Megacity Emissions and Lifetimes of Nitrogen Oxides Probed from Space

See allHide authors and affiliations

Science  23 Sep 2011:
Vol. 333, Issue 6050, pp. 1737-1739
DOI: 10.1126/science.1207824


Megacities are immense sources of air pollutants, with large impacts on air quality and climate. However, emission inventories in many of them still are highly uncertain, particularly in developing countries. Satellite observations allow top-down estimates of emissions to be made for nitrogen oxides (NOx = NO + NO2), but require poorly quantified a priori information on the NOx lifetime. We present a method for the simultaneous determination of megacity NOx emissions and lifetimes from satellite measurements by analyzing the downwind patterns of NO2 separately for different wind conditions. Daytime lifetimes are ~4 hours at low and mid-latitudes, but ~8 hours in wintertime for Moscow. The derived NOx emissions are generally in good agreement with existing emission inventories, but are higher by a factor of 3 for the Saudi Arabian capital Riyadh.

Nitrogen oxides, which are produced mostly by combustion processes, play a key role in tropospheric chemistry: They are toxic, contribute to acid rain, can act as aerosol precursors, and are catalysts for ozone formation (“summer smog”) (1). Thus, global and regional chemistry models require accurate NOx emission inventories, in particular for megacities (2), which are immense sources of numerous pollutants (35). However, substantial differences among state-of-the-art emission databases have been found (6), and uncertainties are high (7).

Satellite measurements of NO2 tropospheric columns have revolutionized our insight into the distribution and magnitude of sources of nitrogen oxides over the past decade (814) and generally enable top-down NOx emission estimates on regional and global scales (810). This, however, requires a priori information on the NOx lifetime τ, which, at daytime, is mainly determined by the reaction of NO2 with the hydroxyl radical OH, leading to formation of HNO3 (15). τ can generally be obtained from atmospheric chemistry models (16), but the accuracy of the estimated emissions is then limited by the models’ capability to represent OH concentrations accurately, which is especially difficult for megacities owing to highly nonlinear small-scale chemistry.

For typical background OH concentrations (~106 molecules/cm3), τ is about a day (1), but within pollution plumes with active photochemistry and enhanced OH levels, it is only a few hours, as determined by measuring the downwind evolution of NOx during aircraft transects of plumes from cities and power plants (17, 18). Similar lifetime information can also be gained from the downwind evolution of NO2 as observed from satellite instruments (8, 13, 19) (20).

Here we present a method for the simultaneous determination of NOx emissions and lifetimes for megacities and other strong “point” sources. In contrast to earlier analyses of satellite data, this investigation of photochemistry within urban plumes was rendered possible by the spatial resolution of the Ozone Monitoring Instrument (OMI) (21) with ground pixel sizes down to 24 km by13 km.

The essential element of our approach is that the mean NO2 distribution is calculated separately for different wind direction sectors, rather than taking the overall average. By making this separation, we avoid the neutralization of outflow patterns of opposite wind directions, which would result in low average wind speeds and rather symmetric NO2 patterns. Instead, situations with high mean winds (for each wind direction sector considered separately) and correspondingly distinct outflow patterns are achieved.

In this study, we consider OMI NO2 tropospheric columns, i.e., vertically integrated concentrations, from “cloud free” (cloud fraction below 30%) measurements at ~2 p.m. local time for 2005 to 2009 (v1.02, TEMIS) (22). Each OMI ground pixel is linked to wind data (below 500 m) from the European Centre for Medium-Range Weather Forecasts (ECMWF) (23). For details see (24).

To demonstrate our approach, we focus on Riyadh, the capital of Saudi Arabia, with a population of about 5 million (city) to 7 million (greater area) inhabitants. Riyadh is particularly suited for this study for several reasons: First, NO2 tropospheric columns in Riyadh are large (~20 × 1015 molecules/cm2). Second, Riyadh is isolated (no other large NOx sources within ~200 km radius), and the contrast between the polluted city center and the relatively low background (~1 × 1015 molecules/cm2) is high. Thus, Riyadh can be considered as a nearly idealized point source of NOx. Third, Riyadh is far from the coast, and the surrounding terrain, and thus also wind fields, are rather homogeneous. Finally, Riyadh is only rarely covered by clouds, allowing undisturbed satellite observations down to the ground.

Figure 1 (left) displays mean NO2 tropospheric columns for the Middle East region for calm conditions (wind speeds w < 2 m/s). On the right, enlarged plots for Riyadh and the surrounding area are shown, where NO2 tropospheric columns have been averaged separately for different wind direction sectors (southeast, south, southwest, etc.); each OMI measurement is assigned to one of these sectors according to the respective ECMWF average wind direction below 500 m. The resulting spatial patterns clearly illustrate the outflow of NO2 from Riyadh, consistent with ECMWF winds. In the analysis below, each OMI observation (also for w < 2 m/s) is assigned to a wind direction sector, to increase sample sizes and reduce noise.

Fig. 1

Wind dependency of NO2 column densities around Riyadh. (Left) Mean NO2 tropospheric columns in the Middle East from OMI measurements during 2005 to 2009 for calm (w < 2 m/s) conditions with <30% cloud cover. The gray box indicates the area around Riyadh displayed in the enlarged plots on the right. (Right) Mean NO2 column densities around Riyadh (white cross) for different wind conditions, i.e., calm (center panel) and eight main wind direction sectors (surrounding panels; arrows indicate the mean of the respective ECMWF winds).

For each wind direction sector, the mean column density maps (two-dimensional, 2D) are reduced to 1D “line densities” along the respective main wind direction (x) by integration across the wind direction (y), respectively (fig. S1). The integration implicitly accounts at the same time for the spatial extent of the source, the OMI ground pixel size, and wind velocity fluctuations in the y direction. Figure 2 shows line densities (light colors) as a function of the distance x to the Riyadh city center, exemplarily for easterly (red) and westerly (blue) winds. A clear asymmetry of the spatial patterns due to transport can be seen: The maximum is shifted in the wind direction, and the curve is less steep downwind than upwind.

Fig. 2

Downwind evolution of NO2. Light colors: Zonally integrated NO2 column densities (mean ± SME) for westerly (blue) and easterly (red) winds as function of the distance to Riyadh City center x. Dark colors: the respective fit result M(x). The numbers indicate the mean wind velocities from ECMWF (w) and the lifetimes τ resulting from the least-squares fit with 95% CIs.

Assuming a pseudo first-order loss of NO2, the mean NOx lifetime τNOx and NOx emissions ENOx can be determined from the observed downwind decay. For this purpose, we perform a nonlinear least-squares fit of a simple model function M(x) (dark colors in Fig. 2) to the observed spatial pattern of the NO2 line density, as a function of the distance x. M(x) describes, in essence, a spatially smoothed exponential downwind decay (24). Fitted parameters are the e-folding distance x0, the spatial smoothing parameter σ, the location of the apparent point source relative to the city center X, total emissions E, and a constant background B. σ accounts for any kind of spatial smoothing in the main wind direction, caused by the actual spatial extent of the source, the ground pixel size of OMI observations with different spatial samplings, and wind variations. For details, see (24).

For Riyadh, the observed mean downwind patterns can be described successfully by this simple model function: Correlation coefficients between observation and fit are about R2 = 0.98 for all wind direction sectors. From the fitted e-folding distance x0, the mean daytime lifetime τ is derived by division by the mean wind speed w. This actually reflects the NOx lifetime, as long as the [NO2]/[NOx] ratio does not change much within the downwind plume (17). For each wind direction sector, τ is computed to be about 4 hours with a typical 95% confidence interval (CI) of about ±0.5 hours. Averaging the fit results for all wind direction sectors yields τ = 4.0 hours with a standard mean error (SME) of 0.4 hours.

The derived emissions E (in terms of NO2) are 187 ± 14 mol/s (mean ± SME) with a typical CI of 16 mol/s for the individual wind direction sectors. For the determination of NOx emissions ENOx, E is scaled by a factor of 1.32, according to typical [NO]/[NO2] ratios of 0.32 under urban conditions at noon (1). The other fitted parameters are X = 15 ± 4 km, B = (0.27 ± 0.03) × 1023 molecules/cm, which corresponds to a background tropospheric column of (1.1 ± 0.1) × 1015 molecules/cm2, and σ = 21 ± 1 km, which is of the expected order of magnitude according to the spatial extent of Riyadh and the OMI ground pixel size.

The CI and SME reflect the fit performance and the consistency of the result for the different wind direction sectors, respectively, and both can be regarded as measures of uncertainty. In addition, the absolute values for E (but not τ) are also directly affected by the uncertainty of OMI tropospheric columns of about 30% (22), as well as by the uncertainty in the NOx/NO2 scaling factor (~10%). Further sources of uncertainty of both τ and E are the settings applied for spatial integration (~10%) (fig. S2) and the choice of wind data (~30%). Total errors of τ and E are estimated as the quadratic sum of these uncertainties, assuming them to be independent (24).

Our method for determining NOx lifetimes and emissions is generally applicable for any strong, localized source. Figure 3 compares the resulting NOx emissions to the EDGAR (V4.1) emission inventory (25, 26) for the year 2005 for a set of (mega)cities and one other strong point source, the Four Corners (27) power plants, which have passed an automated performance check (see SOM text and table S1). In addition, the derived mean lifetimes are color-coded in the scatterplot.

Fig. 3

Resulting NOx emissions and lifetimes. Scatterplot of the resulting NOx emissions for the considered megacities and power plants versus the respective EDGAR emissions (V4.1, integrated over 250 km by 250 km). Error bars show the total uncertainties for ENOx (24), whereas EDGAR errors are set to 50% according to the expert judgment (“medium magnitude of uncertainty”) (7). Resulting NOx lifetimes are color coded.

Despite the large uncertainties on the order of 50% for both our results and EDGAR emissions (7), the derived emissions are generally in good agreement with the EDGAR inventory. One possible reason for the deviations is that, due to the short lifetimes, our method is only sensitive to daytime emissions (see SOM text and fig. S4), whereas EDGAR emissions are 24-hour annual averages. For Riyadh, however, where our fit performs best, resulting emissions are higher by a factor of 3. This indicates that EDGAR NOx emissions are too low for Riyadh, though the reason for this is not yet clear; however, from a trend analysis of the satellite observations, we can exclude that this discrepancy is just due to emission changes since the EDGAR reference year of 2005.

The simultaneously derived daytime lifetimes are almost free of a priori assumptions or model input, as they are based on the relative downwind patterns. They are in the range of 2.3 to 6.4 hours (with typical uncertainties of 40 to 60%), in good agreement with prior measurements (16, 17). This range constrains the average OH concentration to be between about 10 × 106 and 4 × 106 molecules/cm3. This directly reflects the oxidizing capacity within the fresh urban plume, which is also relevant for the depletion of other pollutants like CO, SO2, and volatile organic compounds.

An additional analysis of seasonal mean lifetimes and emissions generally reveals similar results (fig. S3). But at higher latitudes (Moscow) in winter, when heterogeneous night-time reactions of NOx (i.e., conversion to HNO3 via NO3 and N2O5) are probably the dominating loss processes, the derived lifetimes are considerably longer (8 hours).

Owing to the global coverage of satellite observations, our method can be applied to various major “point sources” such as megacities around the world. With the ongoing time series of current and especially future (geostationary) satellite instruments with better spatial resolution, our method can serve as a robust, independent tool to validate NOx emission inventories of megacities, and might also be applied to other trace gases in the future as retrievals continue to improve.

Supporting Online Material


SOM Text

Table S1

Figs. S1 to S4

References and Notes

  1. In this study, we define “megacity” as a city with more than 5 million inhabitants, following the definition used within the MEGAPOLI project (
  2. Other loss processes are, e.g., wet and dry deposition and the formation of peroxyacetylnitrate (PAN).
  3. Either explicitly, or implicitly via inverse modeling approaches.
  4. Because the respective satellites have repeat cycles of only 1 to 6 days, it is generally not possible to track a particular air mass, as done with in situ aircraft transects. Instead, spatial patterns of temporally averaged (seasonal/multi-annual) NO2 tropospheric columns have been analyzed under special conditions, resulting in lifetimes of 4 hours (12) up to a day (7, 17). The latter estimates are biased toward high values because they ignore effects of the satellite’s coarse spatial resolution.
  5. Detailed methods are available as supporting material on Science Online.
  6. Though specific emission inventories exist for various megacities (with highly variable quality), we present a comparison to the global EDGAR inventory, which is widely used in global atmospheric chemistry models.
  7. Acknowledgments: The research leading to these results received funding from the European Union’s Seventh Framework Programme FP/2007-2011 within the project MEGAPOLI (grant 212520). We acknowledge the free use of tropospheric NO2 column data from the OMI sensor from We thank the ECMWF for providing wind fields. EDGAR NOx emissions for 2005 were provided by the European Commission, Joint Research Centre (JRC)–Netherlands Environmental Assessment Agency (PBL) (EDGAR version 4.1,, 2010). We thank S. Dörner for valuable assistance in processing ECMWF data.
View Abstract

Navigate This Article