmospheric composition measurements from Earth-orbiting satellites has strongly improved.Sensors such as the Scanning Imaging Absorp-tion spectroMeter for Atmospheric Chartography (SCIA-MACHY; Bovensmann et al., 1999) and the Tropospheric Monitoring Instrument (TROPOMI; Veefkind et al., 2012) deliver global datasets of multiple species.The satellite observations from SCIAMACHY have been used in combination with inverse modelling techniques to test and improve emission inventories (Konovalov et al., 2014;Mijling and van der A, 2012;Reuter et al., 2014;Silva et al., 2013).By combining observations of different species (e.g.CO, CO 2 , and NO 2 ), information about common sources and, potentially, information about emission ratios is obtained (Hakkarainen et al., 2015;Miyazaki et al., 2017;Reuter et al., 2019;Silva and Arellano, 2017).
In this study, measurements from TROPOMI are used to investigate the burning efficiency in megacities.TROPOMI is a push-broom grating spectrometer on board Sentinel-5 Precursor (S5P), which was launched by ESA on 13 October 2017 (Veefkind et al., 2012).We use the ratio of the TROPOMI-retrieved tropospheric column of NO 2 and the total column of CO, which is formally not equivalent to the CE but can nevertheless serve as a useful proxy of the burning conditions (Silva and Arellano, 2017;Tang and Arellano, 2017).The reason for this is that the NO x emission increases with combustion temperature, which is high during efficient combustion.In contrast, CO is a product of incomplete combustion and is produced when the CE is low (Flagan and Seinfeld, 1988).The combination of these effects makes the NO 2 /CO ratio highly sensitive to the CE.To correct for differences in the NO 2 and CO background concentrations, the enhancement ratio NO 2 / CO is used.Here NO 2 and CO represent concentration increases compared with their respective backgrounds.
The NO 2 / CO ratio is insensitive to atmospheric transport, as NO 2 and CO emissions are dispersed in a similar manner by the wind.Therefore, the impact of transport cancels out in the ratio.Consequently, TROPOMI-observed ratios close to emissions sources can be directly related to emission ratios.The aim of this study is to investigate the local relation between TROPOMI-retrieved NO 2 / CO ratios and emission ratios in a quantitative manner, focusing on megacities that show significant concentration enhancements in the TROPOMI data.In past studies, NO 2 from the Ozone Monitoring Instrument (OMI) and CO from the Measurement of Pollution in the Troposphere (MOPITT) instrument have been used to derive CO/NO 2 ratios (Silva and Arellano, 2017;Tang and Arellano, 2017).MOPITT also has a short-wave infrared (SWIR) channel (or near IR), and the multispectral (thermal infrared/near-infrared, TIR/NIR) product, with near-surface sensitivity over some land regions, was used in both Silva and Arellano (2017) and Tang and Arellano (2017).TROPOMI provides a unique opportunity to measure CO and NO 2 using the same instrument at an unprecedented high spatial resolution (7 × 7 km 2 at nadir) and daily global coverage (Borsdorff et al., 2018b;van Geffen et al., 2019), making this instrument ideally suited for the investigation of NO 2 /CO ratios from space.Additionally, TROPOMI CO retrievals make use of the SWIR, improving the sensitivity to surface emissions of CO compared with the TIR sounders, MOPITT and the Infrared Atmospheric Sounding Interferometer (IASI).However, TROPOMI NO 2 retrievals are less sensitive to the lower troposphere, causing NO 2 / CO to be influenced by vertical sensitivity (Eskes and Boersma, 2003).We derived a correction factor to take this influence into account, as will be explained in detail in Sect.2.5.
This paper is organized as follows: Sect. 2 provides detailed information about the TROPOMI CO and NO 2 retrieval, the approach used to quantify the NO 2 / CO column enhancement ratio over megacities, and how to relate it to the corresponding emission ratio.Results comparing satellite-derived and emission inventory-derived ratios are presented in Sect.3. Finally, Sect. 4 summarizes our findings and presents the main conclusions.
2 Data and method

TROPOMI CO retrievals
For this study, we use the TROPOMI CO scientific beta data product provided by the SRON Netherlands Institute for Space Research (ftp://ftp.sron.nl/open-access-data-2/TROPOMI/tropomi/co/7_7/, last access: 18 November 2018).The output is identical to that of operational data product provided by the European Space Agency (ESA), but it also provides the TM5 a priori profiles (http://tm5.sourceforge.net/,last access: 18 November 2018) that are used in the retrieval.The SRON CO product also supplies more data for the early months of the mission which are not included in the operational product.Total column densities of CO (molec.cm −2 ) are retrieved from spectral radiance measurements from the TROPOMI SWIR module at 2.3 µm using the SICOR algorithm (Landgraf et al., 2018).In this profile scaling algorithm, the TROPOMI-observed spectra are fitted by scaling a reference vertical profile of CO using the Tikhonov regularization technique (Borsdorff et al., 2014).The reference a priori CO profile is derived from the TM5 transport model (Krol et al., 2005), as described in Landgraf et al. (2016).The averaging kernel (A) is an essential component of the CO retrieval and quantifies the sensitivity of the retrieved CO column to a change in the true vertical profile (ρ true ) following Borsdorff et al. (2018c): where CO is the error in the retrieved CO columns.

TROPOMI NO 2 retrievals
The UV-Vis module of TROPOMI is used to retrieve NO 2 in the 405-465 nm spectral range.NO 2 slant column den-sities are processed using the TROPOMI NO 2 DOAS software developed at the Royal Netherlands Meteorological Institute (KNMI; van Geffen et al., 2019).The retrieval algorithm is based on the NO 2 DOMINO algorithm (Boersma et al., 2011) which has been improved further in the QA4ECV4 project (Boersma et al., 2018).The algorithm subtracts the stratospheric contribution to the slant column densities and then converts the residual tropospheric slant column density into the tropospheric vertical density via the air mass factor (AMF).The AMF is computed using co-sampled, daily NO 2 a priori vertical profiles from output of the TM5-MP chemistry transport model at a 1 • × 1 • resolution (Williams et al., 2017).The AMF depends on the surface albedo, terrain height, cloud height, and cloud fraction (Eskes et al., 2019;Lorente et al., 2017).We used the offline Level 2 NO 2 data (mol m −2 ) that are available at https://s5phub.copernicus.euand http://www.tropomi.eu(last access: 18 November 2018).The TROPOMI NO 2 product has been successfully used in various other studies (Griffin et al., 2019;Reuter et al., 2019); however, there are indications that NO 2 is biased low by approximately 30 % in the tropospheric columns due to issues with the cloud pressure and the a priori NO 2 profile used in the AMF calculation (Boersma et al., 2004;Lorente et al., 2017).

Data selection
We used TROPOMI CO and NO 2 retrievals from June to August 2018 due to the large number of clear-sky days during this period over the megacities of interest.Megacities are strong sources of air pollution and can readily be observed in TROPOMI data (Borsdorff et al., 2018c).Since CO and NO 2 are retrieved from different instrument channels using different algorithms, the filtering criteria and spatial resolutions are also different.To facilitate data filtering, both algorithms provide a quality assurance value (q a value).The q a value for both products ranges from 0 (no data) to 1 (high-quality data) For our data analysis, we selected NO 2 retrievals with q a values equal to or larger than 0.75, indicating clear-sky conditions (Eskes and Eichmann, 2019), and CO retrievals with q a values equal to or larger than 0.7, representing measurements under clear-sky conditions or the presence of lowlevel clouds (Apituley et al., 2018).The application of the SICOR algorithm to SCIAMACHY CO retrievals with lowlevel clouds increases the number of measurement, with a limited impact in the ability to detect CO sources (Borsdorff et al., 2018a).CO retrievals are filtered for stripes, as described in Borsdorff et al. (2018c).The CO retrieval has a spatial resolution that is a factor of 2 coarser than the NO 2 retrieval (7 × 7 km 2 versus 3.5 × 7 km 2 ).To co-locate NO 2 and CO retrievals, we combine the NO 2 pixels with centres that fall within a CO pixel, selecting only those pixels for which both the NO 2 and CO retrievals pass the filtering criteria.The total CO column and tropospheric NO 2 columns are converted into the dry column mixing ratio XCO (ppb) https://doi.org/10.5194/acp-20-10295-2020Atmos.Chem.Phys., 20, 10295-10310, 2020 S. Lama et al.: Quantifying burning efficiency in megacities and XNO 2 (ppb) using the dry-air column density calculated using the co-located surface pressure data included in the CO data files, as described in Borsdorff et al. (2018c).

Calculation of NO 2 /CO
This study focuses on the following megacities (with populations exceeding 5 million): Mexico City, Tehran, Riyadh, Cairo, Lahore, and Los Angeles.These six megacities are well isolated from surrounding sources and frequently experience cloud-free conditions, allowing for the retrieval of a large number of XCO and XNO 2 data from TROPOMI.Los Angeles and Mexico City have automated air quality monitoring networks that measure CO and NO 2 at different locations in the city.These measurements are used in Sect.3.3 to validate the results obtained using TROPOMI.In addition, these megacities are expected to span a sizable range of burning efficiencies, as they include urban centres in developed (Los Angeles, USA) and developing countries (Mexico City, Mexico; Cairo, Egypt; Riyadh, Saudi Arabia; and Lahore, Pakistan).
The concentration gradient between the background and the city centre is used to determine the XNO 2 / XCO enhancement ratio.To determine this ratio, we divide each city into a core city area and a background area.Every city has a different size and different neighbouring CO and NO 2 emission sources; therefore, the appropriate choice of radii for the background and outskirts varies between cities (see Sect.S1 in the Supplement for details).Since the same regional definition is used for NO 2 and CO, the enhancement ratio is not sensitive to the details of the selection of the region.Thus, capturing the local enhancement in CO and NO 2 to its full extent is most important for the choice of radii in order to optimize the signal over noise and, in turn, the detection limit for urban emissions.To maximize the size of the city enhancement, we exclude the diffuse outskirts between the city centre and the background.For the location of the city centre, we use the weighted average emission centre of NO 2 derived from the EDGAR emission database (Dekker et al., 2017).The derived centre coordinates as well as the radii of the city core and background area are listed in Table 1.We test the robustness of the satellite-derived emission ratio using two different methods, which are explained in detail below.

Upwind background
To determine the upwind background (UB) column mixing ratio, we select a section of the background region that is upwind from the city centre using the average wind direction over the core city area (see Figs. 1 and S7 for further details).Generally, more than 75 % of all pollutants are emitted between the surface and an altitude of 200 m (Bieser et al., 2011).Therefore, the average wind speed and direction from the surface to an altitude of 200 m are derived from the ERA-Interim reanalysis data, which are pro-vided at a 0.75 • × 0.75 • spatial and 3-hourly temporal resolution.The wind vector components of ERA-Interim are spatially and temporally interpolated to the central coordinate of TROPOMI pixels.Using this information, daily enhancement ratios are calculated as follows: The background area might contain free tropospheric NO 2 from lightning and convectively lofted surface NO 2 from elsewhere.However, these contributions vary on scales that are usually large compared with the scale of a city.Therefore, the calculated XNO 2 and XCO enhancements are predominantly caused by emissions from the city.

Plume rotation
The daily TROPOMI-observed city images are rotated in the direction of the wind using the city centre as the rotation point to align each CO and NO 2 plume in the upwinddownwind direction (Pommier et al., 2013).Rotated images for June to August 2018 are averaged together (see Fig. S8).XNO 2 and XCO are determined by subtracting the average of the first quartile XNO 2 , XCO values in a 100 km × 20 km region upwind from the city centre from the average of the fourth quartile XNO 2 , XCO values in a 100 km × 20 km region downwind from the city centre.Finally, the enhancement of XNO 2 and XCO is calculated as described in Eq. ( 5), and the enhancement ratio is derived using Eq. ( 4).
The downwind-upwind difference where n downwind refers to the number of observations ≥ 75th percentile, and n upwind refers to the number of observations ≤ 25th percentile

NO 2 /CO emission ratio
Local TROPOMI-derived ratios in column abundance are compared with emission ratios derived from the Emission Database for Global Atmospheric Research (EDGAR v4.3.2) at a 0.1  2011).MACCity has been re-gridded to a spatial resolution of 0.1 • × 0.1 • , assuming a uniform distribution of the emissions within each 0.5 • × 0.5 • grid box.Both emission inventories contain total emissions of NO x and CO.NO x emissions are converted into NO 2 by dividing NO x by the conversion factor of 1.32.This conversion factor is based on Seinfeld and Pandis (2006) and represents urban plumes at 13:30 local time (LT).The emission ratios of NO 2 and CO (E NO 2 /E CO ) are calculated from total emissions (the sum of all processes) within the core city area for the EDGAR and MACCity emission inventories.
To compare TROPOMI to inventory-derived ratios, the NO 2 tropospheric column has to be corrected for its limited atmospheric residence time.The CO lifetime is long enough compared with the transport time out of the city domain to be neglected.In addition, we need to account for differences in the vertical sensitivity of TROPOMI to NO 2 and CO, as quantified by their respective averaging kernels (A) shown in Fig. 2. To compare TROPOMI to EDGAR and MACCity, we formulate a relationship between the emission ratio (E NO 2 /E CO ) and the column enhancement ratio ( XNO 2 / XCO) taking the combined effect of atmospheric transport, chemical loss, and the averaging kernel into account.This relationship is as follows (see Appendix A for its derivation): where U is the is the 200 m wind speed (m s −1 ), l x is the diameter of the city centre (m), and K is the rate constant of the reaction of NO 2 with OH of 2.8 × 10 −11 T 300 −1.3 cm 3 molec.−1 s −1 (Burkholder et al., 2015).T (K) and OH (molec.cm −3 ) are the boundary layer average temperature and OH concentration respectively, and A influence is the influence of the averaging kernel on XNO 2 / XCO (see Sect. 3.2).
Copernicus Atmospheric Monitoring Service (CAMS) real-time OH, CO, and NO 2 fields are used to account for the impacts of chemical loss and the averaging kernel.The CAMS data, at a 0.1 • × 0.1 • and 3-hourly resolution, are spatially and temporally interpolated to the TROPOMI footprints.The CAMS CO and NO 2 vertical mixing ratio profiles are converted into vertical column densities using the ERA-Interim reanalysis surface pressure.For CO, the TROPOMI data provide column A values from the surface to the top of atmosphere.For NO 2 , tropospheric A is derived using the AMF for the troposphere as fraction of the total column (Boersma et al., 2016).For further details, see Appendix B.

Uncertainty
To quantify the uncertainty in TROPOMI-derived XNO 2 / XCO ratios for the plume rotation method, we use the error propagation method of Pommier et al. (2013) and bootstrap for the upwind background, as explained further in the following.

Bootstrapping
The bootstrapping method is a statistical resampling method that is used here to calculate the uncertainty in the daily enhancement ratio of XNO 2  XCO .The first step is to generate a new set of samples by drawing a random subset with replacement from the full dataset of N daily XNO 2 XCO ratios.The subset has the same number of samples as the full dataset, from which a mean ratio is calculated.This procedure is repeated 1000 times for each city.Finally, the standard deviation of the resulting ratios is taken and used to represent the uncertainty in the daily XNO 2 XCO .

Error propagation
To calculate the uncertainty in XNO 2 XCO by error propagation, we first determine the uncertainty in the enhancements XNO 2 and XCO, which are derived from the uncertainty in the mixing ratios upwind and downwind of the source as follows: where X is XNO 2 or XCO.
Here, we assume that the upwind and downwind uncertainties are independent.The uncertainty for the column enhancement is 3 Results and discussion 3.1 Detection of NO 2 and CO pollution over megacities The co-located TROPOMI XNO 2 and XCO data have been averaged for June to August 2018 for domains of 500 × 500 km 2 centred around the selected megacities, as described in Sect. 2. The results for Mexico City and Cairo are shown in Fig. 3.The enhancements of XCO and XNO 2 over Mexico City and Cairo are clearly separated from the surrounding background areas and are prominent in several overpasses of TROPOMI (Fig. S9).This demonstrates that a relatively short data averaging period is sufficient for TROPOMI to detect hotspots of CO pollution at the scale of large cities, compared with instruments such as IASI and MOPITT.The orography surrounding Mexico City causes pollutants to become trapped, thereby facilitating detection by TROPOMI.The longer lifetime of CO compared with NO 2 causes the urban influence of CO to be propagated further in the westward direction.As can be seen in Fig. 3, the retrieved XCO and XNO 2 signals of emissions from Mexico City and Cairo correlate quite well with each other, confirming that it should be possible to obtain useful information about burning efficiency by studying XNO 2 XCO .An industrial area is located to the east of Cairo (29.797351 • N, 32.148266 • E), showing a clear enhancement in XNO 2 but not in XCO (Fig. 3c, d).It demonstrates that variations in the column enhancement ratio can already be seen by eye when comparing TROPOMI-retrieved XCO and XNO 2 images.

Comparison of TROPOMI-and inventory-derived ratios
Here, we attempt to compare TROPOMI-derived NO 2 /CO column enhancement ratios to emission ratios from EDGAR and MACCity for the six selected megacities (see Fig. 4).
As explained in Sect.2, column enhancement ratios from TROPOMI are obtained using the upwind background (UB) and plume rotation (PR) methods.These estimates differ by 5 % to 20 % across the six cities, providing an initial estimate of the accuracy with which the column enhancement ratio can be derived (see Table S1 in the Supplement for details).The EDGAR and MACCity inventories show a substantial variation in emission ratios between cities, with relatively high emission ratios for Riyadh and the lowest emission ratios for Lahore.TROPOMI-derived XNO 2 / XCO column enhancement ratios for the UB and PR methods show similar patterns to EDGAR and MACCity, with Pearson correlation coefficients of 0.85 and 0.7 respectively (Fig. S10 in the Supplement).However, inventory-derived emission ratios are clearly 60 % to 85 % larger than TROPOMI-derived column enhancement ratios, which is largely explained by the impact of the limited NO 2 lifetime and the averaging kernel, as will be discussed further after an explanation of the differences between EDGAR and MACCity.Emission ratios from MACCity are 10 % to 75 % lower than those from EDGAR.To understand the differences in emission ratios between MACCity and EDGAR, we selected two cities, Cairo and Mexico City, which present the largest and smallest differences in the emission ratio.The CO and NO 2 emissions are categorized into seven sectors: agriculture, residential, energy, industrial, transportation, shipping, and waste treatment.Sectors that contribute most to the total emission are compared.In the case of Cairo and Mexico City, these are the transportation, industrial, energy, and residential sectors (Fig. S11a, b).For Cairo, the total CO emission is a factor of 2 lower in EDGAR than in MACCity, whereas the total NO 2 emission is 10 % higher in EDGAR.This results in an emission ratio that is a factor of 3 higher.The largest discrepancy between EDGAR and MACCity CO emission is due to the residential sector, followed by energy.For NO 2 , the energy, transportation, and residential sectors explain most of the difference between EDGAR and MACCity.In Mexico City, EDGAR total CO and NO 2 emissions are both a factor of 2 higher than MACCity values; thus, the total emission values cancel out in the ratio, leading to the best agreement of all selected megacities.However, it is complicated to identify the main factors explaining the differences be-tween EDGAR and MACCity at the sector level due to the combined influence of differences in activity data, EFs, and the methods used to disaggregate country totals.To understand the disaggregation of emission in EDGAR and MAC-City, we compared the country total CO and NO 2 of Mexico City (Mexico) and Cairo (Egypt).The comparison shows that the EDGAR and MACCity country CO total and the NO 2 total for Mexico show a small difference (∼ 12 %), whereas the difference is about factor of 2 in Mexico City (Fig. S11c).For Egypt, the EDGAR and MACCity CO total shows a similar difference to Cairo, whereas the EDGAR NO 2 country total emission value is a factor of 2 lower (Fig. S11d).This shows that EDGAR attributes CO and NO 2 emissions to the city, whereas MACCity smears them out over the country.
The difference between satellite-derived column enhancement ratios and inventory-based emission ratios can be explained in part by the relatively short lifetime of NO 2 that reduces columnar NO 2 /CO ratios compared with the emissions.In addition, the sensitivity to the planetary boundary layer is lower for NO 2 than for CO TROPOMI measurements, further reducing the satellite-observed column enhancement ratio.Taking these influences into account using Eq. ( 6) leads to the upwind background-corrected emission ratio (UBCER) and the plume rotation-corrected emission ratio (PRCER) in Fig. 4, which have been calculated on a daily basis before averaging over the full period.Due to the short lifetime of OH, its concentration depends strongly on the local photochemical conditions (de Gouw et al., 2019).Therefore, to account for the local lifetime of NO 2 , we need an https://doi.org/10.5194/acp-20-10295-2020Atmos.Chem.Phys., 20, 10295-10310, 2020 Figure 4. Comparison of TROPOMI-derived NO 2 / CO enhancement ratios, calculated using the different methods shown in shades of blue, with the corresponding emission ratios from the EDGAR (shown using shades of red) and MACCity (shown using shades yellow) emission inventories for six megacities.Dark solid shades for emission inventories represent the annual average inventory-derived ratio, whereas the lighter shades represent the June to August averaged inventory-derived ratios.The upwind background-corrected emission ratio (UBCER) and the plume rotation-corrected emission ratio (PRCER) account for the impact of photochemical NO 2 removal and the averaging kernel.Error bars for the TROPOMI-derived NO 2 / CO enhancement ratios represent 1σ uncertainties calculated using bootstrapping (upwind background) and error propagation (plume rotation method).The error bars for UBCER and PRCER account for the uncertainty in the methodology and TROPOMI data (for details, see Table S3).
estimate of the OH that is representative for the photochemical conditions inside cities. Figure 5 shows the boundary layer OH concentration at the TROPOMI overpass time from CAMS for Mexico City, averaged over June-August 2018.
Figure 5 shows a clear enhancement of OH in the city centre, confirming that the spatial resolution of CAMS is sufficient to resolve the urban influences on OH in megacities.The UB and PR column enhancement ratios increase by 60 % to 85 % when accounting for the NO 2 lifetime (see Table S1).The boundary layer OH concentrations and mean wind speeds for the six cities are listed in Table 2.The impact of differences between the XNO 2 and XCO averaging kernels is calculated using vertical profiles of NO 2 and CO taken from CAMS.These profiles were used to calculate XNO 2 and XCO using either the TROPOMI A values or A values replaced by identity matrices.The relative difference • 100 % quantifies the impact of differences between the averaging kernels (see Appendix C for the derivation).The CAMS-simulated city enhancements for CO from June to August 2018 did not compare well with TROPOMI for Tehran, Cairo, Riyadh, and Lahore, which was possibly due to the coarse resolution of CAMS (see Figs. S14, S15, S16, and S17).Therefore, A influence has been determined for Mexico City and Los Angeles to calculate the averaging kernel impact (Figs.S12, S13).To test the accuracy of A influence , a few days were selected for Tehran, Cairo, Riyadh, and Lahore when CAMS CO and NO 2 enhancements compared relatively well with TROPOMI.For the six megacities, TROPOMI-derived NO 2 / CO ratios are 10 % to 15 % lower than the "ideal" NO 2 / CO ratio that would be measured if both retrievals had uniform vertical sensitivities, i.e. every molecule in the column received equal weight.Details about the selected days and the corrections calculated for each city are listed in Table S2.
After correction, UBCER and PRCER for Tehran and Mexico City are close to EDGAR and MACCity (10 % to 25 %).This confirms that the EFs for these cities are well represented in the EDGAR and MACCity emission inventories.The difference between corrected and uncorrected ratios in Fig. 4 highlights the importance of the correction, in particular the influence of OH, for assessing emission ratios using TROPOMI.For Cairo the correction also reduces the difference between TROPOMI and the emission inventories, although the EDGAR ratios remain about 65 % higher for Cairo than UBCER and PRCER.For MACCity, the emission ratios are close to the TROPOMI-derived UBCER and PRCER for Cairo (within 20 %), pointing to a more accurate representation of emission ratios in MACCity than in EDGAR.For Riyadh, UBCER and PRCER are close to MACCity (∼ 10 % to 20 %), whereas EDGAR is 35 % higher.However, for Lahore, PRCER is close to the EDGAR ratio, whereas MACCity is a factor of 2.5 lower.For Los Angeles, the ratios from EDGAR and MACCity are 55 % and 70 % higher than UBCER and PRCER after correction respectively, suggesting poorer burning conditions than represented by the emission inventories.To further investigate this discrepancy for Los Angeles, we included the Hemispheric Transport of Air pollution version 2 (HTAP-v2) emission inventories for 2010 in the comparison.HTAP-v2 has a resolution of 0.1 • × 0.1 • and makes use of emission estimates from the Environmental Protection Agency (EPA) for the USA (Janssens-Maenhout et al., 2015).The HTAP-v2-derived emission ratio over Los Angeles is 0.074, which is close to UBCER and PRCER (within 20 %).This result provides further confidence in the TROPOMI-derived emission ratio.However, different sources of uncertainty play a role, as discussed further below.
Seasonal variations in EFs may influence our comparison between the seasonal averaged TROPOMI data and annual average EDGAR emissions.To account for the influence of seasonally varying EFs, we compute a seasonal correction factor based on EDGAR v4.3.2 2010, as monthly data are not available for EDGAR 2012 (see Fig. 4).Except for Lahore, the June to August (JJA) EDGAR ratio is 5 % to 12.5 % lower than the annual average EDGAR ratio.The MACCity ratio for JJA, however, is 10 % to 71 % higher than the annual average, indicating that EDGAR and MACCity disagree on the seasonality of the NO 2 /CO emission ratio.For MACCity, the agreement with TROPOMI improves the most when seasonality is taken into account (see Fig. 4).
The ozone concentration and the photolysis rate impact the partitioning of NO and NO 2 (Jacob, 1999), thereby influencing the applied conversion factor of 1.32.To further investigate the uncertainty introduced by this factor, we analysed CAMS surface NO and NO 2 at the TROPOMI overpass time (see Table 2).The CAMS-derived conversion factor varies by less than 10 % compared with the standard value of 1.32, introducing an uncertainty of less than 10 % in the inventory-derived emission ratio.However, given the uncertainty in the CAMS-simulated urban NO, NO 2 , and OH concentrations (Huijnen et al., 2019), the actual uncertainty is probably higher.Additionally, TROPOMI underestimates the NO 2 column by 7 % to 29.7 % relative to MAX-DOAS ground-based measurement in European cities (Lambert, et al., 2019).However, as we currently do not know how representative this estimate is for the cities studied, the impact of this bias has been accounted for as an additional source of uncertainty of 30 % in the TROPOMI-inferred NO 2 /CO ratio (see Table S3).Compared with this number, other sources of uncertainty, such as the wind direction and speed (Figs.S18, S19), the boundary layer OH concentration (Table 2), the A influence correction (Table S2), and the predefined background setting (Fig. S20), only make small contributions to the TROPOMI-derived emission ratio.The total uncertainty in the TROPOMI-derived emission ratio is calculated using error propagation (see Table S3) and ranges between 33 % and 35.6 %.
We also acknowledge that our treatment of the photochemical removal of NO 2 is simplified.In reality, NO 2 is influenced by several other factors including meteorological parameters, such as temperature, wind speed, and radiation (Lang et al., 2015;Romer et al., 2018), causing the formation and loss of NO 2 to vary spatially and temporally.In the corrected ratio, we only consider the first-order loss of NO 2 by OH forming HNO 3 .Several studies show that the loss of NO 2 via the formation of alkyl and multifunctional nitrates (RONO 2 ) can play a more important role than nitric acid production in cities surrounded by forested areas (Browne et al., 2013;Farmer et al., 2011;Romer Present et al., 2020;Sobanski et al., 2017).In addition, the secondary production of CO from volatile organic compound (VOC) oxidation may play a role.However, this only affects our ratios if it changes the CO gradient between the city and the background.Hence, to further improve the accuracy of the TROPOMI-supported evaluation of emission ratios, a more sophisticated treatment of urban photochemistry is required.The validation results are presented in Fig. 6 for spatially averaged, hourly CO and NO 2 measurements for Mexico City and Los Angeles collected during the noon hours (12:00 to 14:00 LT).To determine the enhancement in CO and NO 2 due to local emissions for each ground-based station, the fifth percentile of hourly CO and NO 2 measurements is used as background.CO and NO 2 enhancements for individual monitoring stations are calculated as X = X individual − X background .For comparison with TROPOMI, all measurement sites are spatially averaged.

Validation using ground-based measurements
Ground-based CO and NO 2 at Mexico City and Los Angeles are strongly correlated, with a Pearson correlation coefficient of r = 0.95 and 0.80 respectively, confirming that the observed signals reflect NO 2 and CO emissions from common sources.The slope of the regression line for Mexico City is 0.048, which is 45 % higher than the TROPOMIderived column enhancement ratios using the UB and PR methods.The NO 2 / CO ratio that is observed at ground level is likely less influenced by the photochemical removal of NO 2 than the TROPOMI-retrieved columns and is, therefore, closer to the inventory-derived ratio, which is consistent with our results.This comparison suggests that the removal of NO 2 reduces the ratio for ground-based measurements by 35 % compared with EDGAR and MACCity.Overall, the emission ratios in EDGAR and MACCity for Mexico City are consistent with both the ground-based measurements and TROPOMI, i.e. within the uncertainty introduced by the chemical removal of NO 2 .
For Los Angeles, the regression slope is 0.042, which is 10 % to 20 % larger than the TROPOMI-derived column enhancement ratios using the UB and PR methods.However, the EDGAR and MACCity ratios are a factor of 5 higher than the NO 2 / CO ratio observed at ground level.The ground-based measurements point to similar ratios for Mexico City and Los Angeles, confirming the HTAP-v2supported TROPOMI finding that the emission ratios in EDGAR and MACCity are too high for Los Angeles.There- fore, the ground-based measurements for Los Angeles provide independent support for the TROPOMI-derived ratios, pointing to poorer burning conditions in Los Angeles than indicated by the emission inventories, and confirm the value of TROPOMI with respect to monitoring the burning efficiency of megacities.

Conclusions
In this study, we investigate the use of TROPOMI XCO and XNO 2 retrievals for monitoring the burning efficiency of fossil fuel use in megacities.To improve the accuracy of the global emission inventories, the burning efficiency and the EF are quantified using co-located XCO and XNO 2 enhancements over the megacities of Tehran, Mexico City, Cairo, Riyadh, Lahore, and Los Angeles.TROPOMI is very capable of detecting XCO and XNO 2 enhancements over these megacities with a relatively short averaging time and shows the expected spatial correlation.
TROPOMI-derived column enhancement ratios have been compared with emission ratios from EDGAR and MAC-City.The TROPOMI-derived column enhancement ratios are strongly correlated with the EDGAR and MACCity inventory-derived emission ratios (r = 0.85 and 0.7 respectively), showing the highest emission ratio for Riyadh and the lowest emission ratio for Lahore.This shows that Lahore has the poorest burning efficiency, whereas fossil fuel burning is the most efficient over Riyadh (of all megacities that were analysed).The impact of the short NO 2 lifetime and differences in the vertical sensitivity of the TROPOMI XCO and XNO 2 retrieval on the NO 2 / CO enhancement ratio has been quantified.Correcting for these factors significantly improves the agreement between ratios derived from TROPOMI and emission inventories.The comparison indicates that the emission ratios in MACCity and EDGAR are well represented for Mexico City and Tehran.For Lahore, the EDGAR emission ratio agrees better with TROPOMI, whereas the MACCity emission ratios are closest to the TROPOMI-derived emission ratios for Cairo and Riyadh.Emission ratios in EDGAR and MACCity are significantly higher (by 55 % to 70 %) than TROPOMI for Los Angeles.The total uncertainty on TROPOMI-derived emission ratios ranges from 33 % to 35.6 %.The bias in S5P TROPOMI NO 2 retrievals has the most important contribution to the uncertainty in the TROPOMI-derived emission ratio.TROPOMI-derived XNO 2 / XCO column enhancement ratios for Mexico City and Los Angeles have been validated using ground-based measurement from local air quality monitoring networks.For Mexico City, the enhancement ratio derived from ground-based measurements is consistent with EDGAR, MACCity, and the TROPOMI-derived emission ratio.For Los Angeles, TROPOMI-derived enhancement ratios are consistent with the ground-based measurements as well as the HTAP-v2 inventory based on EPA statistics, whereas EDGAR-and MACCity-derived emission ratios appear to be overestimated by a factor of 5 compared with ground-based measurements.This demonstrates the potential of TROPOMI data for monitoring burning efficiency and evaluating emission inventories.

Figure 1 .
Figure 1.Average wind speed and direction from the surface to 200 m from ERA-Interim at the TROPOMI overpass time (a), and the TROPOMI-derived total column CO over Mexico City (b) for 4 June 2018.The black star represents the centre of the city.In the right panel, the white circle is the background area for Mexico City, and the blue section represents the upwind background area that we selected depending upon the wind direction (θ ) in the core city area.P0, P1, P2, and P3 are the points where the northern, eastern, western, and southern wind directions intersect with the background area.P0 new is the new point generated by rotating P0 with θ in reference to the city centre.

Figure 2 .
Figure 2. TROPOMI averaging kernels (A) for CO total column and tropospheric NO 2 over Mexico City on 1 June 2018.The error bars represent the standard deviation of the mean A at each vertical level.

Figure 3 .
Figure 3. Co-located TROPOMI-retrieved XNO 2 (a, c) and XCO (b, d) data over Mexico (a, b) and Cairo (c, d) averaged for June to August 2018.De-striping is applied to CO total columns (Borsdorff et al., 2018b), and CO and NO 2 retrievals have been re-gridded to 0.1 • × 0.1 • .The white stars represent the centres of Mexico City and Cairo respectively.The red circle in (c) and (d) points to an industrial area east of Cairo.

Figure 5 .
Figure 5.The boundary layer average OH concentration at the TROPOMI overpass time during June to August 2018 over Mexico City.The white star represents the centre of Mexico City.

Figure 6 .
Figure 6.The relation between NO 2 and CO in surface measurements from Mexico (a) and Los Angeles (b).The red dots represent spatially averaged hourly measurements collected during the day (12:00 to 14:00 LT).

Table 1 .
Selected megacities and specifications used for emission ratio quantification.
• × 0.1 • spatial resolution for the most recent year of 2012 and the database provided by Monitoring Atmospheric Chemistry and Climate and CityZen (MACCity) for 2018 that is available at a 0.5 • × 0.5 • resolution(Granier et al., To further evaluate TROPOMI's ability to quantify burning efficiencies, TROPOMI-derived XNO 2 / XCO ratios have been compared with ground-based measurements from Mexico City and Los Angeles.For this purpose, 20 ground-based stations in Mexico City with hourly measurements of CO and NO 2 have been selected from the AIRE CDMX network (http://www.aire.cdmx.gob.mx/, last access: 17 July 2019).Similarly, 12 ground-based stations from the South Coast Air Quality Management District (AQMD) monitoring network (http://www.aqmd.gov/,lastaccess:20 July 2019) have been selected for Los Angeles.For details (names and locations) on these sites, see TableS4.For Mexico City, data were only available for June 2018.For Los Angeles, data for the June to https://doi.org/10.5194/acp-20-10295-2020Atmos.Chem.Phys., 20, 10295-10310, 2020

Table 2 .
Average wind speed and boundary layer CAMS OH concentration for June-August 2018 that was used to correct for the limited lifetime of NO 2 .The errors presented represent the 1σ uncertainty calculated by the bootstrapping method.