Spaceborne evidence for significant anthropogenic VOC trends in Asian cities over 2005–2019

Trends of formaldehyde (HCHO) linked to anthropogenic activity over large cities located in the Asian continent are calculated for the period 2005–2019 using the Quality Assurance for Essential Climate Variables dataset from the Ozone Monitoring Instrument aboard the Aura satellite. Contributions due to anthropogenic emissions are isolated by applying a correction based on near-surface temperature in order to account for interference from local biogenic emissions. Strong positive trends are derived over the Middle East and the Indian subcontinent (up to 3.6% yr−1 and 2.4% yr−1 respectively) where regulations of anthropogenic non-methane volatile organic compound (NMVOC) emissions are currently limited. Weaker trends are observed over cities located in China, where the air pollution action plan (2013) may have mitigated NMVOC trends early on, but targeted legislature concerning VOC emissions was only recently introduced. HCHO trends for cities located in South and Equatorial Asia are mostly not significant or very uncertain. Cities located in Taiwan and Japan (regions in Asia where legislation has been in place since the early 2000s) display mostly negative trends.


Introduction
Anthropogenic non-methane volatile organic compounds (AVOCs) are key atmospheric constituents affecting air quality and the climate through their role in the formation of ozone and secondary organic aerosols. Emissions of AVOCs are significant in urban and industrialized areas and their major sources include fossil fuel and biofuel production and use, industrial processes, solvent use, and road transport (Ehhalt et al 2001). Bottom-up estimates indicate an increase of ca. 40% of the global AVOC emissions between 1970 and 2012, with Asia being the largest contributor (Huang et al 2017). However, regional and global AVOC emission inventories have large uncertainties, due to the strong spatiotemporal variability and high variety of their sources, and to the relative lack of accurate information, particularly in Asia. In comparison, NO x emissions and trends are relatively better quantified, to a large extent thanks to the monitoring of tropospheric NO 2 columns from satellites (Duncan et al 2016). Due to the harmful effects of AVOCs on human health and the environment, those compounds are increasingly attracting the attention of policy makers (He et al 2019, Zhao et al 2020. This has led to a gradual tightening of VOC emission regulations, especially in the US, Europe and Japan. However, these regulations are very diverse across Asian countries. Whereas Japan and Taiwan promulgated AVOC emission controls since the early 2000s (Botta and Yamasaki 2020, EPA-Taiwan (Environmental Protection Administration) 2020), China and Korea formulated VOC emission regulations only recently (Tsai 2016, China VOCs Management 2021. In India, clean air action plans for cities are currently developed (Bhave andKulkarni 2015, Ganguly et al 2020) but no VOC regulations have been adopted so far, except for benzene (Sekar et al 2019). In the Middle East and central Asian countries, regulations are limited and emission standards vary among countries (Sekar et al 2019). The success of VOC regulations largely depends on monitoring, assessing and controlling strategies, which are country-dependent (San et al 2018). In addition to regulations, factors like population growth, economical and industrial development, and new technologies, strongly influence the AVOC emissions and their evolution. The estimation of emissions and their trends becomes therefore very challenging.
A detailed comparison between available bottomup and top-down inventories (Elguindi et al 2020) showed a general agreement that AVOC emissions from China are increasing, although there are important discrepancies among the reported trends. Similarly, the REASv3 bottom-up inventory over Asia (Kurokawa and Ohara 2020) reported that most inventories indicated increasing trends over China after the early 2000s. Most inventories showed continuously increasing AVOC emissions over India while for Japan, emissions have decreased in recent decades. Despite these similarities in AVOC emissions across different inventories, large differences are also found.
Spaceborne observations of formaldehyde (HCHO), a high yield intermediate in the oxidation of the majority of VOCs released into the atmosphere, have been widely used to constrain the VOC emissions from biogenic and biomass burning origin at global and regional scale (Millet et al 2008, Barkley et al 2013, Bauwens et al 2016. However, the application of inverse methods to derive AVOC emissions is challenging due to the generally weak HCHO signal due to anthropogenic sources relative to the biogenic and pyrogenic emissions. A few studies focused on heavily polluted regions where AVOCs are expected to have a strong contribution to the HCHO signal. Barkley et al (2017) reported significant positive HCHO trends in anthropogenic point sources in the Middle East over 2005-2014 using observations from the Ozone Monitoring Instrument (OMI, Levelt et al 2006). OMI HCHO observations were used to detect urban and industrial plumes in Nigeria (Marais et al 2014), eastern Texas (Zhu et al 2017), and eastern China (Shen et al 2019), and as constraints in inversion studies over China (Stavrakou et al 2017, Cao et al 2018 and India (Chaliyakunnel et al 2019).
Using spaceborne observations to monitor the efficacy of environmental regulations for controlling VOCs is hampered by the masking effect of other VOCs sources-primarily terrestrial vegetation (Palmer et al 2006, Stavrakou et al 2009. Nevertheless, this study shows that emission trends can be accurately monitored based on HCHO column data retrieved from the OMI satellite instrument over 15 years, from 2005 to 2019 around large cities in Asia. Since the interannual variability of HCHO columns is affected by climate variability (Palmer et al 2006), primarily due to the dependence of biogenic VOC (BVOC) emissions on temperature and solar radiation, and because this contribution might be significant even around large cities, we subtract the contribution of climate variability to the long-term observed HCHO trends using the temperature correction method suggested in Shen et al (2019). This study focuses for the first time on the detection of AVOC trends over a large number of Asian cities using satellite data, and aims to complement the current body of literature on AVOC emissions in Asia (e.g. Zheng et al 2018, Kurokawa andOhara 2020).
The satellite-based urban HCHO trends are discussed and compared to the estimated AVOC emission trends in bottom-up inventories and to the respective OMI NO 2 trends calculated over the same cities. VOC and NO x are the key ingredients in the formation of tropospheric ozone (Calvert et al 2015), of which abundances have been generally increasing over East and South Asia in the last decades (Nelson et al 2021). Although road transport and other sectors contribute significantly to both AVOC and NO x emissions, the AVOC/NO x emission ratio is sectordependent and some of the major sources of NO x and AVOC (power generation and solvent use, respectively) are markedly different (Li et al 2017). Moreover, emission controls might affect AVOC and NO x differently, resulting in important differences regarding their temporal evolution.

Methodology
We use HCHO and NO 2 data retrieved from the OMI nadir-viewing instrument, both developed in the framework of the EU FP7 project Quality Assurance for Essential Climate Variables (Boersma et al 2017 for the period 2005 to the present. OMI follows a sun-synchronous, low-Earth orbit with an Equator local overpass time of ca. 13:45 and has a resolution of 13 × 24 km 2 at nadir (Levelt et al 2006). The HCHO data set is based on the differential optical absorption spectroscopy algorithm described in De Smedt et al (2018). Here we use observations with cloud fractions lower than 20% in order to minimize potential effects from clouds. The NO 2 data set is based on revised spectral fitting features, accounting for improved absorption cross sections, instrument calibration, and surface effects (Boersma et al 2018, Zara et al 2018. The data are processed according to the data quality recommendations with a cloud filter set at 40%. For monthly mean OMI HCHO and NO 2 column data, the detection limit is estimated at 2.5 × 10 15 molec cm We focus on the summer months, from May to September, as the prompt oxidation of VOCs in this season provides a stronger HCHO signal compared to winter, where HCHO levels are lower due to the weak photochemical source of HCHO. Furthermore, the retrieval errors for both HCHO and NO 2 are lower in summer, and are estimated to 40%-60%, and 20%-55%, respectively, for the cities considered here. We selected cities with a population of over 500 000 inhabitants for China and India, or over 300 000 inhabitants for other Asian countries based on the Geonames geographical database (www.geonames.org, last access: 10 November 2021). We select cities where (a) the AVOC emission flux based on the EDGARv4.3.3 emission inventory (Huang et al 2017) over May-September is higher than a threshold value, taken equal to 6 g m −2 , and (b) the ratio of anthropogenic to BVOC emissions is higher than 3. Those thresholds were chosen to ensure that the AVOC emission of the city is sufficiently important and is significantly larger than the local biogenic emissions. The biogenic emissions (sum of isoprene, monoterpenes, and methanol) are obtained from the MEGAN-MOHYCAN model (Müller et al 2008, Stavrakou et al 2018. The above criteria lead to a total of 133 cities in 31 Asian countries for further analysis (listed in table S1 (available online at stacks.iop.org/ERL/17/015008/mmedia)). To generate the OMI HCHO time series for these cities, all observations in an overpass area of a 30 km radius around the city centers are averaged per month. The monthly mean is not considered when less than five observations are available in a given month. A smaller radius (e.g. 15-20 km) would strongly reduce the amount of available observations per month and would lead to noisier monthly averages. Moreover, in order to remove potential interferences from fire emissions, we exclude specific regions and months with fire episodes. Such regions are the North China Plain (32 • -40 • N, 112.5 • -120 • E) in June and the Indo-Gangetic Plain (26 • -34 • N, 70 • -86 • E) in May, due to agricultural residue burning (Liu et al 2015, Bray et al 2019, and the entire equatorial band (10 • S-10 • N) in September, which coincides with the beginning of the fire season.
To take out the effects of climate variations on the interannual variability of the HCHO columns, we apply a temperature-correction method on the monthly HCHO columns following the approach of Zhu et al (2017) and Shen et al (2019). To this purpose, the monthly mean HCHO columns for the selected cities between May and September over 2005-2019 are regressed onto the monthly averaged daily maximum surface air temperature, obtained by the ECMWF ERA5 (European Centre for Medium-Range Weather Forecasts Reanalysis Fifth Generation) reanalysis fields (Hesbach et al 2020). The fitted temperature dependency is then subtracted from the original time series. Trends are calculated using a linear regression of the annual May-to-September mean corrected columns and are considered significant when their absolute value exceeds their 1 − σ uncertainty.
As for the HCHO columns, we use OMI NO 2 data within 30 km of the city centers of the selected cities (table S1) between May and September for all years of the 2005-2019 period and derive urban NO 2 trends using a linear regression method.

Results
The HCHO trends for the selected cities are illustrated in figure 1, whereas the trends detected as being significant are detailed in figure 2. Of the 133 studied cities, significant trends are found for 77 cities (table S1). Positive HCHO trends are found for most urban areas, and clear regional patterns can be identified.
The trends observed over China are relatively weak and mostly positive, typically between +0.5% yr −1 and +1.5% yr −1 and negative in only two cities, namely Dongguan in the Pearl River Delta and Chengdu in western China. In the Indian subcontinent, strong positive trends are observed, estimated on average at 1.4% yr −1 and ranging between 0.7 ± 0.5% yr −1 (Karachi) and 2.4 ± 0.7% yr −1 (Peshawar). Positive trends are also found over the Middle East and central Asia reaching up to 3.6 ± 0.6% yr −1 (Karaj, Iran). No significant trends were observed over Istanbul and Ankara in Turkey.
Trends are often not significant in southeast and equatorial Asian countries, except over Bandung (Indonesia), Cebu city (Philippines) and Hanoi (Vietnam), with trends exceeding 2% yr −1 .
Significant negative trends are found in Japan, estimated on average at −1.5% yr −1 for all considered cities. In Taiwan, the HCHO trends are also mostly negative, except in Taichung. Figure 3 compares the temporal evolution of (temperature-corrected) HCHO columns and NO 2 columns, relative to 2005, for a selection of 25 cities. The 2005-2019 NO 2 trends for all 133 cities are shown figure S1. The Chinese megacities generally exhibit strong reductions in NO 2 levels between 2005 and 2019 (up to −60%). For most cities, NO 2 columns increase in the first part of the period (2005-2010/2014) and decline afterwards. Over large industrial centers such as Shanghai, Guangzhou and Shenzhen, the NO 2 decline started already after 2007. This is in contrast with the HCHO column evolution in Chinese cities, with generally either positive (figure 2) or negligible trends (e.g. Shanghai and Guangzhou, figures 3(c) and (d)).
A notable exception is Chengdu, where the HCHO trend is negative (−1.2 ± 0.6% yr −1 ). The net reduction over Chengdu between 2005 and 2019 is 20% for HCHO and 60% for NO 2 .
Over large Indian and Pakistani cities (figures 3(k)-(o) and (w)), NO 2 columns display only small trends between 2005 and 2019, and declines are even found after 2013 in Delhi and Chennai, whereas HCHO columns increased over most cities. Over

Discussion
The HCHO trends over China presented above for 2005-2019 are consistent with the predominantly positive anthropogenic trends derived over China between 2005 and 2016 by Shen et al (2019) (up to +2.2% yr −1 ). In line with their analysis, stronger trends are observed over the Yangtze River Delta and the North China Plain. The important reduction in HCHO and NO 2 observed over Chengdu and in the Pearl River Delta (Dongguan) shows that local efforts to counteract pollution are also evident from space observations. In those regions, the implementation of air pollution control policies has shown improvements in recent years (Gao et al 2019(Gao et al , 2021, whereas AVOCs have become one of key pollution control targets (Gao et al 2019). In August 2017, an Air Pollution Prevention campaign, addressing AVOC emissions, was performed in Chengdu leading to immediate AVOC emission reduction of 25% (Tan et al 2020), likely contributing to the HCHO column decline observed over Chengdu in 2017 ( figure 3).
In India and Pakistan, the strong positive HCHO trends are in line with the strong economic growth and near-absence of VOC emission regulations in the fast growing cities of the Indian subcontinent (Ganguly et al 2020). The strong trends observed over Delhi (+2.1 ± 0.6% yr −1 ) and Kanpur (+1.2 ± 0.5% yr −1 ) are consistent with the independent satellite-based analysis of Vohra et al (2020), which derived similar HCHO trends (+1.9% yr −1 for Delhi, +1.0% yr −1 for Kanpur). Note that their work corrected for the background contribution of longerlived VOCs but not for biogenic contributions. Furthermore, our results for the Middle East are qualitatively in line with Barkley et al (2017), who also deduced strong HCHO trends in urban locations over 2005-2014 using OMI observations. In agreement with our analysis, they reported significant HCHO trends along the Persian Gulf coast, between 1% and 3% yr −1 , and in Riyadh (1.97% yr −1 vs 2.5 ± 0.7% yr −1 in this study). In Bagdad, they found an increased NO 2 trend, not seen in HCHO data, which was attributed to increased NO 2 emission from the Daura refinery located close to the city (Barkley et al 2017).
In south and equatorial Asia, the HCHO trends are more uncertain and are significant in only 4 out of the 16 cities considered. This can be explained by the comparably lower HCHO levels and larger associated uncertainties in these regions and by the strict cloud filter used (20%) which removes a large fraction of the data. In addition, uncertainties in the biogenic emissions around cities of equatorial Asia might cause larger errors in the derived trends. In Cebu and Bandung, although the trends are significant, their high uncertainty reflects the strong interannual column variation in these cities (+2.9 ± 1.0% yr −1 and +3.2 ± 2.6% yr −1 ).
The observed declining trend in Phnom Penh (−1.1 ± 0.6% yr −1 ) is at odds with the increase of AVOC emissions reported in San et al (2018), which is based on economic activity and industrial data. The discrepancy could be related to the large deforestation rates observed in the humid primary forests located within a radius of 150 km around the city (Hansen et al 2013). When considering the effects of land use changes on isoprene fluxes (Opacka et al 2021), the isoprene emissions around Phnom Penh over 2005-2019 are found to decline by about 1% yr −1 , suggesting that the trend in HCHO columns might be partly due to forest loss in the regions surrounding Phnom Penh.
The significant positive HCHO trend in Hanoi (+2.4 ± 0.7% yr −1 ), one of the fastest growing cities of Asia, is well supported by the increase of 36% in AVOC emissions suggested by the CAMS-ANTv4.2 R1.1 (Granier et al 2019) bottomup inventory between 2008 and 2019. Figure S1 shows relatively weak NO 2 trends over most cities in south and equatorial Asia with the exception of the strong positive trends observed over the fast-developing cities of northern Vietnam.
Finally, the negative HCHO trends in Japan and Taiwan reflect the VOC emission regulations that have been promulgated earlier on, compared to other Asian countries, in combination with a weaker population growth. Over Japan, where the population growth is <+0.5% yr −1 for all cities under consideration (United Nations Department of Economic and Social Affairs 2021), satellite observations are in line with ground-based observations of Wakamatsu et al (2013), which showed continuously declining AVOC concentrations since 1970. It was estimated that VOC emissions from stationary sources in Japan were reduced by 42% between 2000 and 2009 (Wakamatsu et al 2013). In Taiwan, we find negative HCHO trends in all cities except for Taichung. This may reflect the stronger population growth in Taichung (+1.8% yr −1 ), whereas the population growth for the other Taiwanese cities is much smaller (<+1% yr −1 , United Nations Department of Economic and Social Affairs 2021).
The observed HCHO trends agree qualitatively with the AVOC emission trends estimated by the CAMS-ANTv4.2 R1.1 (Granier et al 2019) and the REASv3 (Kurokawa and Ohara 2020) bottom-up inventories (figure 4). Both datasets suggest increasing emissions in most Asian countries and declining emissions in Japan and Taiwan. While the increasing AVOC trends in most Asian countries are consistent with the trends in figure 1, the magnitude of the trends is not always consistent with OMI observations. Over Iran, the continuous AVOC emission decrease according to CAMS (Copernicus Atmosphere Monitoring Service) is not supported by OMI, in particular in Tehran where HCHO is found to increase rapidly after 2011.
More pronounced negative HCHO trends are derived over Japan than over Taiwan, in better consistency with the emission estimates from REASv3 than those from CAMS. Additionally, our analysis suggests similar HCHO trends in India and Pakistan while those over Chinese cities are much weaker. REASv3, however, suggests similar AVOC emission trends for all three countries and CAMS estimates similar increases in AVOC emissions for India and China, but stronger trends over Pakistan (figure 4). Note however that the REASv3 inventory does not extend after 2015, when AVOC emissions over China were at their maximum according to CAMS, in good consistency with the observed evolution of HCHO columns ( figure 3).
The observed differences between trends in HCHO and NO 2 are also present in the AVOC and NO x emissions from the CANT-ANTv4.2 R1.1 inventory ( figure S2). This is especially true for China, where the Air Pollution Prevention and Control Action Plan has been introduced in 2013, but AVOCs were not specifically targeted until 2018 (China VOCs Management 2021). The differences in NO 2 and HCHO trends originate from the fact that they are not emitted by the same anthropogenic activities. For instance, solvents significantly contribute to the total AVOC emissions while they account for only a small fraction of NO x emissions (figure S3). In contrast, the energy sector strongly contributes to the NO x emissions but not to AVOCs. The discrepancies between AVOC emissions and NO x are especially pronounced in China where industrial sources and solvent use were the main drivers of the rise of AVOC concentrations while reduced emissions from residential biofuel use and on-road vehicle exhaust (which account for a large fraction of NO x emissions) mitigated the rapid growth rates in the last decade (Li et al 2019). The increase in AVOCs, in contrast with the large emission decreases of SO 2 (since 2005) and NO x (since 2011) in China, has been put forward as a potential reason for recent increases in surface ozone concentrations in China (Ma et al 2016, Li et al 2019.

Conclusion
We used OMI satellite observations from 2005 to 2019 to derive long-term trends in HCHO columns, corrected for climate variability, over 133 cities in Asia.
Significant positive trends were observed in 64 cities, and negative trends in only 13 cities. Although the general patterns of OMI column trends are consistent with the emission inventories, several important discrepancies were found. AVOC emissions appear to increase in Iran, in particular in Tehran, contradicting the decline found in CAMS. In India as well, the robust positive trend of OMI columns seems underestimated by CAMS. Strong positive HCHO trends are observed over the Middle East. Generally positive, but weaker trends are found in China. As Chinese VOC regulations were only recently implemented, HCHO trends between 2005 and 2018 do not yet validate these efforts. Nevertheless, the effect of AVOC regulations is clearly seen in two important megacities (Dongguan and Chengdu). Finally, over Japan and Taiwan, negative trends are found, demonstrating that air quality regulations targeting AVOC emissions are effective in these countries. In Japan in particular, those regulations appear even more effective than suggested by the CAMS inventory. The contrast between the observed evolution of NO 2 and HCHO columns over Asian cities illustrates well that emission controls targeting NO x sources have little or no impact on AVOC emissions, in particular over China, as the major sources of NO x and AVOC are markedly different, and as the AVOC and NO x emission factors are affected differently by emission controls. Those important differences between HCHO and NO 2 trends highlight the importance of legislation targeting specifically VOC emissions to improve air quality. The limited number of initiatives to regulate AVOC emissions in many Asian countries is evidenced by the detected positive HCHO column trends compared to countries were policy measures have been in place for a longer time, or where they were recently introduced.