Application of satellite observations for timely updates to global anthropogenic NOx emission inventories



[1] Anthropogenic emissions of nitrogen oxides (NOx) can change rapidly due to economic growth or control measures. Bottom-up emissions estimated using source-specific emission factors and activity statistics require years to compile and can become quickly outdated. We present a method to use satellite observations of tropospheric NO2 columns to estimate changes in NOx emissions. We use tropospheric NO2 columns retrieved from the SCIAMACHY satellite instrument for 2003–2009, the response of tropospheric NO2 columns to changes in NOx emissions determined from a global chemical transport model (GEOS-Chem), and the bottom-up anthropogenic NOx emissions for 2006 to hindcast and forecast the inventories. We evaluate our approach by comparing bottom-up and hindcast emissions for 2003. The two inventories agree within 6.0% globally and within 8.9% at the regional scale with consistent trends in western Europe, North America, and East Asia. We go on to forecast emissions for 2009. During 2006–2009, anthropogenic NOx emissions over land increase by 9.2% globally and by 18.8% from East Asia. North American emissions decrease by 5.7%.

1. Introduction

[2] Nitrogen oxides (NOx = NO + NO2) emitted to the atmosphere mainly through fossil fuel combustion, biomass burning, soil, and lightning play a key role in atmospheric chemistry. Global anthropogenic NOx emissions are expected to change rapidly over the coming decades due to economic development and emissions controls. Monks et al. [2009] give an overview of available inventories. The bottom-up approach of estimating NOx emissions by aggregating activity data and emission factors is a major undertaking that often suffers from a time lag of years between the occurrence of emissions and completion of inventories. Timely NOx emission estimates are needed for better understanding of air pollution, acid deposition, and climate change.

[3] Satellite observations of tropospheric NO2 columns provide near-real-time and independent information on NOx emissions and their trends. Numerous studies have used space-based tropospheric NO2 observations to examine temporal changes [e.g., Beirle et al., 2003; Richter et al., 2005; Kim et al., 2006; van der A et al., 2006; Zhang et al., 2007; Boersma et al., 2008a; Kaynak et al., 2009; Yoshida et al., 2010] and to provide top-down constraints on surface NOx emissions via inverse modeling [e.g., Martin et al., 2003; Jaeglé et al., 2005; Müller and Stavrakou, 2005; Napelenok et al., 2008; Chai et al., 2009; Zhao and Wang, 2009; Lin et al., 2010]. Here we present an approach to rapidly update bottom-up NOx emission inventories using top-down trend analysis of tropospheric NO2 columns from the SCIAMACHY instrument [Bovensmann et al., 1999]. The SCIAMACHY tropospheric NO2 column retrieval is described in section 2. In section 3, we provide a brief account of bottom-up NOx emissions and the GEOS-Chem model. Section 4 presents our approach to construct a top-down anthropogenic NOx emission inventory.

2. SCIAMACHY Tropospheric NO2 Column Retrievals

[4] The SCIAMACHY instrument aboard the ENVISAT satellite observes solar backscatter that can be applied to retrieve tropospheric nitrogen dioxide (NO2) with a typical spatial resolution of 30 km × 60 km, achieving global coverage every 6 days [Bovensmann et al., 1999]. ENVISAT was launched in March 2002 into a sun-synchronous polar orbit, crossing the equator at 10:00 local time in the descending node.

[5] We retrieve tropospheric NO2 columns for the years 2003–2009 using the algorithms described by Martin et al. [2006] with a few updates including the use of clouds from FRESCO+ [Wang et al., 2008] in the air mass factor formulation. For this manuscript we use monthly NO2 vertical profile shapes for 2006 to keep the bottom-up and top-down emissions independent. The SCIAMACHY NO2 retrievals have been validated with coincident airborne in situ measurements [Martin et al., 2006] and extensively applied to understand NOx emissions [e.g., Martin et al., 2006; Sioris et al., 2007; Napelenok et al., 2008; Kaynak et al., 2009; Walker et al., 2010].

[6] Wintertime retrievals are more error prone due to the reduced sensitivity of satellite measurements to lower tropospheric NO2 at high solar zenith angle (SZA) and by uncertainties associated with snow covered scenes [O'Byrne et al., 2010]. We exclude the data for winter by excluding observations made at >50° SZA. To reduce retrieval errors, we use observations with cloud radiance fraction <20%.

3. Bottom-up NOx Emission Inventory and Model Description

[7] We use the bottom-up NOx emission inventory implemented in the GEOS-Chem global three-dimensional model of atmospheric chemistry ( to conduct sensitivity simulations and to evaluate the hindcast top-down inventory. Global anthropogenic emissions are based on the EDGAR 3.2FT2000 inventory [Olivier et al., 2001] for the year 2000. The global inventory is overwritten by regional inventories that include the CAC inventory ( for 2005 over Canada, the U.S. EPA National Emissions Inventory (NEI) for 2002 and 2005 over the United States, the BRAVO inventory [Kuhns et al., 2005] for 1999 over Mexico, the inventory from Zhang et al. [2007] for 2003 and 2006 over East Asia, and the EMEP inventory for 2003 and 2006 for Europe. Where emissions are not available for 2003 or/and 2006, emissions are scaled from the nearest year of available inventory following van Donkelaar et al. [2008]. We focus on 2006 as the most recent year with emission statistics and on 2003 as the most historical year that overlaps with SCIAMACHY observations.

[8] Figure 1 shows bottom-up anthropogenic NOx emissions from land sources for the year 2003 (Figure 1, top) and 2006 (Figure 1, middle) and emission changes for 2003–2006 (Figure 1, bottom). Additional information on the bottom-up inventory is in the auxiliary material. Global anthropogenic NOx emissions increase by 5.2% from 22.9 Tg N in 2003 to 24.1 Tg N in 2006, with global emissions growth partially counteracted by the reduction in North America and Europe. East Asian emissions increase by 25% over the three years. The changes in anthropogenic emissions in Africa, South America, and Oceania are minor (<1010 atoms N cm−2s−1, <0.1 Tg N).

Figure 1.

Spatial distributions of bottom-up anthropogenic NOx emissions at 1° × 1.25° for (top) 2003 and (middle) 2006. (bottom) The difference between anthropogenic emissions for 2006 minus those for 2003.

[9] We develop a global simulation capability for GEOS-Chem at 1° × 1.25° (all previous global GEOS-Chem simulations were at 2° × 2.5° or 4° × 5°). This development is applied to GEOS-Chem version 8-01-04 driven by GEOS-4 assimilated meteorology from the NASA Global Modeling and Assimilation Office. Anthropogenic NOx emissions from land sources are as described above. Other emissions have been recently described by Lamsal et al. [2010]. We conduct a simulation for the year 2006 and coincidentally sample the model output for analysis of the SCIAMACHY data. The GEOS-Chem simulation of NOx has been recently compared with a variety of in situ and satellite observations [e.g., Jaeglé et al., 2005; Martin et al., 2006; Hudman et al., 2007; Wang et al., 2007; Boersma et al., 2008a, 2008b; Zhang et al., 2008; Lamsal et al., 2008, 2010] and generally agrees to within 30% of measured NOx.

4. Prediction of Emissions

[10] Satellite observations of tropospheric NO2 columns are strongly related to surface NOx emissions due to the short NOx lifetime combined with the high NO2/NOx ratio in the boundary layer. We use the GEOS-Chem model to examine the relationship.

[11] Following Walker et al. [2010], we perform two simulations, one with NOx emissions (E) for the year 2006 described in section 3 and another with anthropogenic NOx emissions perturbed by 15%, to establish the relationship between changes in surface NOx emissions and changes in tropospheric NO2 columns (Ω):

equation image

ΔΩ is the change in simulated tropospheric NO2 columns driven by the change in emissions ΔE. The term β represents the local sensitivity of changes in NO2 columns to changes in NOx, similar to the variable α used in top-down emission inference (E) by Martin et al. [2003]:

equation image

Unlike α (units of s−1) that describes the direct relationship between NO2 columns and NOx emissions, β is a unitless trend factor that describes how a change in NOx emissions changes the NO2 columns. β reflects the feedback of NOx emissions on NOx chemistry. Using a perturbation of 30% (instead of 15%) changes β by <2%. The overall error in the approach arises from the combination of errors in the NO2 column trend and in β, which are the subject of ongoing work.

[12] Figure S1 in the auxiliary material shows the spatial variation of annual averaged values of β for 2006 coincidentally sampled with the SCIAMACHY data. The global mean value is 1.16. β tends to be greater than one in remote regions where an increase in NOx emissions decreases the NOx lifetime (such as through feedback on O3 and OH). In polluted regions, β tends to be less than one since an increase in NOx emissions consumes OH and increases the NOx lifetime. Changes in the NO2/NOx ratio partially compensate for changes in the NOx lifetime since increases in HOx both increase the NO2/NOx ratio and decrease the NOx lifetime. A simulation at higher spatial resolution would better resolve nonlinear NOx chemistry and heterogeneous emission sources and may yield more spatial variation in β. Outside of winter, the seasonal variation of β is <5% and interannual variation is <3%. Increasing anthropogenic VOCs and CO by 15% increases global values of β by 2.8% and 1.0%, respectively. The calculation of β is most accurate for regions with homogeneous emission changes; perturbing NOx emissions for a single grid cell in Ohio affects β in neighboring grid cells by 2–6%. Zhang et al. [2007] directly compared trends over China in bottom-up NOx emissions with satellite NO2 columns and found a larger trend in satellite NO2 than in NOx emissions; it appears that accounting for β could help explain the discrepancy since β values are less than one over regions of China with the largest NO2 columns. A nested simulation at equation image° × equation image° also yields similar β values for East China (Q. Zhang et al., manuscript in preparation, 2011).

[13] We use monthly β values to translate the changes in SCIAMACHY tropospheric NO2 columns to the changes in monthly NOx emissions. The annual changes in NOx emissions are then combined with available bottom-up NOx emissions Ei for the year i to predict emissions Ej for the year j:

equation image

We partition the top-down NOx emissions according to the spatial distribution of the sources in the bottom-up inventory to derive the anthropogenic component of the predicted emissions. The error due to partitioning is minimized here by limiting our analyses to grid boxes with large tropospheric NO2 columns (>1 × 1015) and with anthropogenic sources dominating (>50%) total NOx emissions. This also reduces errors in the relation between NOx emissions and NO2 columns. With these criteria we retain data for only 14% of land areas, but they represent 80% of anthropogenic emissions over land and 74% of total anthropogenic emissions. No assumption is made for trends in the remaining 26% of emissions. Below we demonstrate how the 2006 inventory hindcasted to 2003 using SCIAMACHY observations compares with the bottom-up inventory for 2003, and then proceed to predict emissions for 2009.

[14] The left column of Figure 2 shows the spatial variation of bottom-up and predicted NOx inventories of anthropogenic emissions for 2003. The spatial distribution of the two inventories is highly consistent (r = 0.90, N = 2328). The predicted NOx inventory (17.0 Tg N Y−1) is 6.0% lower than the bottom-up (18.1 Tg N Y−1) for regions dominated by anthropogenic NOx emissions. The two inventories exhibit regional differences of 1.7% over North America and 8.9% over OECD (Organisation for Economic Co-operation and Development) Europe, within the uncertainty in bottom-up emissions of 25% over these regions [Vestreng et al., 2009; C. Hogrefe, personal communication, 2008]. Including changes in anthropogenic VOCs and CO in the calculation of β decreases the hindcast inventory by <1%.

Figure 2.

(left) Annual anthropogenic NOx emissions. (top) The bottom-up inventory for the year 2003. Inventories predicted from SCIAMACHY observations are shown for the years (middle) 2003 and (bottom) 2009. White areas indicate where anthropogenic sources contribute <50% of total NOx emissions and tropospheric NO2 columns are <1 × 1015 molec cm−2. Inventory totals refer to colored regions. (right) The changes in SCIAMACHY-derived anthropogenic NOx emissions during (top) 2003–2006 and (bottom) 2006–2009.

[15] Figure 2 (right) shows the difference between the bottom-up inventory for 2006 and the hindcast inventory for 2003 which indicates the change in NOx emissions inferred from the SCIAMACHY data. The top-down emission changes are broadly consistent with the changes in the bottom-up inventory presented in Figure 1 (bottom). Differences with the bottom-up inventory are shown in Figure S2 of the auxiliary material. Both show significant reductions over the eastern United States and parts of Europe, and increases over eastern China. However, the hindcast inventory exhibits larger spatial heterogeneity and stronger emissions growth of 12.7% compared with 5.5% in the bottom-up inventory. While both inventories consistently suggest a negative trend from 2003 to 2006 over Western and Central Europe, they often yield an opposite trend in the rest of Europe, implying that the current knowledge about emissions in Eastern European countries may be inadequate. Apart from some inconsistencies in southeastern China, the predicted emissions are in rather close agreement with the bottom-up inventory in East Asia, where bottom-up and predicted inventories suggest the increase of 22% and 21%, respectively. Chinese NOx emissions increase by 28% during 2003–2006 at a 9.3% annual growth rate in bottom-up inventory, in close agreement with the increase of 24% at a 8.0% annual growth rate in the predicted inventory. Varying the tropospheric NO2 threshold from 1 × 1015 molec cm−2 to 5 × 1015 molec cm−2 increases the growth in bottom-up Chinese emissions by 28%–31% compared to 24%–39% in the predicted inventory, implying spatial variability in emissions growth. These results are in line with previous studies on Chinese NOx emissions [Richter et al., 2005; Zhang et al., 2007, 2009].

[16] Figure 2 (bottom) presents a forecast of anthropogenic NOx emissions for the year 2009. The predicted 2009 NOx inventory (20.9 Tg N Y−1) is 9.2% higher than the bottom-up 2006 inventory (19.1 Tg N Y−1), with most of the increase arising from East Asia. Changes in anthropogenic NOx emissions during 2006–2009 indicate a decrease of 5.7% in North America and an increase of 18.8% in East Asia, with a 6.7% annual growth rate in Chinese NOx emissions.

5. Conclusions

[17] We developed a method to apply changes in satellite observations of the tropospheric NO2 column for timely updates to bottom-up anthropogenic NOx emission inventories. We retrieved tropospheric NO2 columns from SCIAMACHY for 2003–2009, and to interpret these observations we developed a global simulation capability for GEOS-Chem at a global resolution of 1° × 1.25°. The local annual scale factor was determined by examining the response of NO2 columns to a small perturbation in anthropogenic NOx emissions using the GEOS-Chem model. We combined the SCIAMACHY inferred NOx emissions changes each year during 2003–2009 with the 2006 bottom-up inventory to hindcast emissions for 2003 and to forecast emissions for the year 2009. The forecast inventories for 2007–2009 serve as a temporary dataset until bottom-up inventories are developed to represent those years.


[18] This work was supported by NASA's Atmospheric Composition Program and by Environment Canada.

[19] The Editor thanks three anonymous reviewers.