Inverse estimation of NOx emissions over China and India 2005–2016: contrasting recent trends and future perspectives

Bottom-up emission inventories can provide valuable information for understanding emission status and are needed as input datasets to drive chemical transport models. However, this type of inventory has the disadvantage of taking several years to be compiled because it relies on a statistical dataset. Top-down approaches use satellite data as a constraint and overcome this disadvantage. We have developed an immediate inversion system to estimate anthropogenic NOx emissions with NO2 column density constrained by satellite observations. The proposed method allows quick emission updates and considers model and observation errors by applying linear unbiased optimum estimations. We used this inversion system to estimate the variation of anthropogenic NOx emissions from China and India from 2005 to 2016. On the one hand, NOx emissions from China increased, reaching a peak in 2011 with 29.5 Tg yr−1, and subsequently decreased to 25.2 Tg yr−1 in 2016. On the other hand, NOx emissions from India showed a continuous increase from 2005 to 2016, reaching 13.9 Tg yr−1 in 2016. These opposing trends from 2011 to 2016 were −0.83 and +0.76 Tg yr−1 over China and India, respectively, and correspond to strictly regulated and unregulated future scenarios. Assuming these trends continue after 2016, we expect NOx emissions from China and India will be similar in 2023, with India becoming the world’s largest NOx emissions source in 2024.


Introduction
Emission inventories are important datasets for understanding air quality. Air pollutant emissions are estimated using a bottom-up method, in which fuel consumption, emission factors, and removal efficiency are multiplied. A disadvantage of the bottomup method is the time lag in compiling the dataset because the method relies on statistical data. Emissions change dramatically due to changes in the economy, regulation, and technology, along with numerous other factors. Therefore, the latest emissions estimates are urgently needed for understanding air quality and for more accurate data to generate chemical transport models. Accurate emissions estimates are particularly important for Asia, where the anthropogenic emissions have already exceded those of Europe and the US Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence.
Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Shanghai World Exposition from May to October 2010 (e.g. Hao et al 2011), and the Asia-Pacific Economic Cooperation (APEC) meeting held in Beijing in November 2014 (e.g. Sun et al 2016).
To overcome the disadvantages of bottom-up emission inventories, the main approach is inverse estimation, which optimizes emission inventories based on forward model simulation with observational constraints. Recently, many studies have proposed inverse estimation with advanced data assimilation techniques, including four-dimensional variational assimilation (4D-Var) (Muller and Stavrakou 2005, Yumimoto and Uno 2006, Kurokawa et al 2009, and ensemble Kalman Filter (Yumimoto and Takemura 2011, Miyazaki and Eskes 2013. We previously estimated Chinese CO emissions based on a combination of Green's function method with the tagged CO tracer method (Yumimoto et al 2014). Because of the relatively light computational burden of this inverse method by Green's function, long-term CO emissions were inversely estimated and we observed a decreasing trend in Chinese CO emissions consistent with groundbased observations. The mass balance method combined with the 4D-VAR method has been tested from the perspective of computational demand (Cooper et al 2017, Qu et al 2017. Long-term inverse estimations based on a unified methodology would allow better understanding of atmospheric pollutant behavior. In this study, we estimate the long-term NO x emissions using the newly developed inversion technique with light computational burden because NO x is a key air pollutant that plays a critical role in atmospheric chemistry and is a precursor of ozone (O 3 ), which causes short-lived positive radiative forcing. High O 3 concentrations are also an important environmental problem in Asia (Nagashima et al 2010, Itahashi et al 2013.
Space-based observations of NO 2 vertical column density (hereafter, the NO 2 column) have been used as a proxy for NO x emissions (e.g. Martin et al 2006, Kurokawa et al 2009. This space-based view of the NO 2 column was first reported by Richter et al (2005) (Irie et al 2016). Recently, this increasing trend has changed rapidly. The NO 2 column observed by the Ozone Monitoring Instrument (OMI) during the 11 year period from 2005 to 2015 revealed an approximately 50% increase since 2009 that peaked in 2011, and then a sharp drop of around 40% from 2014 to 2015 over the north China plain, which is the most industrialized and populated region in China. In contrast, above central or southern China in the Yangtze River Delta and Sichuan Basin, emissions reached a peak in 2010, but remained almost constant thereafter (Krotkov et al 2016). The analysis of the peak year of estimated NO x emissions for each province also showed spatial dispersion (van der et al 2017). Recently, bottom-up emissions over China have been estimated to be catching up with these dramatic trends. The emission inventory of the Multi-resolution Emission Inventory for China (MEIC) developed by Tsinghua University (Li et al 2017b, Zheng et al 2018 is available up to 2017. Updates of NO x emissions in China are also provided by the GlobEmission project using a top-down approach (Ding et al , 2018. Another important source of emissions in Asia is India. The rapid economic growth and consequent demand for electricity has increased NO x emissions from the power sector by at least 70% from 1996 to 2010 (Lu and Streets 2012). Total NO x emissions from India were estimated as 9.3 Tg yr −1 in 2010, which is approximately 1/3 those of China, making India the third largest emissions source after China (29.0 Tg yr −1 ) and the US (13.4 Tg yr −1 ) according to the Task Force Hemispheric Transport of Air Pollution version 2.2 inventories (Janssens-Maenhout et al 2015). In Asia, China also dominates SO 2 emissions, which is another important air pollutant that poses health risks and is a precursor of sulfate aerosol. A key finding is that, due to regulations introduced in China since the early 2000s, China has now been overtaken by India as the world's largest source of SO 2 emissions and India's emissions continue to grow (Li et al 2017a). In this study, we inversely estimate NO x emissions over China and India to determine the current changes in emissions over Asia and provide future perspectives on these emission trends.

Inversion modeling
We developed an inversion system (Yumimoto et al 2015) for NO x by extending the method proposed by Martin et al (2006) and Lamsal et al (2011). Based on the linear relationships between NO x emissions (E) and NO 2 columns (Ω), Martin et al (2006) estimated NO x emissions with coefficient α taken from the model simulation by using the relationship a = W ( ) E . 1 For example, our previous study also used the linear relationships between NO x and NO 2 , and we estimated NO x emissions from China . In the inversion system developed in the present study, we assumed that the linear relationships between NO x and NO 2 can be obtained over the specific range of emissions variation from E 0 ; hence, Coefficient β, which links the NO x emissions and NO 2 column density, was obtained via the Community Multi-scale Air Quality (CMAQ) sensitivity simulation. We evaluated this coefficient by setting the increase in NO x emissions as 20% in the sensitivity simulation.
The NO x emissions and NO 2 column in the base-case simulation are represented by E f and Ω f , and those in the sensitivity simulation are represented as Then, the NO 2 column was estimated based on the linear unbiased optimum estimation by assimilating the observed NO 2 column. We used OMI observations from the Dutch OMI tropospheric NO 2 (DOMINO) data product version 2.0 (Boersma et al 2011) to constrain the NO 2 column. Level 2 swath data from the DOMINO product was applied with limits of clear sky conditions and good data-quality flags. The observed and modeled NO 2 column in the base-case simulation are Ω o and Ω f , respectively. The errors were assumed to follow a lognormal distribution. Error ε 0 includes errors from the measurements and model, and ε f is the background error. Therefore, the a posteriori column density (Ω a ) and error ( We used an additional treatment for Ω f of the modeled NO 2 column in the base-case simulation. Our previous study (Yumimoto et al 2015) found that the biases caused by the errors in the modeling system can introduce artificial biases in the inversely estimated emissions. To reduce the effect of the model bias, we where B is the estimated model bias. Finally, based on the analogous relation between the a priori (Ω f ) and a posteriori (Ω a ) estimated column densities in equation (3), a posteriori emissions (E a ) corresponding to a posteriori Ω a were calculated As the a priori emissions (E f ), we used the Regional Emission inventory in ASia (REAS) version 2. f For observation error, we used the information in the observation dataset (retrieval uncertainty) and the representation errors estimated from the standard deviation of the grid averaging. As shown in the following section, Chinese anthropogenic emissions rapidly increased from 2009 to 2011, reflecting economic development. Using the REAS emissions in 2008 as a priori emissions in this period may cause negative bias in the a posteriori emissions because the a priori emissions are much lower than the actual emissions. To overcome this limitation, for Chinese emissions from 2009 to 2011, we developed the sequential update technique, in which the a posteriori emissions in the previous year are set as the a priori emissions (i.e. perform the inverse modeling in 2010 with the a posteriori emissions in 2009 as the a priori emissions).
Our inversion system has four new features. First, the method is simple for updating and extending the bottom-up emissions quickly with relatively light computational burden. Second, the errors of the model and observations are included in the inverse estimation by applying a linear unbiased optimum estimation. Third, the model bias is considered by introducing it into the modeled NO 2 column. Fourth, the sequential update technique captures the rapid increase of Chinese emissions. The consideration of the model bias by introducing B and the development of the sequential update technique are an extension of previous works by Martin et al (2006) and Lamsal et al (2011).
Our previous study (Yumimoto et al 2015) partly applied this inverse modeling system to extend the bottom-up emissions and to investigate the long-term trend in China during 2009-2012. The tropospheric NO 2 columns predicted with the estimated emissions reproduced the spatial distribution, seasonal cycle, and interannual variation in the observations, and achieved better agreement with the satellite-observed columns than those predicted with the bottom-up emissions.

Forward model
The forward model simulation was performed with the regional chemical transport model simulation of the . Lateral boundary conditions were obtained from the monthly mean of the global chemical transport model called CHASER (Sudo et al 2002). This modeling system was used to study the air quality in Asia in our previous work (e.g. Morino et al 2015Morino et al , 2017. In addition to these base-case simulations, sensitivity simulations were conducted by increasing all NO x emissions sources by 20% to obtain the sensitivity coefficients (β in equation (4)  The variation over China is shown in figure 1, and we focused on eastern China, particularly over the centers of population and economic activity. Figure 2 shows the annual trends from 2011 to 2016. Generally, there were decreasing trends, although there were some increasing trends over parts of Sichuan Province, Fujian Province, and northeast China. The observed and modeled NO 2 column showed similar variations and the trends with the largest changes of greater than −10%/ year were centered on eastern China (i.e. Henan, Hubei, and Anhui Provinces). The inverse estimated emissions on the province scale showed trends of around −3%/ year and there were large differences according to location. Our inverse estimate of NO x emissions from China as a whole showed the peak in 2011, whereas the estimation by Zheng et al (2018) showed the peak in 2012 (figure 1). Figure 3 shows the peak year of the observed and modeled NO 2 column and inverse estimated NO x emissions on the province scale in China.
The peak year of the observed and modeled NO 2 column corresponded well; generally, the peak year was around 2011 (light orange), Beijing and Guangzhou exhibited an earlier peak between 2005 and 2008 (blue to green), and northeast China exhibited a later peak between 2013 and 2016 (red).

Continuous increase over India
The observed NO 2 column and estimated NO x emissions over India with estimated error are shown in       and table 1). The current decreasing trends are consistent with the projected regulation in the PC1 scenario with decreases of 0.7 Tg yr −1 (3.9%/year). In contrast, the increasing trends over India match the unregulated scenarios of 0.7 Tg yr −1 (5.5%/year) for BAU, 0.3 Tg yr −1 (2.9%/year) for CLE, and 0.6 Tg yr −1 (4.2%/ year) for REF. Based on the confirmation of our recent estimation trends over China and India with various future scenarios, we assume continuous trends of −0.83±0.14 Tg yr −1 over China and +0.76±0.16 Tg yr −1 over India after 2016. Under this assumption, NO x emissions from China and India will be similar in 2023 with around 19-20 Tg yr −1 , and NO x emissions from India will surpass those from China in 2024. Based on the range of the trends, the excess NO x emissions from India rather than China will occur between 2022 (maximum decrease of −0.97 Tg yr −1 over China and maximum increase of +0.92 Tg yr −1 over India) and 2025 (minimum decrease of −0.69 Tg yr −1 over China and minimum increase of +0.60 Tg yr −1 over India). Because the US, which is the second largest source, has shown a decline of −0.65 Tg yr −1 during 2011-2016 (https://epa.gov/air-emissions-inventories/airpollutant-emissions-trends-data) or even in the possible slowdown trend (Jiang et al 2018), India will become the world's largest NO x emission source. These are the first perspectives for future emission changes to two importance sources in Asia, and we should continue to monitor these changes closely.

Conclusion
We developed an immediate inversion system to estimate NO x emissions using the constraint of NO 2 column density satellite observations. Our inversion system, which is based on the linear unbiased optimum estimation, improves upon previous studies because it can be used to update emissions quickly and simply, it considers errors for both the model and observations, and it includes model bias. In this study, we used our system for the inverse estimation of longterm variation of NO x emissions over China and India from 2005 to 2016. The results showed that NO x emissions from China peaked in 2011, and subsequently declined by around 1.0 Tg yr −1 (around 3%/ year), consistent with other estimates. Compared with the large variation found over China, NO x emissions from India increased continuously by 0.1-0.6 Tg yr −1 (1 to 6%/year). Based on these contrasting trends over China and India, we predicted the following future trends in NO x emissions over Asia. Our inverse estimation indicated a decrease of 0.83 Tg yr −1 over China and an increase of 0.76 Tg yr −1 over India from 2011 to 2016, corresponding to strictly regulated and unregulated future scenarios, respectively, as reported in several studies. Assuming continuous trends after 2016, we expect that NO x emissions from India will surpass those from China in 2024, and India will become the largest global NO x emission source.

Acknowledgments
This work was partly supported by the Global Environment Research Fund (No. S-12, 5-1903) of the Ministry of the Environment, Japan. This work was also funded by a collaborative research program through the Research Institute for Applied Mechanics (RIAM) at Kyushu University (No. 30 AO-14, 2019 AO-11). This work was also supported by JSPS KAKENHI JP16H02946. Part of the research was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration. The authors acknowledge T Noguchi for his technical support.

Data availability statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.