NO2 can be observed from space with satellite instruments based on its strong absorption features in the 400–465 nm wavelength region . By comparing observed spectra with a reference spectrum, the amount of NO2 in a portion of the atmosphere between the instrument and the surface can be derived. The TROPOspheric Monitoring Instrument (TROPOMI), on board the European Space Agency’s (ESA) Sentinel-5 Precursor (S-5P) satellite, is one of those instruments, providing daily measurements of NO2 around 13:30 local time (LT; the time zone for all instances in the text is LT). Its high spatial resolution (originally 3.5×7 km2 at nadir, improved to 3.5×5.5 km2 since 6 August 2019) allows one to observe some of the fine-scale structures of NO2 pollution, such as within cities or near power plants and industrial facilities . Tropospheric vertical column densities (VCDs or simply columns) are provided by the algorithm after measurement by the instrument, which represents the vertically integrated number of NO2 molecules per surface unit between the surface and the tropopause. An algorithm also provides an air mass factor (AMF), which is used to convert slant column densities into vertical column densities. This factor depends on many parameters, including the albedo of the viewed surface, the vertical distribution of the absorber and the viewing geometry. It is a source of structural uncertainty in NO2 measurements , which becomes non-negligible in polluted environments. The large swath width of the instrument (∼ 2600 km) makes it possible to construct NO2 images of VCDs on large spatial scales. We use TROPOMI NO2 retrievals from 2019 to 2022 (S5P-PAL reprocessed data with processor version 2.3.1 from January 2019 to October 2021, offline (OFFL) stream with processor version 2.3.1 from November 2021 to October 2022 and OFFL stream with processor version 2.4.0 from November to December 2022) over Qatar. The arid climate of the country, which offers a large number of clear-sky days throughout the year, enables the calculation of monthly averages based on multiple observations. Its intensive anthropogenic activity, concentrated on a small number of areas, allows one to observe high NO2 concentration patterns. Due to its short lifetime (about 1 to 10 h), background NO2 levels can be orders of magnitude lower than levels near polluted areas. Pollution patterns are therefore characterized by large signal-to-noise ratios above main emitters. Finally, TROPOMI products provide a quality assurance value qa, which ranges from 0 (no data) to 1 (high-quality data). For our analysis of concentrations, we selected NO2 retrievals with qa values greater than 0.75, which systematically correspond to clear-sky conditions , and gridded these retrievals at a spatial resolution of 0.0625∘×0.0625∘. From August 2019, this resolution is lower than that of the instrument, thus providing a grid for which NO2 VCDs correspond to one or more measurements. The observed plumes remain correctly resolved.

The domain of interest extends between the between the parallels of 24 and 27∘ N and the meridians of 50 and 52∘ E. TROPOMI tropospheric columns show polluted areas, such as the Ras Laffan area, but also the large urban area of Doha. Observed VCDs are high (5 to 30×1015 molec. cm-2), which facilitates the observation of NO2 plumes. An example of a TROPOMI image is provided in Fig. . Within the domain, the meridional component of the wind is generally oriented towards the south. The zonal component varies throughout the year. Strong wind events can cause the pollution from Bahrain to reach the coasts of Qatar. Finally, several areas in the west of the country are not visible by TROPOMI most of the time because the quality of the corresponding measurements is too low. This phenomenon, which is also present in the east of the country (slightly south of downtown Doha) but with a smaller spatial extent, is studied in detail in Sect. 5.3.

The horizontal wind vector field w=(u,v) is taken from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 data archive (fifth generation of atmospheric reanalyses) at a horizontal resolution of 0.25∘×0.25∘ on 37 pressure levels . The hourly values have been linearly interpolated to the TROPOMI orbit timestamp and re-gridded to a 0.0625∘×0.0625∘ resolution.

The Copernicus Atmospheric Monitoring Service (CAMS) global near-real-time service provides analyses and forecasts for reactive gases, greenhouse gases and aerosols. Data are gridded on 25 vertical pressure levels with a horizontal resolution of 0.4∘×0.4∘ and a temporal resolution of 3 h . Here, CAMS concentration fields of hydroxyl radical (OH) are used to calculate NO2 sinks from TROPOMI observations. We also use CAMS temperature field T to account for the reaction rate for HNO3 production through the calculation of the NO2 lifetime according to . The hourly values are also linearly interpolated to the TROPOMI orbit timestamp and re-gridded to a 0.0625∘×0.0625∘ resolution.

The Emissions Database for Global Atmospheric Research (EDGARv6.1) and the CAMS global anthropogenic emissions (CAMS-GLOB-ANT_v5.3) are global inventories that provide 0.1∘×0.1∘ gridded emissions for different sectors on a monthly basis. EDGARv6.1 emissions are based on activity data (population, energy production, fossil fuel extraction, industrial processes, agricultural statistics, etc.) derived from the International Energy Agency (IEA) and the Food and Agriculture Organization (FAO), corresponding emission factors, national and regional information on technology mix data, and end-of-pipe measurements. The inventory covers the years 1970–2018. CAMS-GLOB-ANT_v5.3 is developed within the framework of the Copernicus Atmospheric Monitoring Service . For this inventory, NOx emissions are based on various sectors in the EDGARv5.0 emissions up to 2015, which are extrapolated to 2021 using sectorial trends from the Community Emissions Data System (CEDS) inventory up to 2019. From one inventory to another, the names and definitions of the sectors may differ. In EDGARv6.1 and CAMS-GLOB-ANT_v5.3, the emissions for a given country are derived from the type of technologies used, the dependence of emission factors on fuel type, combustion conditions, and activity data and low-resolution emission factors .

As the power sector is one of the main drivers of NOx, we use electricity production and consumption data at several timescales, as detailed in Sect. 5.5. Daily load profiles from February 2016 to January 2017 are taken from and used to calculate monthly ratios between the average power demand and the power demand during the overpass of TROPOMI. These ratios are assumed to be valid from 2019 to 2022. We also use monthly electricity generation time series from 2019 to 2022, which are provided by the Qatar Ministry of Development Planning and Statistics . From 2019 to 2021, these time series can be completed with monthly reports from Kahramaa, which transmits and distributes the electricity generated by each of the country's power plants with great accuracy . The corresponding data for 2022 are not available yet.

In satellite retrievals, the NO2 signal from a sparsely populated area or a small industrial facility may be difficult to identify due to high noise levels or natural emissions. As a consequence, detecting traces of non-natural emissions in TROPOMI NO2 images is not a straightforward process. In the absence of anthropogenic sources, the NO2 columns that are observed constitute a tropospheric background that varies between 0.2 and 1.0×1015 molec. cm-2. At the global scale, this background is mostly due to soil emissions in the lower troposphere . In the upper troposphere, NO2 sources include lightning, convective injection and downwelling from the stratosphere , but the factors controlling the resulting concentrations are poorly understood. According to state-of-art estimates, anthropogenic NOx accounts for most of the emissions at the global scale, whereas natural emissions from fires, soils and lightning are smaller . In Qatar, the desert climate limits lightning and fire, and the share of the corresponding emissions within the background is thus expected to be low. As a first step, we estimate this background by excluding the part of the domain which is outside the territory of Qatar, as well as other regions with human activity, which include Bahrain and the neighbouring part of Saudi Arabia. The remaining pixels constitute an external mask within which we calculate the 5th percentile of the TROPOMI columns. The corresponding value is defined as the tropospheric background. Figure  displays the external mask, as well as an internal mask that gathers only pixels above Qatar's territory. This external mask is used in Sects. 5 and 6.

As a second step, we derive top-down NO2 production maps with the flux-divergence method, which was originally proposed by . It consists of applying the continuity equation in a steady state as follows: 1eNO2=divΩNO2w+ΩNO2/τ. The previous equation highlights a transport term D=divΩNO2w, obtained by multiplying NO2 VCDs with horizontal wind speeds and the NO2 sinks S=ΩNO2/τ. As the local overpass time of TROPOMI is close to 13:30, sinks are dominated by the chemical loss due to the reactions of NO2 with OH (leading to the formation of HNO3), which can be described by the chemical lifetime τ as follows: 2τ=1kmeanT,[M]⋅[OH]. Here, kmean is the kinetic reaction rate that characterizes the different first-order reactions between NO2 and OH. provide a general expression of this reaction rate with respect to atmospheric conditions (temperature T and total air concentration [M]).

In the atmosphere, OH is a dominant oxidant species. It is mainly produced during daylight hours by the interaction between water and atomic oxygen produced by ozone dissociation , and its concentration is therefore strongly correlated with solar ultraviolet radiation . The direct measurement of OH is possible using spectroscopic methods, but the spatial representativeness of the data is limited due to its short lifetime. Most global analyses thus estimate OH budgets from other variable species with high associated uncertainties . In polluted air, another mechanism for OH production is the reaction between NO and HO2. This reaction, referred to as the NOx recycling mechanism, illustrates the nonlinear dependence of the OH concentration on NO2 . Here, considering OH to be the only sink assumes that other sinks are negligible. We consider such an approximation to be valid . This hypothesis can be validated a posteriori by analysing the calculated NOx emission maps and verifying the absence of highly negative emissions (which would correspond, all other things being equal, to an underestimation of the sink term) or emissions having the shape of a plume (which would correspond, all other things being equal, to an overestimation of the sink term).

Finally, it should be noted that anthropogenic combustion activities produce mainly NO, which is transformed into NO2 by reaction with ozone O3. NO2 is then photolysed during the day, reforming NO . This photochemical equilibrium between NO and NO2 can be highlighted with the concentration ratio L=[NOx]/[NO2]. NOx emissions are therefore obtained by multiplying NO2 production eNO2 by L. As diurnal NO concentrations in urban areas are generally above 20 ppb, the characteristic stabilization time of this ratio never exceeds a few minutes . This time being lower than the order of magnitude of the inter-mesh transport time (about 30 min considering the resolution used and the mean wind module in the region), we can reasonably neglect the effect of the stabilization time of the conversion factor on the total composition of the emissions and treat each cell of the grid independently from its direct environment. In , this ratio was estimated using the CAMS NO and NO2 concentration fields in Egypt. For Qatar, CAMS data show many outliers for these concentration fields: for some pixels, NO concentrations can be equal to zero. They can be also 2 times higher than NO2 concentrations in places without any significant feature that would explain it. We therefore do not use these fields and choose a fixed value of 1.32 for this ratio, as used by . After removal of outliers in the statistics, CAMS values for L suggest that the average value for this ratio ranges between 1.17 and 1.60 with small spatial variations.

Through the OH concentration, we enable a variability in the chemical lifetime. This variability is not allowed in the original version of the first-divergence method by , which relied on heavy averaging over time to infer emissions at the scale of cities and power plants. Here, seasonal and spatial variations of lifetimes are resolved, thus limiting the errors in the estimation of the daily sink term. Although errors remain high when estimating daily emissions, averaged monthly emissions are correctly resolved above the main emitters.

We apply Eq. () to daily TROPOMI images and average the resulting emissions to obtain monthly daytime emissions. This process can be hindered due to the presence of poor-quality measurements that prevent the calculation of the divergence in the transport term each day. This happens when a significant part of the country is covered by clouds. Furthermore, the resolution used to grid data (0.0625∘×0.0625∘, i.e. about 6.3×6.9 km2 in this region) is lower than the initial resolution of TROPOMI (3.5×7 km2 until 6 August 2019). For the first 8 months of this study, our gridding therefore results in maps having many pixels without data. In the process of estimating mean monthly emissions, we consider these two situations and do not take into account the averaging of the days for which corresponding TROPOMI images cover more than 70 % of the territory of Qatar (defined by the internal mask) without data or with poor quality data (qa<0.75). Furthermore, Bahrain is located 30 km away from Qatar, and the centres of the capitals of the two countries are separated by about 130 km. Pollution from Bahrain, which is usually transported to Qatar, can reach Doha during strong wind events. In such situations, the errors in ERA5 can alter the estimation of the transport term and thus the NOx estimates. As a consequence, we also remove days with high wind speeds in the Bahrain–Qatar direction, i.e. days for which the average wind over Bahrain and the marine area between the two countries has a speed higher than 30 km h-1 and an angle between -15∘ (E1/4SE) and -75∘ (S1/4SE). The choice of these thresholds is discussed in the Supplement. Figure  shows the amount of days considered in the calculation of the average emissions of a given month between 2019 and 2022. It also shows the number of days that have been discarded and the reasons for the corresponding discards.

Here, monthly average emissions are calculated with between 8 and 25 d of TROPOMI images, with low values for the first 7 months of the study due to the lower resolution of TROPOMI during this period. Excluding these months, averages are calculated with 17.9 d. Cases for which too many pixels are not observed by TROPOMI in the internal mask occur more frequently than cases with strong winds from Bahrain. Overall, we consider most of the inferred monthly averages to be robust. However, at the small scale, there are a small number of pixels with a constant absence of reliable VCD estimations due to low-quality measurements, resulting with pixels having estimations of NOx with a poor reliability. The treatment of such pixels is shown in Sect. 5.3.

We use the flux-divergence model to obtain NOx emission maps within our domain. For each applicable day (as described in Sect. 5.1), a background is calculated using the external mask, and daily emissions are estimated. These daily emissions are then averaged to obtain a representation of the emissions at 13:30 on a monthly scale. For October 2022, a maximum of 25 daily emissions have been average to calculate the corresponding mean monthly emissions map, which is displayed in Fig. . When the transport term is integrated over large spatial scales, it cancels out due to the mass balance in the continuity equation between NO2 sources and NO2 sinks. Although the sink term is responsible for most of the total NOx budget within the domain, the transport term can reach high values at a small scale, highlighting hotspots where emissions are concentrated. Such hotspots comprise gas power plants in the northeastern part of the country and industrial facilities in the southeast. High emissions are also observed in Doha, but in that case, the sink term accounts for most of the NOx budget due to the spread of sources in the urban area. The western part of the country, where two cement plants are located, also shows significant emissions. High emissions outside of Qatar are observed in the urban areas of Manama (Bahrain) and Dammam (Saudi Arabia). In the southeastern corner of the domain, an area of high emissions also seems to stand out from the desert. This area corresponds to a cross-border centre between Saudi Arabia and the United Arab Emirates. It is possible that important road traffic there is responsible for such emissions, but we have no information to support this hypothesis. Finally, we observe high emissions at sea, up to about 30 km from the eastern coast of the country. These emissions of about 2.0×1015 molec. cm-2 h-1 cannot be solely attributed to shipping activity . Because in this region, the centre of the 40 km × 44 km CAMS pixel is close to the coast, the NO2 lifetime is probably underestimated. The rest of the domain has relatively low emissions, which is consistent with the absence of major sources of NOx.

Nevertheless, it can be noted that inferred emissions in the domain remain relatively noisy. From monthly maps, we calculate average emissions at 13:30 for the period 2019–2022, which are displayed in Fig. . The hotspots identified earlier remain visible with a improved signal-to-noise ratio. In particular, emissions in the northeastern region are superior to 3.0×1015 molec. cm-2 for only seven pixels, which are among the pixels closest to the power plants in the region. In the eastern part of the country, NOx emissions reflect the main directions of urban expansion in Doha (from the coast towards the west and the southwest), with emissions ranging from 2×1015 to 10×1015 molec. cm-2 h-1. The desert areas display very low emissions (about 0.2×1015 molec. cm-2 h-1), indicating a slight overestimation of the average sink term or an underestimation of the NO2 background. Emissions at sea remain abnormally high on the east coast, but the effect is lower than what is observed in Fig. . However, unlike for October 2022, for which the emissions from cement plants could be identified, we observe low emissions in the western part of the country.

High-quality measurements by TROPOMI are infrequent above the western and southeastern parts of Qatar, with pixels regularly corresponding to measurements with qa<0.75. For these pixels, the inferred emissions are only calculated with an average over a very small number of measurements. In many cases, no measurements are available, and emissions cannot be attributed to the pixels. Figure  shows the mean TROPOMI coverage over the domain for the year 2020, counted as fraction of days with qa<0.75 within the year. In addition to the west and southeast (corresponding to the southern part of Doha) already identified, the northeast is also affected but to a lesser extent. This situation is the same for the years 2019 and 2021, and the identified areas have a fraction of observable days lower than 40 %. The situation is different for the year 2022, during which the west had a fraction of about 70 % and the southeast had a fraction of about 50 % (see the Supplement). The effect on the total inferred emissions is not negligible since pixels that are concerned include intensive emitters, such as an industrial area in the southeast and the two large cement plants in the west. For instance, during the 36 months of the period 2019–2021, inferred emissions above these cement plants were lower than 1.5×1015 molec.cm-2 h-1 for 8 months and greater than this value for 8 months. Emissions could not be inferred for at least one of the plants for the remaining 20 months. Conversely, emissions could be inferred for 10 months out of 12 in 2022, and they were higher than 1.5×1015 molec. cm-2 h-1 for 6 of those months. In the TROPOMI products, the value of qa includes automated quality assurance parameters related to different algorithms concerning the presence of clouds, aerosol particles, pressure levels and other physical quantities. For situations with qa<0.75, retrievals are only sensitive to the NO2 concentrations above the clouds and will depend on model assumptions for the missing part. They still contain useful information, but this information should be carefully interpreted. Here, all areas that are affected by these low values are located near the coasts. These areas are not characterized by persistent clouds, but they are all located within or close to urban or industrial centres, in which aerosol emissions can be high. However, other regions in Qatar and the neighbouring countries are characterized by high emissions of aerosols without qa values being particularly high there. For instance, many studies have been providing estimates of NOx emissions in Riyadh without indicating anything similar. We observe that these high frequencies of low qa values are found in several urban and industrial areas on both coasts of the Persian Gulf (especially in Saudi Arabia, the United Arab Emirates and Iran). This suggests that the identified persistent low values for qa come from the calculation of one or several quality assurance parameters. Because the areas displaying low values usually have a rectangular shape, it also suggests a low resolution for these parameters.

Here, we observe that the use of the OFFL stream (version 2.3.1) from November 2021 to October 2022 generates fewer pixels for which the value of qa falls below 0.75 compared to the use of the S5P-PAL data from January 2019 to October 2021. Furthermore, the last 2 months of 2022, calculated with the version 2.4.0 of the NO2 product (operational since July 2022), do not show any area with particularly low-quality flag values (see the Supplement). The latest version of the user manual indicates that, in this last version, the surface albedo climatology (SAC) used in the NO2 fitting window and derived from OMI and GOME-2 was replaced by a SAC derived from TROPOMI observations . This new TROPOMI SAC is consistently applied in the cloud fraction, in the cloud pressure retrievals and in the air mass factor calculation. In the low-qa areas identified in Fig. , it is therefore possible that high levels of aerosol-generating industrial activities in a small number of pixels in the grid results in a value of qa below 0.75 in those pixels but also in all the surrounding pixels due to the low resolution of the parameter responsible for exceeding the threshold without these pixels being located over particularly emissive zones. This effect would be absent in the version 2.4.0 of the TROPOMI product. For the first 34 months of our study, the calculated monthly emissions for these areas would be then underestimated because they would correspond to the days with low industrial activity.

To evaluate the impact of this effect, we apply the same model using a threshold value of qa corresponding to 0.7. This includes many pixels that would be rejected with a threshold of 0.75. As the TROPOMI manual recommends not using this threshold level, we do not consider these emissions to be representative of human activity in Qatar, and they are not treated as emissions estimates. Nevertheless, it allows us to evaluate the effect of non-detection of cement plants and power plants in the southeast of the domain. The use of a threshold value of qa=0.7 increases the inferred fluxes by 43.8 % above cement plants in the west and by 33.4 % above the industrial area in the southeast. Fluxes for the Ras Laffan area in the north and central Doha, which are less affected by the effect described here, are increased by 2.6 % and 3.0 % respectively. It is difficult to determine whether these increases are realistic because they were obtained using columns where NO2 might have been estimated above aerosol and cloud layers. However, they indicate that our model, run with TROPOMI data before November 2021, leads to a systematic underestimation of NOx emissions in two of the four most NOx-intensive areas of the country. We note, however, that, on average, these increases are compensated for: with a threshold of qa=0.7, total fluxes within the internal mask are reduced by 2.1 % because fluxes outside the main emissive areas, which comprise most of the mask, are reduced by about 17 %.

In Qatar, the official rest day is Friday, and the economic activity of the country is lower during this day than during the other days of the week. We therefore try to characterize this feature by evaluating the weekly cycle of NOx emissions. We use the TROPOMI-inferred emissions to obtain averages per day of the week. We use the flux-divergence method to calculate the emissions within the internal mask. We remove the days for which most of the area contains poor-quality measurements and those for which the wind from Bahrain blows to the southeast. This is the same way the method was conducted in Sect. 5.1.

For a given day, the empty pixels (i.e. those to which it was not possible to assign emissions) are filled with the average of the emissions obtained for the remaining pixels. This choice leads to an underestimation of the emissions when the main emitting areas are covered by clouds or aerosols and to an overestimation in the opposite case. To limit this effect, we remove from our estimate of annual averages the emissions below the 5th percentile and above the 95th percentile. Figure  shows the resulting daily emissions for the 4-year period. Each year, a Friday minimum is observed, defining a weekly cycle. This trend is also observed for mean NO2 column densities but not for OH concentrations. It is also more pronounced for residential areas, for which the Friday activity is 40 % lower than the weekly average. Conversely, the effect is weaker above the power plants in the north of the country, which are located more than 30 km away from major residential areas (see the Supplement). At the scale of the country, Fridays have average emissions of 9.04 t h-1, which is lower than average emissions for the rest of the week, which reach 14.15 t h-1. In Middle-Eastern countries, this “week-end” effect had already been inferred from satellite measurements by and and from ground measurements by on NO2 concentrations but with a lower difference between Fridays and the other days.

In this study, we estimate the emissions at around 13:30, which corresponds to the moment when TROPOMI overpasses the country. However, anthropogenic activity is not uniform throughout the day, and the emissions inferred from parameters calculated at 13:30 do not correspond to the average emissions of the country. Using the power consumption data from , Fig.  shows the mean load curve for the country on April 2016, July 2016, October 2016 and January 2017. These curves show that the average power injected into the grid is slightly lower than the power injected at 13:30. The ratio between the two powers is minimal in June, where it reaches 0.911, and maximal in February, where it reaches 0.976 (see the Supplement). Moreover, according to the traffic congestion index in Doha , the road traffic around 13:30 shows a congestion peak that is slightly lower than the other peaks that occur at the beginning and the end of the daytime. This suggests that the average emissions from the road transport sector are also close to those at 13:30, with a similar ratio to that of the power sector.

We therefore re-scale the inferred NOx emissions at the scale of the country by multiplying the monthly estimate for 13:30 with the power ratio to estimate mean monthly emissions. We acknowledge that this is a simplification: the power sector has a variable emission factor due to the different technologies in the power plants that are used to respond to the electricity demand over the year, and the road congestion level does not necessarily reflect traffic emissions. Moreover, other sectors, such as cement production, whose seasonality is unknown to us, might behave differently.

Within the internal mask, Fig.  shows total NOx emissions in Qatar using our method. Since the NO2 background has been removed from VCD calculations, the total NOx budget calculated for each month corresponds to the NOx production by human activities. The annual variability is marked, with higher emissions in summer and lower emissions in winter. On average, NOx emissions reach 9.56 kt per month. With mean emissions of 9.29 and 10.08 kt, 2019 and 2022 appear to be the years with lowest and highest mean emissions respectively. No seasonal cycle seems to appear for VCDs, except above the greater Doha area, which undergoes NO2 levels that are about 23 % higher than the 2019–2022 average from October to January, while they are 27 % lower from June to August. To obtain the observed cycle in NOx emissions, these differences in VCDs over Doha suggest a more intense oxidation in summer, which is captured by higher OH values for summer months. This is consistent with nonlinear effects in the NO2 loss rate, with an absolute growth rate that is higher for OH than for NO2 VCDs at high NO2 levels inside the plume . However, because wind speeds tend to be higher during summer in Qatar, the observed variations in VCDs over Doha can also be partially explained by a spread of the pollution beyond the shores of the country. The uncertainties, detailed in Sect. 7, are higher for the months for which emissions could not be attributed for the western and southeastern parts of the territory. This hinders year-on-year comparisons. Consequently, although emissions in the March–April–May (MAM) 2020 period are 96 % and 87 % of the level of emissions in MAM 2019 and MAM 2021 respectively, we cannot assert whether this effect can be attributed to the restrictions put in place from March to May 2020 to tackle the Covid-19 pandemic. This caution is all the more necessary as these restrictions have been less stringent than in other countries . Similarly, higher emissions during 2022 could be due to the different values reached by qa in the 2.4.0 version of the TROPOMI product with respect to the 2.3.1 version. However, it cannot be excluded that these higher emissions were due to a more intense activity in 2022 in the context of organizing the FIFA World Cup.

In Qatar, all electricity production comes from gas-fired power plants, with three plants located in the northeast (Ras Laffan area). Other gas power plants are located in the urban area of Doha and in the southeast of the country. Figure  shows the total electricity production and the share of the power plants in Ras Laffan between 2019 and 2021 as provided by Kahramaa. Data regarding the electricity generation by power plant in 2022 are not yet available. The electricity generation is 2 to 2.5 times higher during summer months than during winter months. This increased generation is mostly due to the use of air conditioning , which is also captured in Fig. .

The inferred emissions have a behaviour that is similar to electricity generation, with higher emissions in the warmer months. The comparison of the time series of NOx emissions and the electricity generation data for the 4-year period provides a correlation coefficient of R2=0.400. This value is relatively low, but it should be noted that monthly emission maps can be very noisy due to averaging over too few days. For instance, in April 2019, the spatial distribution of emissions for this month shows non-negligible emissions in the central part of the country where no significant emitter is located. We can therefore consider it to be the case that the average is not calculated with enough data to limit noise effects. When we only keep in the analysis the mean monthly emissions calculated on the basis of more than 18 d (the average over the 4-year period being 17.1), 19 points out of 48 are retained, and the correlation is improved, with a coefficient which then reaches R2=0.657, with a slope of 1.773 tNOx GWh-1. This value would be equal to the emission factor of the power sector, provided that it was the only source of variable NOx emissions during the months that are considered. This value must therefore be considered as an upper-limit estimate. Figure  shows the comparison between electricity generation and emissions in the two cases. The points that are retained in the second case mostly correspond to months with high emissions, notably in autumn. Among these 19 points, only 2 correspond to winter or spring months (December to May), whereas 7 correspond to summer months (June to August) and 10 to autumn months (September to November). The improved correlation must be interpreted with caution, but it might be possible that other NOx-producing sectors have a seasonality which is the opposite to that of the power sector. If the transport sector has no particular seasonal cycle, it is possible that industry has a higher production during winter and spring months, leading to higher industrial NOx emissions during this period. As mentioned before, midday temperatures in summer are high in Qatar, preventing labour in outdoor activities, and this interpretation could be investigated. With the exception of the two cement plants in the west of the country, most of the industries are located in the urban area of Doha or Ras Laffan, where emissions from several sectors overlap over distances smaller than the resolution of TROPOMI.

The method can also be applied to a smaller area of the domain. The northeastern part of the country is interesting in this respect: it is very sparsely populated, and the human activity there is mainly associated with oil and gas activities and the production of electricity. In this region, three power plants (Ras Laffan A, B and C) have a total capacity of 3.35 GW and are responsible for about 45 % of the total electricity generation throughout the year, as shown in Fig. . These power plants are independent water and power plants (IWPPs), serving as both desalination plants and power plants. Other than these plants, the region also has gas liquefaction facilities as Qatar is a major producer of liquefied natural gas. Refineries are also present. In the life cycle of gas, only extraction and combustion are highly NOx-emitting processes: emissions from midstream (liquefaction and transport) and downstream (refining) activities are 20 to 30 times lower per unit of fuel . As the oil and gas extraction processes are located outside this area (mostly offshore), it can be considered that the majority of the NOx emitted in this region comes from the combustion of gas in the power plants. This is not the case for other power plants in the country, which are located in urban areas where emissions from different sectors overlap. We focus on this area and calculate the corresponding monthly emissions. The three power plants are located within the same pixel, and the difficulty for TROPOMI to visualize this region with good-quality measurements is less pronounced than for other regions identified in Sect. 5.3. Because Ras Laffan plants are not the only ones participating in the country's electricity generation, the use of load curves is no longer valid to infer total emissions in a given period, and only emissions at 13:30 can be considered. Assuming a correct calculation of emissions, the observed emissions display a Gaussian distribution around the power plants. Using a zonal cross-section of obtained emissions in a 28 km band centred around the mean latitude of the plants, we estimate the mean emissions by fitting a Gaussian curve eNOx(λ)=B+E0σ2πexp⁡-(λ-λ0)22σ2 to the profile, with B being the observed emissions above desert areas and seas, λ0 being the mean longitude of the power plants, σ being the standard deviation (measured as an angle) and E0 being the total emissions. λ is the zonal angle. The best approximation is displayed in Fig. .

Monthly profiles peak around the location of the power plants whose heights vary without displaying any particular seasonality. The fit of the averaged profile leads to a low positive background of 67.8 kg h-1 per unit of band surface, i.e. about 0.390 mg m-2 h-1, and average emissions of 1.86 t h-1 for the three power plants. This value is 1 order of magnitude similar than that of emissions for power plants in other regions estimated with the flux-divergence method . In particular, it corresponds approximately to the value reported by , which estimated an output of 1.81 t h-1 despite various differences in the parameterization of the method.

Here, only a small part of the domain is considered, for which the power sector is by far the major contributor to NOx emissions. It is therefore not necessary to consider the seasonality of the other sectors in the estimation of an emission factor, as was done earlier when the emissions of the whole domain were considered. However, as the scaling assumption can no longer be used, other assumptions must be made. One assumption would be that these plants serve as intermediate-load power plants: with 13:30 being a peak of consumption at any time of the year, the Ras Laffan plants would be running very close to their maximum capacity, and the annual variability of power generation by the three plants observed in Fig.  would be due to the decrease in power generation during low-demand hours. This would be consistent with the lack of seasonality detected in the monthly profiles in Fig. . In this case, comparing the emissions at 13:30 with the total capacity of the plants provides an emission factor of 0.557 tNOx GWh-1. As the total capacity is used in the calculation, this value must be considered as a lower-limit estimate.

Another possible assumption is that the three plants are used as peak-load power plants: in this case, they would be used at 13:30 to respond to the daily peak. This use would increase with the height of the peak and therefore with the temperature. In this case, the annual variability of power generation by the three plants would be due to their variable use at 13:30. The absence of a clear annual variability in our TROPOMI-inferred emissions would be explained by the large uncertainties in the calculation of our emissions, which concern only six pixels of our domain. Here, the large share of the three power plants in the electricity production seems to rule out a peak-load functioning. However, it is also possible that some of the plants correspond to different functions within the national grid.

We compare TROPOMI-derived NOx emissions to the Emissions Database for Global Atmospheric Research (EDGARv6.1) for 2018 and the CAMS global anthropogenic emissions (CAMS-GLOB-ANT_v5.3) from 2019 to 2022. The two inventories provide 0.1∘×0.1∘ sectorial gridded emissions on a monthly basis. After aggregating the different sectors of activity, the CAMS-GLOB-ANT and EDGAR inventories directly provide the anthropogenic NOx emissions over the same domain. Both inventories display NOx emissions that are significantly higher than our estimates. According to CAMS-GLOB-ANT, the annual emissions for 2019, 2020, 2021 and 2022 are 191.7, 191.8, 192.5 and 193.7 kt respectively, whereas we only estimate emissions of 111.5, 111.8, 114.4 and 121.0 kt for these same years. According to EDGAR, NOx emissions are even higher, reaching 193.8 kt in 2018. It should be noted that the same version of EDGAR estimates emissions of 111.1 kt in 2007, which is similar to the reported terrestrial emissions for the same year, as well as our estimated emissions for 2019–2022.

The sectoral shares of emissions estimated by EDGAR and CAMS-GLOB-ANT differ. In EDGAR, the transport sector accounts for 42 % to 54 % of the emissions, whereas it has a nearly constant share of 50 % in CAMS-GLOB-ANT. In the remaining emissions, industry accounts for about 2.8 kt per month in both inventories. These emissions are slightly lower during summer, which can be interpreted as a lower production due to a reduced outdoor activity at high temperatures. The corresponding share is more variable in EDGAR (between 13 % and 23 % of terrestrial emissions) than in CAMS-GLOB-ANT (about 17 % of terrestrial emissions). These values are lower than reported industrial shares for 2007, where the cement sector alone accounted for 23.4 % of total emissions. Finally, power sector emissions in both inventories account for a significant share of total emissions but with a higher value for EDGAR (35 % of the total budget for 2018) than for CAMS-GLOB-ANT (32 % of total emissions in 2019–2022). The two values are respectively higher and lower than the reported share for power emissions, which was estimated at 28.1 % for terrestrial emissions in 2007. Another major difference concerns the seasonal cycle of those power emissions. The two inventories do not show any correlation with the corresponding electricity generation data. However, in 2018, EDGARv6.1 shows a pronounced variability, with extremes of 7.0 kt emitted in May and 3.5 kt in December (see the Supplement). The ratio between these two values remains lower than the ratio between the extremes of the monthly electricity generations, which has varied between 2.2 and 2.8 over the last 4 years, but it highlights a difference compared to the previous version of EDGAR (EDGARv5.0), which showed almost no seasonality in power emissions. This lack of seasonality is also found in the emissions of CAMS-GLOB-ANT_v5.3, which uses EDGARv5.0 as a basis. We also note that, while these seasonality differences between TROPOMI-inferred emissions and inventories were already highlighted for the previous versions of EDGAR and CAMS-GLOB-ANT in Egypt, the difference between mean annual NOx emissions was much lower .

The report for emissions in 2007 indicates an emission factor for the power sector of 0.392 tNOx GWh-1, and the corresponding power plants are still operating today. In the previous sections, we estimated an emission factor of 1.773 tNOx GWh-1 using all monthly emissions in the internal mask and assuming the power sector was the only source of variable NOx emissions. We also estimated a value of 0.557 tNOx GWh-1 for Ras Laffan power plants using mean emissions at 13:30 under the hypothesis of a intermediate-load power plant functioning. To infer emission factors from inventory estimates, power emissions in EDGAR and CAMS-GLOB-ANT and total electricity generation provided by the Planning and Statistics Authority (PSA) between 2018 and 2022 can also be used to infer emission factors of 1.406 and 1.219 tNOx GWh-1 respectively. Table  summarizes all the different emission factors that have been calculated. Even though EDGAR's emission factor is particularly high compared to other estimates, the five listed values are plausible, as the majority of gas-fired power plants have a NOx emission factor value between 0.1 and 10 tNOx GWh-1 .

The comparison between these different emission factors should be made with caution. The factor reported by the report for 2007 emissions takes into account all of the power plants in the country, the oldest having been built in 1980. Moreover, some units of one of the Ras Laffan plants were built between 2010 and 2011, i.e. after the publication of the report. It is also possible that other power plants have been upgraded in the last years. Finally, the reported emission factor for 2007 accounts for the public electricity production and the water production, but the monthly reports from Kahramaa do not specify whether the electricity generation from the Ras Laffan power plants accounts for the large amount of electricity that is used in desalination processes . Monthly reports show that the water production from the IWPPs in the country does not have a clear annual cycle, with the production generally ranging between 50×106 and 60×106 m3 per month in the entire country. The amount of electricity needed for this water production varies with the technologies used in the IWPPs and accounts for about 20 % of the consumption in total electricity . Consequently, if the electricity used for water production is not accounted for in the electricity generation for monthly reports, then all inferred emission factors are overestimated.