Air quality and urban climate improvements in the world’s most populated region during the COVID-19 pandemic

In this study, we assessed air quality (AQ) and urban climate during the mobility restrictions implemented in the Greater Tokyo Area, Japan, the world’s most populated region, in response to the COVID-19 pandemic. Observations from dense surface networks were analyzed using an interpretable machine learning approach. In parallel with a ∼50% reduction in mobility and an altered lifestyle of the population, we found limited reductions in nitrogen dioxide; decreases in fine particulate matter not entirely driven by local mobility; minor variations in ozone, with a positive (negative) tendency in areas with high (low) emissions; a decrease in air temperature consistent with mobility; and pollution levels and air temperature changes with well-defined, common spatiotemporal patterns. Specifically, cooling mainly occurred in urbanized areas with an improved AQ. Overall, although reductions in mobility were moderately effective in improving the typical indicators of urban AQ, including those known to negatively impact human health, the reductions in waste heat had a stronger impact on Tokyo’s urban heat island, suggestive of a strategy to minimize exposure to heat stress. These findings can help guide urban planning strategies and policies aimed at addressing climate change.


Introduction
About half of the world's population lives in urbanized areas, a proportion expected to reach 66% by 2050 (UN 2015).However, the air quality (AQ) in urbanized areas is often unhealthy.Pollutants such as fine particulate matter (PM 2.5 ), ozone (O 3 ), and nitrogen oxides (NO x ) lead to respiratory and other diseases and are important causes of morbidity and mortality (Schraufnagel et al 2019, Fuller et al 2022, Feron et al 2023).Increasingly strict emission control strategies aimed at reducing air pollution have been implemented worldwide over the past several decades, following international guidelines.Assessing the resulting improvements in AQ and the potential for further future improvement is thus of primary importance.
A further concern in urbanized areas is the urban heat island effect.Commonly, temperatures are warmer in cities than in rural areas due to the prevalence of heat-absorbing materials, the lack of vegetation, and the waste heat associated with human activities (e.g. transportation and heating-cooling systems).As the intensity, frequency, and duration of heatwaves increases (e.g.Feron et al 2019, Perkins-Kirkpatrick and Lewis 2020, Trancoso et al 2020), reducing the urban heat island effect has become an essential aspect of urban adaptation strategies (Larsen et al 2015) aimed at improving the quality of life and reducing the adverse health outcomes caused by increased temperatures (Arbuthnott et al 2017).
While the COVID-19 pandemic was a worldwide tragedy, it also created a global natural experiment enabling an evaluation of the impact of reduced anthropogenic emissions in urbanized areas, and thus the possibility of a 'green' future (e.g.Shakil et al 2020, Zheng et al 2020, Gkatzelis et al 2021, Laughner et al 2021, Cooper et al 2022, Damiani et al 2022).The mobility restrictions imposed by most governments reduced urban traffic (Yabe et al 2020) and improved AQ worldwide (Venter et al 2020).However, in many studies of the COVID-19-AQ connection, the focus was on reactive pollutants, such as nitrogen dioxide (NO 2 ), as their short lifetimes and predominantly anthropogenic origin make them good proxies of anthropogenic emissions (Bauwens et al 2020, Petetin et al 2020, Venter et al 2020, Cooper et al 2022, Damiani et al 2022, Levelt et al 2022).Yet, in contrast to the reported reductions in the emissions of those indicators, the levels of other pollutants, such as O 3 , formaldehyde (HCHO), and carbon monoxide (CO), as well as aerosols (e.g.PM), often showed minor changes, due to their long lifetimes, secondary origin, or mixed sources (Le et al 2020, Sicard et al 2020, Ghahremanloo et al 2021, Huang et al 2021, Keller et al 2021, Liu et al 2021).
Exploring the temperature response to the COVID-19 lockdown can provide valuable indications to minimize exposure to heat stress by reducing urban mobility and adopting remote work.A recent study of the mobility restrictions in several Chinese cities (Liu et al 2022) showed that the COVID-19 lockdown was also associated with a transient decrease in air and surface temperatures, due to reductions in waste heat caused by transportation and heating-cooling systems.Nevertheless, in cities where aerosol concentrations are frequently very high, an increase in the shortwave radiation flux due to reduced aerosol scattering could produce a contemporaneous temperature rise (Shepherd 2022).Combined analyses of AQ and urban climate variations in areas usually unaffected by extreme pollution could shed further light on these observations.The Greater Tokyo Area is a vast flat area of central Japan that is home to ∼43 million inhabitants and includes the capital city of Japan (figure 1).To limit the spread of COVID-19, this region was under a state of emergency between 7 April and 25 May 2020, when authorities requested citizen self-restraint.A recent study (Damiani et al 2022) provided a first look at the changes in the levels of trace gases and lightabsorbing aerosols that occurred during this period, mainly focusing on satellite data.A modulation of Tokyo's urban heat island, potentially associated with the reduced mobility during the pandemic, has also been reported (Fujibe 2020).
The aim of the present study was to assess the changes in pollution and temperature that occurred as a result of the altered lifestyle of the Greater Tokyo Area population during the state of emergency imposed by the Japanese government during the early phase of the COVID-19 pandemic.While the variation in the satellite-based surface land temperature during the pandemic has been explored for a few regions (e.g.Chakraborty et al 2021, Kenawy et al 2021), the focus of our study was the air temperature, which is more relevant for public health considerations.We provide a novel view of the spatiotemporal relationship of temperature and pollution that may lead to adaptation strategies to climate change that are applicable to not only the Greater Tokyo Area but also other regions worldwide.

Methods
A business-as-usual (BAU) scenario at daily resolution derived from an interpretable machine learning (ML) approach was compared with recently updated and corrected observations from dense surface networks (i.e.Atmospheric Environmental Regional Observation System) and with the Agro-Meteorological Grid Square Data (resampled over a grid of 0.05 • × 0.05 • ).The observed target parameters were reproduced using an ML algorithm trained on the previous 6 years (figure 1; see also Petetin et al 2020).The data were divided into five training and test datasets, yielding five BAU models.The first reproduced the observations in 2020; the remaining four reproduced the observations in 2016-2019, to estimate the uncertainty of the 2020 BAU simulation, based on the standard deviations of their residuals (figure 1).Only meteorological variables were ingested into the ML model as predictors to reproduce the AQ targets, while a climate analogs method was used to estimate the air temperature BAU scenario (for further details on datasets and methods please see supporting information).Data over the period 2010-2020 were mainly used.

AQ improvements
Studies based on satellite and ground observations showed that NO 2 levels in Tokyo decreased during the state of emergency (Ito et al 2021, Damiani et al 2022) but the changes that occurred on a daily scale were not examined.Figures 2(a) and (b) shows the time series of daily observations of surface NO 2 (red line) and partial column NO 2 (green line) and the respective meteorology-driven BAU scenarios during the first half of 2020, when mobility was strongly reduced (see supporting information for details of the datasets and ML approach).Data were recorded around Chiba (figure 1), a city ∼40 km east of Tokyo and in an area not directly impacted by road traffic pollution.The average NO 2 level from 13 AQ sites, located within 10 km from of the remote-sensing-based partial column NO 2 measurement site and where  6).Urban areas (in black) are from the corresponding built-up class of the JAXA high-resolution land use and land cover map at 0.5 × 0.5 km.The simplified scheme of the employed ML approach shows that the BAU modeled scenario for 2020 was trained with data from 2014-2019.The remaining four BAU models were used to estimate the uncertainty of the 2020 BAU scenario based on their residuals (i.e.observation minus BAU model).Details are provided in supporting information.
the data better correlated with the partial column NO 2 data, was calculated.Our ML-based model (thin lines) well reproduced the day-to-day variations in the NO 2 partial column data.Before the period when the state of emergency went into effect, the model tended to well fit the observations, with both indicating negligible NO 2 reductions.NO 2 levels lower than those of the simulations appeared only in late April, around the time of Japan's major spring festivity (Golden Week, starting on 29 April 2020; day of the year = 120 in figures 2(a) and (b)).This change lasted until the end of May and occurred again in June.Although the surface data showed similar features, the corresponding BAU model was less successful in reproducing the changes, especially in February.The large error bars indicated that the model generally tended to fail during this period and at the measurement location, whereas during the emergency period the errors were smaller than the model-observation differences.Overall, NO 2 reductions of ∼25% lasted for ∼2 weeks in May and June, but the general NO 2 trend only partially reflected the trend in mobility (i.e.Google Mobility data, see the gray line in the opposite axis in figures 2(a) and (b) and the supporting information).
For comparison with the Chiba findings, figure 3 shows the spatial distributions of the changes (observation minus BAU scenario) during the state of emergency (7 April-25 May) for surface NO 2 (a), (d), O 3 (b), (e), and PM 2.5 (c), (f) at each AQ site within the Greater Tokyo Area.Surface NO 2 changes were negative almost everywhere except over most of the Boso Peninsula, where (excluding the area around Chiba city) the population density and urbanization are limited, such that emissions are fewer and more sporadic than in the west (figure 3(a)).The highest NO 2 levels (black circle) were usually measured in central Tokyo, consistent with the city's high emissions, resulting in more significant absolute changes (figure 3(d)).NO 2 levels decreased by 20%-30% for most sites.While the reductions in NO 2 emissions were usually more prominent at roadside than at ambient AQ sites, the changes in central Tokyo were homogeneously distributed.It should be noted that satellite observations (shown in the background in figure 3(a)) were roughly representative of daily changes.Unlike the spatial pattern obtained from the satellite data, there was no evidence of positive surface NO 2 changes in the region north of Tokyo.This limited discrepancy can be explained by the low temporal resolution of the satellite datasets, and by the difference between surface and free tropospheric NO 2 levels.However, unlike ground observations, satellite observations in 2020 were compared to 2019, as in Damiani et al (2022); therefore, meteorology also likely played some role.
The changes in O 3 (figures 3(b) and (e)) were usually limited, with most sites showing no difference.However, there was a tendency of positive changes in the main urbanized areas and negative changes in suburban and rural locations.The increase in O 3 levels in central Tokyo (black circle) spatially coincided with the highest (absolute) decrease in NO 2 levels (figure 3(d)).As a large urban area, Tokyo is under volatile organic compound (VOC)-limited conditions for most of the year (Akimoto 2017, Irie et al 2021).Therefore, the increased O 3 levels may have been related to the decreased O 3 titration by NO resulting from the reduced mobility, in agreement with a previous study (Sicard et al 2020).The limited negative variations in southern Yokohama were likely driven by local conditions, as these changes started before the pandemic and persisted throughout the year.
PM 2.5 (figures 3(c) and (f)) levels decreased at nearly all measurement sites, with 20%-30% variations in urbanized areas.This reduction is consistent with recent research (Damiani et al 2022).The pattern of the PM 2.5 changes is coherent with that of NO 2 and shows more limited changes in the South, where even slightly positive changes can be found.
Figure 4 shows the time series of surface NO 2 , O 3 , and PM 2.5 levels in central Tokyo (figure 1), their corresponding modeled BAU scenarios, and the relative differences averaged at 26 AQ sites.As seen in the top panels, the simulation well reproduced the NO 2 concentrations in January and February but it produced overestimations since late February, roughly when remote work and school closures were mandated by the local government.Although this period was well before the emergency declaration, a limited change in NO 2 levels on the order of 10% occurred when Tokyo's mobility was already reduced by 20%.Also, the expected uncertainty of the BAU model (red bars in figure 4(a)) was well within these differences.With the start of the state of emergency, there was a sudden drop in NO 2 levels of 15%-20%.Finally, as noted above, a further drop occurred during a national holiday (Golden Week), with reductions of ∼30% lasting for ∼2 weeks.During the following weeks, and even after the end of the emergency, NO 2 levels remained below those of the modeled scenario.
Figures 4(c) and (d) shows that the modeled BAU scenario perfectly matched the O 3 levels until the declaration of the state of emergency, after which it underestimated them.Although the differences were limited, on the order of 10% at most, the results were consistent with the reduced NO titration, which could have led to higher O 3 levels.Nevertheless, care should be taken in evaluating such changes, as the  uncertainty of the scenario was significant and, in some periods, larger than the potential differences (e.g.around the end of the state of emergency).
The PM 2.5 trend is shown in the bottom panels of figure 4. PM 2.5 levels declined throughout most of the investigated period.Reductions began in early February, as in the case of NO 2 , but thereafter the two indicators diverged, especially in early May, when NO 2 decreased while PM 2.5 remained at the levels predicted by the model.Several earlier studies also did not find a clear association between the reduced mobility associated with COVID-19 restrictions and PM 2.5 concentrations (e.g.Venter et al 2020, Huang et al 2021).This was likely because the anthropogenic contribution was smaller than natural contributions of dust, salt, and other sources, and because the longdistance atmospheric transport of PM 2.5 can be significant.Moreover, local restrictions likely did not affect potential anthropogenic sources such as agriculture and energy production.Finally, it is worth noting that reductions in NO x often led to increases in O 3 , which in turn increased atmospheric oxidizing capacity and enhanced formation of secondary PM (Huang et al 2021).
In January, and thus well before local mobility restrictions, changes in PM 2.5 were associated with a slight uncertainty.Japan is affected by transboundary pollution transport from the Asian continent (Itahashi et al 2022).While the impact is more relevant on the west side of the country, a contribution can also be expected in the investigated region (Damiani et al 2021).Since China was under a strict lockdown in January-February and emissions were strongly reduced at the national level (Bauwens et  The altered lifestyle of the population in the study area and its effect on AQ are evident in figure 5. Figures 5(a) and (b) shows the weekend effects of O 3 (defined here as the relative difference between the Sunday and weekday means) from March to June averaged over 2010-2019 and 2020.Usually, O 3 levels in the Greater Tokyo Area are higher during the weekends, when NO x levels are reduced (Sadanaga et al 2012).In Tokyo's center, the increase in O 3 levels was >30% whereas in other areas it was 10%-20% (figure 5(a)).However, in 2020, there were no relevant changes in O 3 at most stations and only a limited increase (5%-10%) in the city center (figure 5(b)).Moreover, the values at some sites in the semirural areas on the region's west side were negative (figure 5(b)).The annual O 3 weekend effects from 2010 to 2020 were ranked and the year with the corresponding smallest (or even negative) change at each site was plotted (figure 5(c)).In 2020, the weekend effect in urbanized regions was unprecedented.

Urban climate improvements
Next, we examined the AQ changes into the context of urban climate variations.Figure 6 shows the daily temperature departures from the modeled BAU scenario in the Greater Tokyo Area from 20 January to 28 June 2020; the mean, minimum, and maximum air temperature are reported together with the concomitant changes in mobility and NO 2 levels, as a proxy of pollution (opposite axis).Overall, the temporal evolution of the temperature changes was consistent with a mobility reduction.After some modulation in late January, the mean and minimum air temperatures decreased beginning in early March,  after the end of the state of emergency), suggesting poorer agreement between the model and observations during the monsoon rainy season (daily weather conditions were evaluated computing the clear-sky index; Damiani et al 2018).The bottom panel of figure 6 shows the linear correlation between mobility and temperature.Mobility correlated most strongly with the minimum temperature (r = 0.92), followed by the mean temperature (r = 0.89) and maximum temperature (r = 0.76).These results are consistent with the more stable early-morning conditions, when the temperatures is usually lowest, rather than earlyafternoon conditions, when the temperature is generally highest.A similar correlation between mobility and NO 2 (r = 0.87) was also noted, while the correlation between the changes in mean temperature and NO 2 was slightly lower (r = 0.68, not shown).
Figure 7 presents the spatial pattern of the change in the mean air temperature during the emergency period.To focus on regions with more reliable changes, only areas where temperature variations were larger than 3 sigma are shown (see supporting information for details).Overall, temperature decreases were evident over most urbanized regions and communication routes.The temperature decline was less than −0.6 • C in central Tokyo, with smaller decreases east of Tokyo, around Chiba, and south of Yokohama.To put these values into context, the spatial distribution of the weekend effect on air temperature was determined (figure S2).The generalized cooling and the gradient from central Tokyo to the suburbs were roughly comparable with the pattern shown in figure 7, although the absolute changes were smaller (maximum of about −0.2 • C).Therefore, the impact of COVID-19 on temperature was more significant than the usual temperature reduction during the weekends.Figure 6 also shows the changes in surface NO 2 , as a pollution proxy, at the AQ measurement sites (filled circles as in figure 3(d)).The spatial correlation between the two datasets (r = 0.55) further implied that reduced mobility resulted in similar patterns of temperature and NO 2 ; note that the changes in PM 2.5 followed a similar pattern but NO 2 is shown due to the larger number of sites with available data.The improved AQ was coupled with an improved climate over most urbanized areas during the state of emergency, whereas in areas with limited or even positive NO 2 variations (e.g.most of the Boso Peninsula) there were no associated temperature decreases, with a few exceptions (the changes over the entire area without the associated sigma restriction are shown in figure S3).

Discussion and conclusions
Pollution and air temperature affect the quality of life for residents of urbanized areas.High levels of air pollution are associated with increased morbidity and mortality while urban heat island effects worsen heat stress.COVID-19 provided the conditions for a global natural experiment that enabled evaluations of the impact of achieving reduced anthropogenic emissions in urbanized areas via reduced mobility, and the interactions between altered pollutants and urban climate.
The aim of our study was to examine the spatiotemporal variabilities in AQ and urban climate and their joint impact in the world's most populated urbanized region.The analysis took advantage of the hundreds of AQ and weather monitoring stations in the study area.The data were analyzed using an interpretable ML approach that also enabled an assessment of the behavior of the physical parameters embedded in the model, thereby showing the utility of interpretable ML approaches in AQ studies (see supporting information).
The results showed that the COVID-19-induced changes in AQ were associated with spatial and temporal temperature variations.Both pollution levels and air temperature improved following welldefined, common, temporal and spatial patterns.Specifically, cooling mainly occurred in urbanized areas with an improved AQ.These conclusions were derived from a common ML approach applied to two independent datasets and using different predictors.The identified association between temperature and AQ demonstrates the suitability of the adopted method and points to reduced mobility as the shared causative factor.The limited discrepancies in the temporal evolution of temperature and AQ were related to the fact that while the largest decreases in the mean and minimum temperature occurred during the first half of the state of emergency, in agreement with the temporal evolution of mobility (figures 6(a) and (b)), pollution levels followed a slightly different course.Figures 4(a) and (b) shows the sharp reductions in the NO 2 levels in central Tokyo during the second half of the state of emergency.Since mobility was consistently reduced by ∼50% over the entire emergency period, the reductions in road transport emissions were likely coupled to more significant reductions in factory emissions during the second half (which partially coincided with a holiday period).Indeed, a recent report showed decreasing trends in NO x emissions from Japanese road transport while industrial NO x emissions remained unchanged during recent years (Kurokawa and Ohara 2020).As a result, the relative contribution of industrial NO x emissions over the total NO x emissions has increased.
The trend in mobility changes in Tokyo according to Google Mobility mostly coincided with that inferred from other local data sources (Yabe et al 2020, Nagata et al 2021).However, while traffic counts decreased by ∼50% in central Tokyo, the decrease in other areas was smaller (∼20%, Takane et al 2022).Nevertheless, the advantage of using Google Mobility is that it allows a more straightforward comparison of local changes with those in other cities or countries.For example, Venter et al (2020) compared NO 2 , O 3 , and PM 2.5 variations with Google Mobility data in different countries during the pandemic.They found a linear correlation between NO 2 and mobility, with a NO 2 reduction of ∼50%, corresponding to a mobility reduction comparable to that estimated for Tokyo.Although this correlation is subjected to significant variations, the difference compared with the NO 2 changes estimated in this study is worth noting.
Most of the investigated region showed minor variations in O 3 levels, with a tendency toward positive (negative) changes in areas with higher (lower) emissions.The positive tendency in O 3 levels in Tokyo was roughly similar to the expected general behavior of O 3 in other countries with the same levels of mobility (Venter et al 2020).
As further evidence of the impact of the altered lifestyle of the population in urbanized regions, the O 3 weekend effect in 2020 was unprecedented (i.e.negligible or even reversed) in the latter, suggesting that O 3 provisionally approached the transition level during this period.This was further confirmed by the HCHO to NO 2 ratio, an additional indicator of O 3 sensitivity determined from satellite data.
The reductions in PM 2.5 in the Greater Tokyo Area were more significant than the changes resulting from the same mobility reduction in other countries (Venter et al 2020).A positive mean bias error (MBE) between the observations and the BAU scenario was found for each of the previous 4 years whereas in 2020 the MBE was negative (see supporting information).Therefore, the PM 2.5 trend in 2020 was unusual, although due to its poor agreement with local mobility it could not be entirely explained by mobility restrictions.Rather, other factors not accounted for in the model, such as transboundary transport (Ikeda et al 2015), may have played a role.
The reduction in PM 2.5 during 2020 raises the question whether its impact on air temperature was due to reduced aerosol scattering and the consequent increase in ground-level radiation.However, the temperature decrease in Tokyo was well within the variations reported for most Chinese cities (Liu et al 2022).In additional experiments (not shown), including PM 2.5 as a proxy of aerosol scattering in the ML model did not substantially change the results or contribute to explaining the temperature change.Therefore, aerosol scattering was unlikely to have played a relevant role in the temperature variation identified in this study.
Studies assessing the AQ and urban climate benefits caused by the pandemic help put our analysis into perspective.Venter et al (2021) found a total of 49 900 excess deaths potentially avoided during lockdowns in 34 countries.Only in China, 8911 (3214) NO 2 (PM 2.5 )-related deaths were avoided (Chen et al 2020), while in a country with population and pollutant reduction comparable to Japan (i.e.Germany), 667 excess deaths were avoided (Venter et al 2021).These findings suggest the potential health benefits of reducing BAU emissions.
During the last decades, Japan's deaths in July and August increased by 1.1% for each one • C increment of summer mean temperature (Fujibe and Matsumoto 2021).A recent study suggested a more substantial influence of a mobility reduction on urban air temperature in summer than in spring due to the more diffuse use of cooling systems (Takane et al 2022).Thus, the considerable cooling homogeneously spread over a large region we observed already in spring suggests a way to minimize exposure to heat stress.Nevertheless, an association between maximum temperature and mobility remained somewhat elusive in our study.
Our analysis showed that reducing urban mobility reduced NO 2 and (at least in part) PM 2.5 , enhanced O 3 , and decreased air temperature.It indicates that AQ mitigation in the investigated region should be achieved through a coordinated and balanced policy for controlling several pollutants.
In addition to introducing trees, green roofs, and vegetation, which are known to reduce urban heat island effects, a few studies evaluated the impact of the urban plan.In a scenario in which offices in Tokyo are distributed in the suburbs instead of in the center, simulations rendered a moderate temperature decrease in the city's core (Adachi et al 2012, Takane et al 2022).Although similar conclusions can be drawn from analyzing the weekday-weekend temperature difference (figure S2), evaluating such findings in the real world is hard.Our study confirms these previous results and points to reducing urban mobility and adopting remote work as a further way to minimize exposure to both pollution and heat stress.Therefore, urban planning strategies aimed at adapting to climate change should consider our findings.

Figure 1 .
Figure 1.Inset: map of the investigated region comprising the Greater Tokyo Area (defined as the region between 35.5-36.0N and 139.5-140E). White circles show the zones examined in detail (i.e.Tokyo and Chiba city).Red arrows indicate areas used as a predictor of the Greater Tokyo Area's temperature (red dashed box; see also figure6).Urban areas (in black) are from the corresponding built-up class of the JAXA high-resolution land use and land cover map at 0.5 × 0.5 km.The simplified scheme of the employed ML approach shows that the BAU modeled scenario for 2020 was trained with data from 2014-2019.The remaining four BAU models were used to estimate the uncertainty of the 2020 BAU scenario based on their residuals (i.e.observation minus BAU model).Details are provided in supporting information.

Figure 2 .
Figure 2. The panels show the time series of remote-sensing-based partial column NO2 (green) and in situ surface NO2 (red), recorded by MAX-DOAS and AEROS, respectively, from 20 January to June 28, 2020, as measured at Chiba.The average from the 13 AQ sites located 10 km from Chiba city that best correlated with remote-sensing observations were used to compute the relative difference (a) and absolute time series (b).The thick line indicates the observations, and the thin line the modeled scenario.Google mobility data are shown as a gray line (opposite axis) in (a).A 15-day rolling average was applied to the time series.The red bars show the 1-sigma error (see supporting information).

Figure 3 .
Figure 3. Average changes (observation minus BAU scenario) in the in situ surface NO2 (a), (d), O3 (b), (e), and PM2.5 (c), (f) levels at each AQ site of the AEROS network during the state of emergency (from April 7 to May 25).(a) The changes in surface NO2 plotted against the spatial distribution of the relative changes in the TROPOMI NO2 column for the same period (but estimated from the difference between 2020 and 2019, as in Damiani et al 2022).In the other panels, the values are plotted over a terrain map.Top panels: relative differences; bottom panels: absolute differences.Black circles show Tokyo and Chiba city.

Figure 4 .
Figure 4. Surface NO2 (a), (b), O3 (c), (d), and PM2.5 (e), (f) levels averaged over AQ sites of the AEROS network located 8 km from central Tokyo from January 20 to June 28, 2020.Thick line: observation; thin line: BAU modeled scenario.(a), (c), (e): absolute values; (b), (d), (f): relative values.Mobility data for Tokyo are shown as a gray line on the opposite axis.Red bars show the 1-sigma error of the BAU model.Data smoothed by a 15-day rolling average.

Figure 5 .
Figure 5. Mean O3 weekend effect (Sunday minus weekday mean) for 2010-2019 (a) and 2020 (b) between March and June based on AEROS observations.The year corresponding to the smallest annual weekend effect for each site is shown in (c).
al 2020, Le et al 2020), the reduced transport of pollutants toward Japan (Itahashi et al 2022) probably contributed to the early decline in PM 2.5 levels in the Greater Tokyo Area.A recent study (Schroeder et al 2022) investigated the changes in O 3 levels in Los Angeles in relation to the pandemic.Based on a trend analysis using two indicators of O 3 sensitivity, i.e. the O 3 weekend effect, and the satellite-based HCHO to NO 2 ratio, the authors showed that 2020 was the first year on record in which the spring was NO xlimited.Following Schroeder et al (2022), we analyzed the HCHO-to-NO 2 ratio using satellite data averaged over the most populated area (figure S1).While the NO 2 trend decreased from 2005 to 2022, the HCHO trend slightly increased (figure S1(a)).Then, in 2020, the HCHO-to-NO 2 ratio (figure S1(b)) suddenly raised and shifted from VOC-limited conditions toward a ratio value similar to that of the transition region (i.e. 1 < HCHO/NO 2 < 2) of O 3 formation (Hoque et al 2022).

Figure 6 .
Figure 6.Mean absolute changes (i.e.observation minus the BAU scenario) in air temperature averaged over the Greater Tokyo Area (dashed box in figure 1) during the first half of 2020.The changes in the mean (a), minimum (b), and maximum (c) air temperature.Error bars represent 1-sigma (see supporting information).The opposite axis shows the changes in mobility (based on Google Mobility) in Tokyo and the surface NO2 level (as in figure 4(b)).The scatter plot (d) shows the correlations for the same data as in the upper panel.The data were smoothed with a 29-day rolling average to account for the more significant fluctuation of the daily temperature differences.

Figure 7 .
Figure 7. Spatial pattern of mean air temperature changes (observation minus BAU scenario) during the emergency period.Only areas with temperature changes larger than 3 sigma are shown (see figure 1 and supporting information for details).The circles show the absolute changes in surface NO2 at each AQ site for the equivalent period (same colors for negative and positive changes as in figure 3(d) but ranging between −5 and +5 ppb).The background urban areas (gray) are those from the corresponding built-up class of the JAXA high-resolution land use and land cover map.The AMGSD temperature dataset was resampled over a grid of 0.05 • × 0.05 • .