Yongjoo Choi This email address is being protected from spambots. You need JavaScript enabled to view it.1, Yugo Kanaya1, Hisahiro Takashima1,2, Kihong Park3, Haebum Lee3, Jihyo Chong3,4, Jae Hwan Kim5, Jin-Soo Park6 1 Research Institute for Global Change, Japan Agency for Marine-Earth Science and Technology, Yokohama 2360001, Japan
2 Faculty of Science, Fukuoka University, Fukuoka 814-0180, Japan
3 School of Earth Sciences and Environmental Engineering, Gwangju Institute of Science and Technology, Gwangju 61005, Korea
4 Environmental Management Division, Yeongsan River Basin Environmental Office, Gwangju 61945, Korea
5 Department of Atmospheric Sciences, Pusan National University, Busan 46241, Korea
6 Climate & Air Quality Research Department, National Institute of Environmental Research, Incheon 22689, Korea
Received:
April 10, 2022
Copyright The Author(s). This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are cited.
Revised:
October 25, 2022
Accepted:
March 6, 2023
Download Citation:
||https://doi.org/10.4209/aaqr.220145
Choi, Y., Kanaya, Y., Takashima, H., Park, K., Lee, H., Chong, J., Kim, J.H., Park, J.S. (2023). Changes in Tropospheric Nitrogen Dioxide Vertical Column Densities over Japan and Korea during the COVID-19 Using Pandora and MAX-DOAS. Aerosol Air Qual. Res. 23, 220145. https://doi.org/10.4209/aaqr.220145
Cite this article:
We investigated the impact of human activity during COVID-19 on the tropospheric nitrogen dioxide vertical column density (NO2 TropVCD) at three urban sites (Gwangju and Busan in Korea and Yokosuka in Japan) and one remote site (Cape Hedo in Japan) from Multi-Axis Differential Optical Absorption Spectroscopy (MAX-DOAS) and Pandora. Compared to the monthly mean NO2 TropVCD from 2015 to 2018 and in 2019, the values were lower in 2020 due to social distancing in Korea and Japan. High negative relative changes were observed from May to September (−30% to −18%) at the three urban sites; Cape Hedo, a remote site, did not show a significant difference in relative changes between previous years and 2020, suggesting that only anthropogenic emission sources decreased dramatically. In the case of Yokosuka, the 15-day moving average of the NO2 TropVCD exhibited a good relationship with transportation (R = 0.48) and industry (R = 0.54) mobility data. In contrast, the NO2 TropVCD at the Korean sites showed a moderate to low correlation with the industrial sector and insignificant correlations with transportation. The differences in correlations might be caused by the different social distancing policies in Korea (voluntary) and Japan (mandatory). By applying generalized boosted models to exclude meteorological and seasonal effects associated with NO2 TropVCD variations, we revealed that the decreasing trend from 2019 to 2020 was much steeper than that from 2015 to 2020 (a factor of two), and a significant change was identified in January 2020, when the first cases of COVID-19 were observed in both Korea and Japan. This result confirmed that the reduction in NO2 can be largely explained by the NOx emission reduction resulting from social distancing for COVID-19 rather than annual meteorological differences; however, in December 2020, NO2 recovered suddenly to its previous level due to an increase in human activities.HIGHLIGHTS
ABSTRACT
Keywords:
COVID-19, Nitrogen dioxide, MAX-DOAS, Pandora
Nitrogen oxides (NOx = NO2 + NO) are major components of air quality degradation in urban/ industrialized areas due to their role as catalysts in tropospheric ozone formation and as precursors of secondary inorganic aerosols, with consequences for climate and human health (Crutzen, 1970; Bond et al., 2001; Myhre et al., 2013; Lelieveld et al., 2015). Typically, NOx is emitted from vegetation fires, lightning, and soils, but emissions from fuel combustion account for ~50% of global total NOx emissions (Delmas et al., 1997). Because the emission of NO2 is highly linked with human activities, the NO2 concentration is comparably higher in urban areas than in rural areas; additionally, NOx is characterized by a diurnal and weekly cycle in most countries (Beirle et al., 2003; Kanaya et al., 2014), with high NO2 emissions in rush-hour periods on working days and low NO2 emissions on weekends or public holidays. As a result of the efforts to limit the effects of deteriorated air quality on public health, emission regulation policies for NOx and other pollutants have contributed to the reduction of the NO2 tropospheric vertical column density (TropVCD), which has been confirmed by long‐term satellite and ground-based observations (Duncan et al., 2016; Geddes et al., 2016; Georgoulias et al., 2019; Choi et al., 2021). After coronavirus disease 2019 (COVID-19) first broke out in December 2019 in Wuhan, China, it became an ongoing global pandemic and was deemed a public health emergency of international concern by the World Health Organization (WHO); more than 0.17 billion cases were observed in 212 countries or regions as of 6 June 2021 (WHO, 2020). The dramatical increase in newly confirmed COVID-19 cases globally led to shrinkage in human activities in many countries, which was associated with travel restrictions, self-quarantine measures, curfews, and the closure of businesses and factories to suppress the spread of the virus (Baldasano, 2020; Bauwens et al., 2020; Berman and Ebisu, 2020). From model simulations and satellite observations, the strict regulation period for COVID-19 led to large reductions in the NO2 TropVCD compared to the level in the same period in 2019; this difference was especially notable for the transportation and industrial sectors (Bauwens et al., 2020; Ding et al., 2020; Forster et al., 2020; Koo et al., 2020; Liu et al., 2020; Miyazaki et al., 2020; Shi and Brasseur, 2020; Venter et al., 2020; Doumbia et al., 2021). The reduction in the NO2 TropVCD during the unprecedented COVID-19 period was complicated by several factors, not only changes in anthropogenic emissions rates (e.g., transportation and industrial sectors) but also chemical and meteorological seasonal cycles, generally leading to a decline in NO2 from winter to spring in the Northern Hemisphere (Martin et al., 2003; Ding et al., 2020; Kroll et al., 2020). Therefore, the reduction in the NO2 TropVCD driven by COVID-19 must be distinguished from the pre-existing trajectory, and meteorologically driven variability must be considered (Li et al., 2019; Vu et al., 2019; Kroll et al., 2020). Although the satellite-based NO2 TropVCD can provide valuable estimates over a broad spatial area, there are inherent limitations (e.g., temporal resolution and/or spatially inhomogeneous distribution) that can complement pollutant trends using ground-based measurements (Bechle et al., 2013; Berman and Ebisu, 2020; Choi et al., 2021). Ground-based remote sensing measurements for trace gases (e.g., Multi-Axis Differential Optical Absorption Spectroscopy (MAX-DOAS) and Pandora) have been regarded as ‘‘ground truth’’ data for target pollutants and provide higher sensitivity in the lower troposphere and a higher temporal resolution than satellite observations. In this study, we investigated the impacts of human activity reductions resulting from the spread of COVID‐19 on the NO2 TropVCD at three urban sites (Yokosuka, Japan and Gwangju and Busan, Korea) and one background site (Cape Hedo, Japan); these sites are directly influenced by Chinese and/or local emission sources. In that context, using MAX-DOAS and Pandora instruments, we compared the difference in the NO2 TropVCD between 2020 (whole year) and that in the previous five years in terms of human activities and meteorological variables. Additionally, we focused on the trend in the NO2 TropVCD at two different time scales (2015 to 2020 and 2019 to 2020) by decomposing seasonal and weather effects so as to precisely evaluate the NOx emission reduction. We carefully selected the four measurement sites by considering the available multiyear measurements obtained since 2015 over East Asia among the representative ground-based remote sensing measurement networks for trace gases: MAX-DOAS over Russia and Asia (MADRAS; https://ebcrpa.jamstec.go.jp/maxdoashp/) and the Pandonia Global Network (PGN). Fig. 1 shows the geographical locations of the four measurement sites and the differences in the NO2 TropVCD from ozone monitoring instrument (OMI) satellite data (OMNO2d_003) between 2019 and 2020 over East Asia, indicating that the NO2 TropVCD was reduced in most regions. The Yokosuka site (139.65°E, 35.32°N) is close to Tokyo and Yokohama, which are two of the largest cities on the Kanto Plain, Japan. According to the Regional Emission inventory in ASia (REAS) version 3.2 (Kurokawa and Ohara, 2020), the NOx emission rates were 37 Tg in Yokohama (~76% from industry and power plants) and 41 Tg in Tokyo (~49% from road transportation) in 2015. Therefore, the Yokosuka site is suitable for investigating the NO2 behaviour in urban areas with complex transportation and industrial effects (Kanaya et al., 2014; Choi et al., 2021). Cape Hedo (128.25°E, 26.87°N) is located in the northernmost part of subtropical Okinawa Island. Because there are no major anthropogenic emission sources at the site (0.02 Tg; ~98% from road transportation in REAS), and this area is distant from local cities (40 km from Nago (population of 60,000) and 100 km from Naha (population of 0.32 million)). Cape Hedo represents a background site in East Asia (Takashima et al., 2009; Kanaya et al., 2014; Choi et al., 2021). Busan (129.08°E, 35.24°N) is the second-largest city in Korea, with a population of 3.4 million within 770 km2. The Busan site is located at Pusan National University (PNU), southeastern Korea, and is in a valley surrounded by small mountains. Similar to those in Yokohama, the major NOx emission sources in Busan are industry and power plants, which account for 65% of the total NOx emissions. The Gwangju site (126.84°E, 35.23°N) is located on the campus of the Gwangju Institute of Science and Technology (GIST) in southwestern Korea. Gwangju has a population of 1.5 million people within ~500 km2 and is one of the metropolitan cities in Korea. The major NOx emission source near the site is transportation, with a similar contribution (~49%) as that in Tokyo, although the total emission rates in Gwangju in REAS v3.2 are lower than those in Tokyo by a factor of two. Information on the MAX-DOAS instrument and its retrieval algorithm has been described in detail elsewhere (e.g., Irie et al., 2008, 2011; Kanaya et al., 2014). MAX-DOAS consists of a light-receiving part and a miniature spectrometer (USB4000; Ocean Optics, Dunedin, FL, USA) connected by a fibre-optic cable bundle. The sky irradiance collected by a telescope is redirected by a prism reflector and quartz fibre to the spectrometer at six elevation angles (EAs; 3, 5, 10, 20, 30, and 90° (70° for Cape Hedo)) for spectral analysis, with a < 1° field of view (FOV). The prism reflector is sequentially rotated every 5 min for each angle; thus, the measurement interval is 30 min. Among the measured wavelength range (230 to 560 nm, with less than 0.7 nm of full width at half maximum), we chose 460–490 nm for retrieving the NO2 and oxygen collision complexes (O2-O2 or O4) column densities in this study. For Yokosuka and Cape Hedo, the five-fold optical axes were installed for simultaneous observations at different EAs, but only a single telescope was used for sequential scanning for each EA. All measured spectra were retrieved by (1) DOAS spectral fittings using QDOAS software version 3.0 and (2) conversion of the differential slant column density (ΔSCD) to TropVCD by optimal estimation methods (Rodgers, 2000). All measurement spectra were first corrected for offset and dark currents and analysed using the DOAS technique to retrieve the ΔSCD of O4 and NO2 with respect to the reference spectrum (measured at an EA of 90° or 70° within 30 min). Here, ΔSCD is defined as the difference in the SCD between the measured spectrum and the corresponding reference spectrum. The absorption cross-sections for retrieving the ΔSCD were adopted from Hermans et al. (2003) for O4 but increased by a constant factor of 1.25 (Wagner et al., 2019), Vandaele et al. (1996) for NO2 at 298 K, Bogumil et al. (2003) for O3 at 223 K, Rothman et al. (2003) for H2O, and Chance and Spurr (1997) for Ring effect. A polynomial of degree three was used to fit the continuum. The ΔSCDs of NO2 and O4 were retrieved using MAX-DOAS profile inversion algorithms with the following procedures. The profiles of aerosol optical depth (AOD) at 476 nm below 5 km were retrieved and then used as constraints to retrieve the NO2 profile below 5 km. To retrieve both the AOD and NO2 TropVCD, an optimal estimation method (Rodgers, 2000) was applied to solve the nonlinear inversion problem with an iteration equation using a lookup table of box air mass factors (AMFs) that was generated by the Monte Carlo Atmospheric Radiative Transfer Simulator (MCARaTS; Iwabuchi, 2006). In brief, the state vector consisted of the target outputs (AOD or NO2 TropVCD) and three parameters (f0–1 km, f1–2 km, and f2–3 km; f represents the fractional contribution to the remaining AOD or NO2 TropVCD from the low-altitude range) used to determine the vertical profiles. Then, each partial column (0–1, 1–2, and 2–3 km) of the target outputs was expressed as f0–1 km × output, (1 − f0–1 km) × f1–2 km × output, and (1 − f0–1 km) × (1− f1–2 km) × f2–3 km × output. The a priori values of f0–1 km, f1–2 km, and f2–3 km were 0.60 ± 0.05, 0.80 ± 0.03, and 0.80 ± 0.03, respectively. The a priori values of the AOD and NO2 TropVCD were 0.21 ± 3.0 and 20% of the largest ΔSCD values for NO2 among the five ΔSCDs during 30 mins, respectively. The overall uncertainty in the AOD has been estimated at 30% (Irie et al., 2008; Takashima et al., 2009; Irie et al., 2011) and that in the NO2 TropVCD has been reported as 17% (Kanaya et al., 2014). The minimum detection limit for the NO2 TropVCD at an altitude of 0–1 km has been reported to be < 0.2 ppbv, corresponding to 5 × 1014 molecules cm−2 (Takashima et al., 2011, 2012; Kanaya et al., 2014). Quality control was carefully conducted to eliminate data measured under abnormal conditions (e.g., changes in integration time and temperature settings, large residuals in the spectral fittings and saturated signal levels) (Kanaya et al., 2014). Pandora spectrometer was installed at Pusan National University in March 2012 to measure the total column NO2 and O3 as part of the Distributed Regional Aerosol Gridded Observation Networks (DRAGON)-Asia campaign in South Korea (Chong et al., 2018). The direct solar beam is measured by Pandora via an optical head sensor attached to a solar tracker and connected to a UV–VIS spectrometer (spectral range of 280–525 nm) by a fibre optic cable attached to the head sensor (Herman et al., 2009, 2019). The direct-sun NO2 total VCD from Pandora was retrieved by applying a spectral fitting algorithm (Cede, 2019) using a near-noon reference spectrum from which the NO2 slant column amount is derived by a statistical calibration approach (Herman et al., 2009). The direct-sun NO2 total VCD has been reported to have high precision and accuracy, and data are available in near-real time as a standard product (Judd et al., 2019); the reported uncertainties of Pandora NO2 total VCD are ~2.7 × 1014 molecules cm−2 for random uncertainty and ~2.7 × 1015 molecules cm−2 for systematic uncertainty (Herman et al., 2009). During the CINDI-2 campaign which participated 36 spectrometers, the NO2 ΔSCD from Pandora instruments at visible range showed good agreement with reference with small mean relative difference (−1.3% to 3.4%) (Kreher et al., 2020). Because the NO2 total VCD contains both stratospheric and tropospheric NO2 VCDs and most of the NO2 is located between 0 and 3 km of altitude with only approximately 2.7 × 1015 molecules cm−2 (0.1 ± 0.05 DU) in the upper troposphere and stratosphere (Dirksen et al., 2011; Herman et al., 2019), the NO2 TropVCD from Pandora was derived from the NO2 total VCD after subtraction of the stratospheric NO2 VCD estimated using OMI satellite data (Chong et al., 2018; Pinardi et al., 2020). In this study, the Pandora retrievals were screened to exclude observations with uncertainties in NO2 total VCD greater than 1.35 × 1015 molecules cm−2 (0.05 DU) and normalized root-mean square of the weighted spectral fitting residuals greater than 0.005 (Judd et al., 2019; Choi et al., 2020; Pinardi et al., 2020). It should be noted that this approach leads to the retrieval of the total tropospheric column from Pandora, while the tropospheric column from MAX-DOAS mainly represents the boundary layer below 5 km. Fig. 2 shows the monthly mean and standard deviation of the NO2 TropVCD at four measurement sites during three different time scales: 2015−2018, 2019, and 2020. The monthly mean NO2 TropVCD in 2020 was generally lower than that in 2015−2018 and in 2019 (some months were not), possibly due to social distancing in South Korea (February) and Japan (April) and the lockdown in China (early in January) to prevent the spread of COVID-19. The relative change in NO2 in 2020 compared to that in the other periods (2015−2018 and 2019) was calculated using the following equation: (NO2 in 2020 − NO2 in other periods)/NO2 in other periods. For the three urban sites, the relative changes in the NO2 TropVCD between 2015−2018 and 2020 were considerably low at −17% for Yokosuka, −25% for Busan, and −22% for Gwangju compared to those between 2019 and 2020 (−11%, −11%, and −12%, respectively), possibly because NOx emissions showed a continuously decreasing trend due to the effectiveness of regulation policies in each country (Zheng et al., 2018; Kurokawa and Ohara, 2020). Regarding monthly variations in the NO2 TropVCD, all sites showed the typical pattern of high NO2 in winter and low NO2 in summer due to the longer (shorter) lifetime of NO2 in winter (summer) associated with its photo-dissociation of NO2 depending on the amount of solar irradiation (Kanaya et al., 2014; Kroll et al., 2020; Choi et al., 2021). The decreased magnitudes were pronounced in May to September with high relative changes (−30% to −18%) compared to those in other months except for March and November (−3.3% to −16.4%). This finding could be partially explained by low emissions of NO2 due to the decrease in human activity, such as activity in the transportation and industry sectors; however, the COVID-19-driven decline must be separated from these pre-existing trends, and meteorologically driven variability must be considered (Li et al., 2019; Vu et al., 2019; Kroll et al., 2020). The difference in temperature and wind speed between 2020 and 2015 to 2019, which was measured at the nearest weather stations operated by the Japan Meteorological Agency and Korea Meteorological Agency, indicated that the warm season exhibited less favourable conditions for NO2 photo-dissociation under relatively low temperatures (−0.41 ± 1.33°C) compared with the cool season (0.17 ± 1.59°C) along with high wind speeds (0.12 ± 0.23 m s−1 in the warm season vs. −0.02 ± 0.14 m s−1 in the cool season). In other words, lower NO2 levels in the warm season might not have been associated with meteorological parameters, although the difference was insignificant. The detailed effects of meteorological parameters will be discussed in Section 3.3. However, the background site, Cape Hedo, exhibited increasing trends from 2015−2018 (14%) and in 2019 (12%), suggesting that only anthropogenic emission sources decreased dramatically and that the influence of NO2 emissions was very local due to its short lifetime under relatively high temperatures (~21°C for the annual mean temperature). Hereafter, we focus more on the three urban sites in Korea and Japan. Fig. 3 presents time series of the 15-day moving average (± 7 days) of the ground-based NO2 TropVCD, human activity changes, and daily newly confirmed COVID-19 cases in Korea and Japan in 2020, including the periods of the Chinese New Year holiday (red shading in Fig. 3(a)) and social distancing in Korea and Japan (blue and red shading in Fig. 3(b), respectively). Google mobility data, including transit and workplace information, were used as proxies for transportation and industry, respectively. Forster et al. (2020) reported that the transit, workplace, retail, and residential movement data from the Google mobility trend correspond well with the surface-transportation, industry, public, and residential sector emission changes reported by Le Quéré et al. (2020) based on official statistics. We used the activity data of Tokyo for Yokosuka, whereas for Busan and Gwangju, we used activity data of the non-Seoul Metropolitan Area because Google did not provide the activity data for Busan and Gwangju. We assumed that the human activities in the two urban sites in Korea would correspond well with the Google mobility data of the non-Seoul Metropolitan Area because the driving activity data from Apple, which are available for Gwangju and Busan, showed moderate correlations with transit activity from the non-Seoul Metropolitan Area (0.36 for Busan and 0.45 for Gwangju). During the Chinese New Year holiday (P2; 24−30 January 2020), the NO2 TropVCD at the urban sites in Korea decreased rapidly by approximately −7% of the relative change in Gwangju and −31% in Busan compared to that in the previous week (P1; 16−23 January) because the NO2 reduction during the Chinese New Year holiday period in China is a yearly phenomenon associated with decreases in business and industrial activities (Tan et al., 2009; Bauwens et al., 2020; Ding et al., 2020; Liu et al., 2020; Myllyvirta, 2020). Therefore, the NO2 reduction in Korea, which observes the same holiday, is reasonable. However, it should be noted that the NO2 TropVCD also decreased in Yokosuka, with a local minimum concentration after the holiday week (P3; 31 January−6 February), similar to Gwangju, although Japan does not observe the Chinese New Year holiday. The relative change of P3 to P1 in Yokosuka (−36%) was much higher than that at the two Korean sites (−15% at Gwangju and −25% at Busan) and even close to the change in China (−35%) (Ding et al., 2020). This magnitude of the decreasing trends was similar to that for the surface NO2 concentration recorded at an air quality monitoring station near Yokosuka (−34%). The NO2 reductions in this period seem to be driven solely by NOx emissions because the difference in temperature between each period at all urban sites was not significant. Typically, a rebound in NO2 begins approximately 7 days after Chinese New Year, the end of the holiday season (Liu et al., 2020). However, the local minimum concentration in P3, not P2, revealed the ongoing reduction in human activities in both Korea and Japan. Bauwens et al. (2020) also reported that the strict lockdowns in Wuhan and Nanjing in China could be mainly attributed to delaying the uptick in the NO2 concentration, unlike in previous years. After P3, the NO2 TropVCD in Yokosuka and Gwangju recovered to levels close to those before the Chinese New Year holiday (~89%) due to the stepwise resumption of work and social activities, although the NO2 TropVCD levels were lower than the January values, primarily due to NOx lifetime changes as the temperature generally increased from January to February. As new cases of COVID-19 were recognized in Korea and Japan (middle of January), the transportation and industry mobility in both countries decreased as public awareness about COVID-19 increased (Fig. 3(b)). Additionally, the NO2 TropVCD started to decrease in accordance with the low emission rates of NOx along with increased temperatures and decreased human activity. During the social distancing periods in Korea (March–May, August–October, and December in 2020) and Japan (April–May in 2020), Japanese mobility for transportation and industry significantly decreased by up to –55%, whereas Korean mobility was relatively high (up to –25%). Because the measures in Korea were mostly voluntary (spontaneous social distancing while maintaining ordinary life); whereas, the Japanese government enforced emergency declaration to stay at home orders, avoid crowded places, and avoid close contacts (Azuma et al., 2020; Tashiro and Shaw, 2020), resulting in conscious behavioural changes in these periods (Fig. 3(b)). In a similar context, the 15-day moving average mobility in Korea showed a much higher negative Pearson correlation coefficient (r) with newly confirmed COVID-19 cases (–0.49 for transportation and –0.36 for industry) compared to Japan (0.10 for transportation and –0.06 for industry), supporting that human activities in Korea were voluntarily restricted. Since February, based on the available Google mobility data, the 15-day moving average of the NO2 TropVCD showed a good relationship with the transportation (r = 0.48) and industry (r = 0.54) mobility data, indicating that emissions from both major contributors to the total NO2 burden were reduced in the Tokyo metropolitan area. In contrast, the NO2 TropVCD at the Korean sites exhibited a moderate to low correlation with only the industrial sector (0.39 for Busan and 0.19 for Gwangju), and the transportation sector did not show a significant correlation with the NO2 TropVCD. This is because voluntary-based social distancing in Korea might be led to a relatively small reduction of human activities which can blur their relationship. However, it is hard to exclude the other factors for the decreasing NO2 TropVCD in Gwangju and Busan such as chemical process, long-range transport, and changes in meteorological variables (e.g., increased temperatures) that influence the lifetime of NO2. To subtract the effects of meteorological and seasonal cycles from the NO2 TropVCD time series, the ‘‘deweather’’ package in R (Carslaw, 2018) was applied because it uses a powerful statistical technique based on boosted regression trees using the generalized boosted models (GBM) package (Ridgeway, 2020). We selected the daily averaged NO2 TropVCD at the three urban sites from 2015 to 2020 (a total of 4466 cases), and three meteorological variables (wind speed, temperature, and relative humidity) were obtained from the nearest weather stations near the three urban sites. Then, we selected four more variables that can represent the variation in the NO2 TropVCD: trend, week, day of week, and measurement site. The hyper parameters were tuned using a training (sample) fraction of 0.8 (3562 cases) and a validated fraction of 0.2 (891 cases). Fig. 4 shows the relationship between the predicted and measured NO2 TropVCD values for the validation cases using the GBM model. Two variables exhibited good correlation, with an R of 0.70 and 0.85 of best-fit lines through the origin; these values were lower than those for the training cases (R = 0.81), and the mean bias and root mean squared error were low at −0.07 and 7.18 × 1015 molecules cm−2, respectively, indicating that the developed model can reasonably reproduce the NO2 TropVCD excluding meteorological effects. Fig. 5 shows the partial dependence plots from the gradient-boosted regression models. As previously discussed, the variable with the largest influence on the NO2 TropVCD was temperature, which is a proxy of photo-dissociation, and the next-most influential variables were the site, wind speed, trend, day of week, RH, and week of year. In the case of temperature, the NO2 TropVCD peaked at ~5°C, not below zero, which could be suppressed condition of photo-dissociation. Considering the similar latitudes of the three urban sites, the low NO2 TropVCD at low temperatures could be interpreted as the effect of air masses from the north passing through relatively clean areas, influenced by the Siberian continental high-pressure system during wintertime (Lee and Park, 1996; Jhun and Lee, 2004; Choi and Ghim, 2021). In contrast, the high NO2 TropVCD at relatively high temperatures in winter might be associated with the combination of transboundary transport and stagnant conditions induced by the eastward movement of migratory high-pressure systems from China mainland (Ghim et al., 2019). The highest NO2 TropVCD was observed in Yokosuka (20.4 × 1015 molecules cm−2), followed by those in Busan (13.1 × 1015 molecules cm−2) and Gwangju (10.4 × 1015 molecules cm−2), which is the same order of population and nearby strong emission sources. Wind speed was also an important variable because low wind speeds can lead to the accumulation of NO2 emitted from human activities in urban areas. Next, the ‘‘trend’’ variable reflects a clear decreasing trend according to the air quality policies in Japan and Korea; thus, a high NO2 TropVCD was observed in the early years of the study period (Choi et al., 2021). Weekdays (Monday–Friday) exhibited a higher NO2 TropVCD than weekends because human activities were more concentrated (Beirle et al., 2003). Elevated NO2 TropVCD levels were observed when the RH was high and during the early and late weeks of the year (winter) due to decreased photo-dissociation corresponding to weakened solar irradiance under cloudy (high RH) and low-temperature conditions. Therefore, the partial dependence of each variable can fully explain the NO2 TropVCD variations at the three urban sites. Fig. 6 shows the linear trends estimated from the Theil-Sen slope (Theil, 1950; Sen, 1968), which has been used to analyse long-term temporal variations in air quality data (Collaud Coen et al., 2013; Choi and Ghim, 2021; Choi et al., 2021). The Theil-Sen slope was obtained from the monthly mean using the R function ‘‘TheilSen’’ included in the package ‘‘openair’’ (Carslaw and Ropkins, 2012). Theil-Sen analysis was used to investigate the multiyear trend of the NO2 TropVCD based on the deseasonalized and deweathered output derived from the GBM in the previous section. From 2015 to 2020, the Theil-Sen slope of the NO2 TropVCD from 2015 to 2020 with a 95% confidence level continuously decreased by −5.40% yr−1 (−5.89 to −4.77% yr−1) in Gwangju, −4.96% yr−1 (−5.67 to −4.3% yr−1) in Busan, and −3.14% yr−1 (−3.56 to −2.62% yr−1) in Yokosuka, and all decreasing trends were significant (p < 0.01). Although the considered time scales were different, the magnitudes of the decreases agreed well with results in previous studies that used ground-based instruments and/or satellites except for Gwangju (Duncan et al., 2016; Herman et al., 2018; Georgoulias et al., 2019; Jamali et al., 2020; Choi et al., 2021). For the time series of the NO2 TropVCD from 2019 to 2020, the Theil-Sen slope indicated a much steeper and significant decreasing trend (by a factor of two) than that from 2015 to 2020: −12.5% yr−1 (−15.4 to −9.62% yr−1) in Gwangju, −11.9% yr−1 (−13.8 to −9.84% yr−1) in Busan, and −7.51% yr−1 (−8.56 to −5.55% yr−1) in Yokosuka (p < 0.01). This result is consistent with that in the previous section and reflects the dramatic decrease in the NO2 TropVCD in 2020. To identify the significant changes, we applied break-point analysis to the deseasonalized and deweathered NO2 TropVCD levels using the ‘‘strucchange’’ package for break-point detection in the R program (Zeileis et al., 2003), which can find a better explanation using two discrete models rather than one general model. This approach does not assume event dates but instead uses changes in linear regression properties in a data series over time to identify likely points of change by providing a more independent measure of events than a classical ‘‘before and after’’ analysis (Ropkins and Tate, 2021). Through this analysis approach, break points associated with statistically significant changes were identified in January 2020 at all three urban sites; this time frame coincides with that of the first observed cases of COVID-19 in both Korea and Japan. The 95% confidence intervals were also located in a similar range from December 2019 to February or March 2020. This result confirmed that the reduction in NO2 was mainly due to social distancing for COVID-19 rather than annual meteorological differences. Although January 2020 overlapped with the first phase of the Korean seasonal PM control policy (December to March of the following year), the sudden reduction in NO2, including Yokosuka, Japan, could be mainly explained by the outbreak of COVID-19. It should be noted that the decreasing trend in summer in 2020 was overwhelmed that in summer in 2019, but the NO2 TropVCD recovered suddenly to the level in the previous year after November; this feature was also observed in the NO2 TropVCD from OMI data. In this study, we investigated the impact of the unprecedented COVID-19 pandemic on the NO2 TropVCD at three urban sites (Yokosuka in Japan and Gwangju and Busan in Korea) and one remote site (Cape Hedo in Japan). Compared to the monthly mean from 2015−2018 and in 2019, that in 2020 was low due to social distancing in Korea and Japan and the lockdown in China to prevent the spread of COVID-19. At the three urban sites, the relative changes between 2015−2018 and 2020 were much larger than those between 2019 and 2020 due to the effectiveness of regulation policies, such as China’s Clean Air Action. High negative relative changes were observed from May to September (−30% to −18%). The differences in temperature and wind speed between 2020 and 2015−2019 indicated that the warm season might be relatively unfavourable condition for NO2 photo-dissociation, although an insignificant difference in the two variables blurred the effect of reduced human activity. Additionally, at Cape Hedo, no significant difference was observed between the levels in previous years and 2020, suggesting that only anthropogenic emission sources decreased dramatically. Regarding the daily variations, the 15-day moving average of the NO2 TropVCD during the Chinese New Year holiday at the three urban sites decreased rapidly compared to that in the previous week because the NO2 reduction during the New Year holiday period in China is a recurring annual phenomenon associated with decreases in business and industrial activities. Japan does not observe the Chinese New Year holiday; therefore, these NO2 reductions seemed to have been driven solely by NOx emissions at the local scale. The transportation and industry mobility levels from the Google database in both countries decreased as public awareness about COVID-19 increased, resulting in a decrease in the NO2 TropVCD. In the case of Yokosuka, the 15-day moving average of the NO2 TropVCD showed notable relationships with those of transportation (0.48) and industry (0.54) mobility data. In contrast, the NO2 TropVCD at the Korean sites showed a moderate and/or low correlation with the industrial sector (0.39 in Busan and 0.19 in Gwangju), and no significant correlation was observed for the transportation sector. These different tendencies might be associated with different social distancing policies in Korea (voluntary) and Japan (mandatory). We applied a GBM to extract meteorological and seasonal contributions from the NO2 TropVCD time series. The variable with the greatest influence on the NO2 TropVCD was temperature, followed by site, wind speed, trend, day of week, RH, and week of year. Using the deseasonalized/deweathered NO2 TropVCD from the GBM, the linear trends were estimated from the Theil-Sen slope. From 2015 to 2020, the NO2 TropVCD at the three urban sites decreased continuously by −3.14 to −5.40% yr−1. During the period from 2019 to 2020, the decreasing trend was much steeper than that from 2015 to 2020 (−7.51 to −12.5% yr−1), confirming the dramatic decrease in NOx emissions in 2020. At the three urban sites, statistically significant changes in the NO2 TropVCD were detected in January 2020, coinciding with the first observed cases of COVID-19 in both Korea and Japan. In particular, the decreased magnitude in summer in 2020 was considerably greater than that in summer in 2019. This result confirmed that the reduction in NO2 was mainly due to reduced NOx emissions resulting from social distancing for COVID-19 rather than yearly meteorological differences; however, during the cold season, the NO2 level increased notably to the previous level due to increases in human activities. In the future, model simulation should be required to understand the quantitative effects of the reduction in NO2 emissions. This research was supported by the GOSAT-GW NO2 NIES-JAMSTEC-NICT collaborative research project. The authors thank the City of Yokohama for maintaining the air quality monitoring stations used in this study. The MAX-DOAS data files for the MADRAS network observations from 2007 to 2021 are available at http://ebcrpa.jamstec.go.jp/maxdoashp. Pandora data at Busan were processed as part of the Pandonia Global Network (PGN; https://www.pandonia-global-network.org/, last accessed 6 November 2020). Meteorological data were obtained from the Japan Meteorological Agency and Korea Meteorological Administration through https://www.data.jma.go.jp/gmd/risk/obsdl/index.php (in Japanese) and https://data.kma.go.kr/data/grnd/selectAsosRltmList.do?pgmNo=36 (in Korean), respectively. The ground-level NO2 concentrations in Yokohama are available at https://www.city.yokohama.lg.jp/kurashi/machizukuri-kankyo/kankyohozen/kansoku/kanshi_center/geppo/geppoarc.html (in Japanese).1 INTRODUCTION
2 METHODS
2.1 Measurement Sites
Fig. 1. The difference in tropospheric NO2 vertical column density (NO2 TropVCD) over East Asia between 2019 and 2020, which measured by Ozone Monitoring Instrument (OMI) products (OMNO2d_003; solar zenith angle < 80° and cloud fraction < 30%) and site locations of four measurement sites (Yokosuka and Cape Hedo in Japan and Gwangju and Busan in Korea).
2.2 MAX-DOAS
2.3 Pandora
3 RESULTS AND DISCUSSION
3.1 Difference in Monthly Variation in the NO2 TropVCD between 2020 and Previous Years
Fig. 2. Monthly variation of NO2 TropVCD at (a) Yokosuka, (b) Busan, (c) Gwangju, and (d) Cape Hedo. The different color symbols with vertical lines indicated the mean and standard deviation of NO2 TropVCD in 2015–2018, 2019, and 2020, respectively.
3.2 Time Series of the NO2 VCD with Human Activity during the COVID-19 Period
Fig. 3. (a) Time-series of 14-days running average of NO2 TropVCD at four sites; Yokosuka, Busan, and Gwangju. The red, blue, and grey shades indicate the week of the Chinese New Year (CNY) holiday, 24–30 January in 2020, the week before (period 1) and after CNY (period 3), respectively. (b) Time-series of 14-days running average of transit and workspace from Google mobility data and daily confirmed new COVID-19 cases. The one red and three blue shade periods indicate emergency measures for social distancing in Japan and Korea, respectively.
3.3 Deseasonalized and Deweathered Analysis
Fig. 4. Comparison of measured with predicted NO2 TropVCD which calculated from the generalized boosted (GBM) model to minimize the seasonal cycle (deseasonalized) and meteorological effects (deweathered). Different colored symbols indicate data from three urban sites (Yokosuka, Gwangju, Busan). The solid and dashed lines mean best-fit line and 1:1 line.
Fig. 5. The partial dependencies of each variable for predicting NO2 TropVCD from gradient boosting from all data from three urban sites. Error shadings and vertical lines for (b) and (e) represent the standard deviation. The order of figure follows the order of influenced variables, (a) temperature (24.0%), (b) sites (20.5%), (c) wind speed (18.7%), (d) trend (11.3%), (e) day of week (10.2%), (f) RH (9.8%), and (g) week (5.4%).
3.4 Trend Analysis with Break-point Detection
Fig. 6. The long-term trend of NO2 TropVCD at three urban sites using deseasonalized and deweathered data using generalized boosted (GBM) model: (a) Yokosuka, (b) Busan, and (c) Gwangju. Symbols with line and solid line indicate monthly mean of NO2 TropVCD and Theil-Sen slope from 2015 to 2020. The enlarged figures on right panel show the same as left panel except for shortening time scale (2019 to 2020). Vertical blue solid and dashed line indicate detected break-point (statistically significant changes) and 95% confidence intervals.
4 CONCLUSIONS
ACKNOWLEDGEMENTS
REFERENCES
