Photochemical sensitivity to emissions and local meteorology in Bogotá, Santiago, and São Paulo: An analysis of the initial COVID-19 lockdowns

Nearly 85% of South America’s population lives in urban areas (United Nations, 2019), which, combined with extreme social, economic, and environmental segregation (e.g., Romero et al., 2012; Carpenter and Quispe-Agnoli, 2015; Guzmán and Bocarejo, 2017), makes these cities highly vulnerable to air pollution and climate change (Garschagen and Romero-Lankao, 2015). Bogotá, Santiago, and São Paulo together have ca. 40 million inhabitants and increasing motorization rates (Table 1). Mobile sources have contributed to air pollution and its evolution in these cities in recent decades (e.g., Nogueira et al., 2015; de Andrade et al., 2017; Barraza et al., 2017; Gallardo et al., 2018; Seguel et al., 2020; East et al., 2021). Emissions from mobile sources depend on the intertwined effects of changes in mobility, land use, population and economic growth, vehicle technology, decision making, and urban governance (e.g., Kelly and Zhu, 2016; Javaid et al., 2020). Despite technological advances that have led to notable reductions in pollutant and greenhouse gas emissions per vehicle and the implementation of vehicle restriction plans, attainment plans, and boosting public transport, motor vehicles—in and around our cities—are still a strong driver of pollutants and climate forcers. This, in turn, affects air composition and the quality of life of citizens.

It is well known that while primary pollution responds largely linearly to changes in emissions, secondary pollution is a nonlinear process (e.g., Parrish et al., 1999). Ozone (O₃) and secondary aerosols are initiated from the oxidation of volatile organic compounds (VOCs) by hydroxyl radicals (OH; e.g., Andreae and Crutzen, 1997). This mechanism involves the net production of nitrogen dioxide (NO₂), whose photolysis produces O₃. Additionally, as oxidation proceeds, the typically high vapor pressures of primary VOCs decrease, favoring partitioning into the aerosol phase (e.g., Goldstein and Galbally, 2007). The VOC to NO_x ratio (VOC/NO_x) is fundamental to the understanding of the nonlinear photochemistry of O₃ because it reflects the competition of VOCs and NO_x for OH (Fujita et al., 2003). In urban atmospheres with typically high levels of NO_x (VOC-limited), the formation of nitric acid represents the dominant radical sink, in which NO₂ removes OH, while NO titrates O₃. Under this regime, reductions in NO_x must be accompanied by comparable or greater reductions in VOCs to prevent the increase in O₃ formed via VOC oxidation. The shifts in O₃ formation regimes from relatively low to high VOC/NO_x have been observed in Santiago and (central) São Paulo (Silva Júnior et al., 2009; Seguel et al., 2012). In these cities, higher levels of O₃ are frequently measured during the weekends when NO_x levels are lower compared with weekdays. Conversely, increases in NO_x may result in decreases in O₃, which were noted in urban areas of São Paulo between 2009 and 2011, when higher ethanol prices induced 2 million drivers in São Paulo to switch to using gasoline (Vara-Vela et al., 2016; Schuch et al., 2019). The resulting increase in NO_x emissions resulted in up to 30% decreases in O₃ concentrations (Salvo and Geiger, 2014); however, the emission of ultrafine particles also rose by the same magnitude (Salvo et al., 2017).

The global COVID-19 outbreak in 2020 led to significant declines in vehicle emissions as lockdowns were imposed, limiting the use of fossil fuel–powered vehicles (e.g., Kroll et al., 2020; Le et al., 2020). Hence, mobility restrictions modified the VOC/NO_x ratio, leading to an increase in O₃, as reported for several cities worldwide (Connerton et al., 2020; Nakada and Urban, 2020; Siciliano et al., 2020; Gkatzelis et al., 2021; Rodriguez-Villamizar et al., 2021; Toro et al., 2021). In several areas, meteorological conditions could nevertheless result in amplification or damping of photochemical effects leading to secondary pollutants (e.g., Le et al., 2020; Ordóñez et al., 2020; Gaubert et al., 2021; Shi et al., 2021).

During the start of the coronavirus pandemic in South America in March–May 2020, adherence to mobility restrictions varied with time as well as between and within cities. There was generally less adherence in poorer communes, where household income depends on daily activities typically carried out far from home either in informal commerce or in domestic service (Comisión Económica para América Latina y el Caribe/Organización Internacional del Trabajo, 2020). Given that large cities are key to national economies (Table 1), governments and economic sectors were reluctant to adopt a complete lockdown, maintaining, in addition to essential activities (food, health, energy, and water provisioning), that the public sector services, banks, and industrial sectors remain open. Notably, in the case of São Paulo, mayor and governor decisions were largely counteracted by the federal government (The Lancet, 2020), leading to less adherence.

In this study, we investigated the changes in emissions from mobile sources that were modified by COVID-19 lockdowns in 3 South American cities: Bogotá (Colombia), Santiago (Chile), and São Paulo (Brazil). We focused on the impacts of the changes in primary and secondary air pollutants and the resulting local photochemistry. To accomplish this goal, we analyzed data from high-resolution vehicular emission inventories and in situ and remote observations to characterize the changes in mobile emissions and air quality compared with previous years (2014–2019) for all 3 cities. Additionally, we used a generalized additive model (GAM) technique to disentangle the roles of local meteorology, day of the week, and seasonality from other drivers of O₃ and NO₂ changes in these megacities. Thus, by analyzing commonalities and differences among Bogotá, Santiago, and São Paulo, we aim to provide a more systemic perspective to address photochemical air pollution in complex and diverse urban environments and its sensitivity to emissions and local meteorology. This is particularly important in terms of developing efficient air quality and climate policies for our cities in the near future.

Nonmethane VOC (hereafter referred to as VOCs), NO, NO₂, PM_2.5, and CO emissions were estimated through state-of-the-science vehicular emission models for Bogotá, Santiago, and São Paulo. Emission factors were generally adapted from the COmputer Programme to calculate Emissions from Road Transport of the European Environment Agency, considering actual fleet technology, fuel quality, environmental conditions, and vehicle activity in Bogotá, Santiago, and São Paulo.

In Bogotá, traffic activity data were derived from a static macroscopic transport model, reflecting Bogotá’s morning rush hour traffic conditions in 2015 (Mangones et al., 2019). Traffic changes during lockdown were estimated based on manual traffic counts in 10 arterial-road intersections in Bogotá from April 20 to May 1, 2020. Further details of traffic counts and data sources can be found in Table S1.

For Santiago, reference traffic activity was based on vehicular flow data for the 2011–2015 period, including origin-destination information for 45 communes in the metropolitan area of Santiago. The effect of lockdown strategies on vehicle traffic for the March–April 2020 period was obtained from the Tomtom traffic index (https://www.tomtom.com/en_gb/traffic-index/). Additionally, information from the Operational Transit Control Unit was considered (Table S1), comprising 49 traffic-count stations from 5 communes, for the periods between March 2 and 8 (before COVID-19 restrictions) and between March 30 and April 5, 2020 (during lockdown restrictions).

In the metropolitan area of São Paulo, vehicle activity (Table S1) and emissions were estimated using the Vehicular Emissions Inventories model (Ibarra-Espinosa et al., 2018; 2020). Hourly traffic counts were aggregated for several types of vehicles from toll stations in São Paulo for March 2020. The lockdown period was defined as March 23–28, and March 2–8 was denoted prelockdown.

We analyzed 4 monitoring stations in Bogotá, 8 in Santiago, and 5 in São Paulo to characterize the impact of emission changes during the variety of measures implemented to decrease the spread of COVID-19 in early 2020 (for simplicity, hereafter referred to as lockdown). The lockdown period in Bogotá was affected by the transport of biomass-burning pollution from Venezuela and the Colombian eastern plains during March (Mendez-Espinosa et al., 2019; Table 1). During the Santiago and São Paulo lockdown periods, the photochemistry was still strong enough to provide insights into the chemical mechanisms controlling secondary pollutant formation.

In Santiago, 3 of the 8 stations studied were located in an area under lockdown over the whole period. The rest of the Metropolitan Region was only under nighttime curfew (22:00−5:00). In São Paulo, the selected sites ranged from suburban-scale (outskirts of the city) to urban-scale stations. The locations of the stations are shown in Table 2 (see also Figure S3), and Table S2 provides more details of each monitoring station.

The air quality networks of Bogotá, Santiago, and São Paulo use instrumentation that complies with U.S. Environmental Protection Agency (EPA) reference methods. The time series used in this study was validated by each national Environmental Agency according to their own protocols. In some cases, preliminary data were used, which were validated after several checks that included completeness, instrument detection limits, revision of isolated extremely high hourly values, and other statistical analyses.

In each city, we defined the baseline period as a (2014–2019) multiyear average of pollutant concentrations. The lockdown period, when mobility restrictions were enforced, was defined as March 23–May 23 in Bogotá, March 16–May 14 in Santiago, and March 24–May 23 in São Paulo. The lockdown periods of each city were compared with the same periods of the baseline. This analysis was carried out with hourly data (air pollutants and meteorological parameters) whose periods had more than 70% validated data for every year at each station. The metrics used for statistical analysis were based on daily means and the maximum daily 8-h average (MDA8).

We used NO₂ tropospheric column observations obtained by the TROPOspheric Monitoring Instrument (TROPOMI) onboard the Copernicus Sentinel-5 Precursor (S-5P) platform for the lockdown periods listed in Table 1. Descriptions and validations of these data can be found in many studies (e.g., Judd et al., 2020; van Geffen et al., 2020; Verhoelst et al., 2021). The TROPOMI data were cloud filtered, and time averages were produced by mapping the observations to a regular oversampling grid of approximately 2 × 2 km. A suitable approximation for the tropospheric column errors was used to scale linearly with the observations (T Verhoelst, private communications, September 3, 2020). Therefore, relative percentage changes in NO₂, comparing 2020 with 2019, are reported in this article. It is worth noting that TROPOMI overpass local time is at 13:30, and thus, the columns are not based on a daily average. This is important to keep in mind when NO₂ columns are compared with in situ observations or emission estimates of NO₂.

To further explore the impact of lockdown on atmospheric composition in Santiago and São Paulo, data from the Aerosol Robotic Network (Holben et al., 1998) were examined. The aerosol optical depth (AOD) at 500 nm and the Ångström exponent (AE) during the March 20–April 3, 2020, period in central Santiago (33.458 S, 70.665 W) and March 24–April 15, 2020, at São Paulo University campus (23.561 S, 46.735 W) were compared with a multiyear average of the same dates for 2015–2019.

An empirical method (Chang and Lee, 2007) was used to estimate the fractions of primary and secondary aerosols in the 3 cities. According to this approach, carbon monoxide is used to trace primary combustion sources, while the daily maxima of O₃ (O_3max) serves as a proxy for the photochemical production of inorganic and organic aerosols. Essentially, the better the PM_2.5-to-CO correlation is, the more primary aerosols there will be, and the higher the ozone levels are, the more secondary particles there will be. Thus, the primary aerosol fraction is calculated for days when O_3max does not exceed the background O₃, and the daily secondary fraction is estimated as the difference between the observed PM_2.5 and the primary fraction. The background O₃ mixing ratio is in turn estimated from the intercept of the regression between O_x = O₃ + NO₂ and NO_x during daytime from an upwind station in each city. The methodology was recently applied in Santiago by Menares et al. (2020).

GAMs are nonlinear regression tools that allow for nonparametric fittings of complex dependences of response variables, such as O₃ and NO₂, on explanatory variables, for example, weather, seasonality (Julian day), day of the week, trends, and so on (Ordóñez et al., 2020; Solberg et al., 2021). Thus, a GAM adopts a sum of arbitrary functions of variables—possibly nonlinear—that represent different features via splines, which all together describe the magnitude and variability of the response variables (e.g., Molnar, 2022). The choice of explanatory variables brings in the physics of the problem at hand (e.g., Solberg et al., 2021). Here, we applied expert knowledge, examples from previous applications, and a ranking of variables according to the Akaike information criterion (Cavanaugh and Neath, 2019). Subsequently, we used partial dependence plots (Friedman, 2001) to interpret the model structure and outputs.

A GAM was applied to estimate MDA8 for O₃ and NO₂ with the following input variables recorded by the monitoring stations (Section 2.2): daily maximum temperature (°C), radiation (in W/m²), wind speed (in m/s), and specific humidity (in g/kg), as well as the day of the week and the day of the year (Julian day). While the former variables characterize changes in local meteorology, the temporal variables should reflect typical changes in urban activity patterns. Thus, the application of this model to data from the lockdown periods can detect changes due to factors other than meteorology and typical urban activity. For example, in this case, there were unprecedented changes in mobility and mobile emissions.

Between March and May 2020, mobility decreased by up to 80% in Bogotá and up to 50%–60% in Santiago and São Paulo, as illustrated in Figure 1. The figure illustrates how mobility changes evolved in the 3 cities and the differences between each city, except for when religious festivities coincided between countries. In the following subsections, the impact of mobility restrictions on vehicle emissions and concentrations of primary pollutants are addressed. Note that these estimates are based on local traffic counting and not the database illustrated in Figure 1.

Traffic in Bogotá during lockdown decreased for all vehicle types and in all zones. Motorcycles and light-duty vehicle circulation decreased by 53% and 73%, respectively, and urban buses decreased by 40% on average. Small and heavy truck circulation decreased by 58% and 28%, respectively. The only increase in circulation noted was for long-distance buses in the city center, which increased by approximately 39%. These changes in traffic are expected to have been observed only during the mandatory lockdown period in Bogotá. National and local authorities imposed personal mobility restrictions and made operational changes in urban and suburban public transport systems.

We estimated that PM_2.5 emissions decreased on average between 77% and 51% across the city during the morning rush hour. NO_x and VOC decreased by 71%–30% and 73%–54%, respectively. The emission reduction in NO_x between normal and lockdown conditions was 52.6 t/d, while the emission reduction in VOCs was 30.8 t/d. Light-duty vehicles contributed more strongly to emissions compared with reductions from buses and trucks due to the traffic behavior during lockdown. As shown in Figure 2, the city center had a lower PM_2.5 emission decrease rate, reaching a 52% decrease on average during the morning rush hour. The emission reduction rate of PM_2.5 also varied throughout the day, with higher emission reduction rates during morning (5:00–7:00) and evening (19:00–21:00) peak hours than during off-peak hours (0:00–4:00). We attribute this behavior to a greater effect of speed on emissions during high traffic congestion (rush hour). Alleviating traffic during heavily congested hours allows a greater number of vehicles to increase their speed compared with off-peak hours. Figure 2 shows PM_2.5 emissions during the baseline and lockdown periods at peak and off-peak hours in Bogotá. NO_x emission graphs are included as Supplemental Material (Figure S1).

The analysis of mobility information in Santiago shows a different behavior between areas with mandatory confinement and those in which it was voluntary. Communes with compulsory restrictions (7 of 52) showed a reduction in motorized trips during lockdown that fluctuated between 73% and 83%, while the rest of the communes varied between 42% and 54%. Public transport decreased by 82% on average over the entire city. Regarding day-of-the-week variation, stable behavior was observed between Monday and Saturday (63%), whereas Sundays showed a slightly greater reduction (72%).

Daily VOC and NO₂ emissions decreased 72% and 59%, respectively, over the entire metropolitan area of Santiago. These reductions correspond to 3.2 t/d and 7.3 t/d, respectively. The NO reduction between the baseline and lockdown conditions was 51%, corresponding to 30.4 t/d. In the communes with mandatory lockdown, VOCs and NO₂ were reduced by 92%, while in the zones with voluntary restrictions, VOCs decreased by 54% (Figure S2) and NO₂ by 44% (Figure 3). These results consider reduced mobility and a slight increase in average circulation speeds. The VOC/NO₂ ratio in the baseline period was 0.36, which changed to 0.26 in the lockdown period. It is interesting to note that lockdown led to a significant reduction in emissions (approximately 64%) of carbon dioxide (CO₂) from mobile sources, from 31.8 kt CO₂/d in the baseline period to 11.5 kt CO₂/d during lockdown. This highlights the importance of the transportation sector in CO₂ emissions in the urban areas of Chile.

Analyses of traffic counts showed that during lockdown, the number of passenger cars decreased by 54%, light commercial vehicles by 33%, light trucks by 22%, medium trucks by 16%, buses by 56%, and motorcycles by 48%, while heavy-duty trucks increased by 13%. VOC and NO₂ hourly emissions and the difference between the lockdown and baseline periods (prelockdown) are shown in Figure 4. The average hourly emissions of VOCs changed from 6.5 to 3.4 kg h^–1 km^–2, while NO₂ changed from 1.5 to 1.3 kg h^–1 km^–2, and the VOC/NO₂ ratio changed from 9.3 to 7.7. However, these patterns varied within the larger São Paulo city, and considering the extension of São Paulo, specific local differences need to be considered. The difference between the lockdown and baseline periods for NO₂ showed sharp decreases in downtown São Paulo (Figure 4).

The seasonal variations in rainfall, temperature, and relative humidity in 2020 and 2014–2019 for the 3 cities are presented in Figure 5. Note the sharp decrease in temperature from March to May (lockdown period) in Santiago as winter approaches, which contrasts with the constant temperatures of Bogotá, which has a more tropical climate.

Because the prevailing meteorology can greatly influence air quality (Gkatzelis et al., 2021), 2020 was compared with 2014–2019 to identify the environmental factors that could amplify or reduce the impact of lockdowns. Typically, decreased precipitation may worsen air quality, and in terms of photochemistry, higher temperatures (lower humidity and precipitation) lead to favorable conditions for O₃ formation. In the 3 cities, March–May 2020 was drier than the previous year averages (2014–2019). São Paulo was the only city to experience lower temperatures in 2020 (March–May). Temperature variations were less noticeable in Bogotá and slightly lower in Santiago. Similar findings can be found from ERA5 reanalysis-derived anomalies for March, April, and May 2020 with respect to 2015–2019 in Sokhi et al. (2021).

To provide a notion of the typical values found in the central stations of Bogotá (UTC-5), Santiago (UTC-4), and São Paulo (UTC-3), Figure 6 shows the pollution levels with diurnal cycles split into the baseline and lockdown periods. The lockdown led to significant changes in criteria air pollutants in the 3 cities (Table 2).

In Bogotá, the median of the daily mean concentrations of NO showed the greatest decline with respect to the baseline period (70%), followed by PM_2.5 (40%), CO (38%), and NO₂ (34%). In contrast, the median MDA8 O₃ mixing ratio increased (51%). The lowest increase in MDA8 O₃ (29%) and the lowest decrease in daily mean PM_2.5 (18%) occurred at the Guaymaral station, located in the northern part of the city. On the other hand, the highest increase in MDA8 O₃ (67%) and the greatest decrease in NO (91%) and NO₂ (52%) occurred near the geographic center at the CDAR station.

In Santiago, the criteria pollutants showed a significant change during the lockdown (Table 2). The median for the 8 stations showed that the NO daily mean decreased the most compared with the baseline period (67%), followed by NO₂ (25%), CO (20%), and PM_2.5 (14%). The median (8 stations) MDA8 O₃ mixing ratios increased by 36%. The comparison of monitoring stations located in areas under lockdown (Independencia, Parque O’Higgins, and Las Condes) with those areas where other kinds of measures were implemented showed no significant differences for NO (Table 2). Minor differences were found for PM_2.5 (–6%) and O₃ (+7%) between stations under lockdown versus the rest of the city, while CO and NO₂ showed the highest difference, both decreasing by 15% at the lockdown stations.

In São Paulo, the median of the daily mean concentrations showed the greatest reductions for the primary pollutants NO (67%) and CO (42%; Table 2). The median NO₂ and PM_2.5 concentrations, both of primary and secondary origin, decreased by 34% and 21%, respectively. In contrast, the median MDA8 O₃ concentration increased by 30%.

Overall, the ozone increases during the lockdown period negatively impacted air quality compliance. Santiago and central Bogotá recorded more days above the World Health Organization (WHO) guideline of 51 ppbv compared with the same periods of previous years (Table 3). This variation was much more intense in the eastern parts of Santiago, where the number of days over the guideline rose to 18 days. In São Paulo, some central stations retained similar exceedances or slightly increased, as seen at Park IBI SP, where the number of O₃ exceedances reached 16 days. It is worth mentioning that WHO guidelines allow 4 days above 51 ppbv per year (99th percentile of the MDA8 O₃).

Despite cloudiness, the reduction in NO₂ columns can be seen over and downwind of the 3 cities in the TROPOMI NO₂, showing parallels with the estimated reductions in mobile emissions (Figure 7). The median declines in NO₂ columns over Bogotá, Santiago, and São Paulo during lockdown were 40%, 35%, and 47%, respectively (Table S3). By relaxing cloud filtering over Bogotá and São Paulo, we estimated the decline in the NO₂ column 15-day periods, as shown in Table S3. In this manner, we found that while Bogotá and Santiago showed the largest decreases in the second half of March, 70% and 44%, respectively, São Paulo achieved the largest reduction (66%) in the first half of May. The decrease in Bogotá coincided with the largest reduction in mobility, as shown in Figure 1. In Santiago, the maximum NO₂ column reduction coincided with the quarantine over the 7 wealthiest districts and nighttime curfew elsewhere in the metropolitan area, which led to an overall mobility reduction of ca. 50%–60% according to Google data. In late May, mobility in Santiago reached its lowest values, coinciding with a national holiday and with the progressive extension of the lockdown over the whole metropolitan area; however, this was not reflected in the NO₂ column due to increased cloudiness that interfered with the satellite retrieval. Additionally, May was typically cold, and thus, residential emissions increased. In São Paulo, mobility was most reduced (50%–60%) in late March and April 1–15, which coincided with a 56% reduction in the NO₂ column. In the first half of May, when the decrease in the NO₂ column was strongest (66%), mobility increased, suggesting there was a decrease in sources other than vehicles.

The substantial reduction in VOC and NO_x emissions resulted in a photochemical regime shift. The ambient NO₂ to NO_x ratio was higher during the lockdown period than during the baseline period in Bogotá, Santiago, and São Paulo (Figure 8). In these 3 cities, less NO was available to titrate O₃ due to fewer vehicle emissions, and the conversion of NO to NO₂ was favored. As a result of these 2 chemical processes, O₃ increased at all stations studied in Bogotá, Santiago, and São Paulo (Figure 8). Furthermore, CO, which can be used as a proxy for VOCs, was useful for inferring that the VOC to NO_x ratio increased in the lockdown period. In Santiago, for example, higher MDA8 O₃ was reached on days when CO was similar to the baseline period mean (Figure 8).

In Santiago, during the lockdown period, the O₃ mixing ratio was also modulated by the temperature, reaching the highest MDA8 values on days with higher daily mean temperature and the lowest MDA8 on cloudy or rainy days (Figure 8). This correlation was particularly relevant for central Chile, where increasing heat waves were leading to more O₃ exceedances (Seguel et al., 2020) and wildfires (Jacques-Coper et al., 2021).

It is also worth noting that from January to April 2020, fire activity in Colombia and Venezuela was much more intense than the 2014–2019 average, peaking in February and March in Colombia and as late as April in Venezuela (Mendez-Espinosa et al., 2020). The impact of biomass-burning events continued during the lockdown period but was less severe than before.

We estimated the secondary fractions of aerosols at representative sites in Bogotá, Santiago, and São Paulo for the base and lockdown periods, as illustrated in Figure S3. The largest fraction of secondary aerosols was found in eastern Santiago, where secondary aerosols made up 55% and 61% of the daily average concentrations of PM_2.5 under the base (approximately 21 µg m^–3) and lockdown (approximately 16 µg m^–3) scenarios, respectively. Southwestern Bogotá had the highest daily averages of PM_2.5 and the lowest fractions of secondary aerosols during the baseline (35% of 34 µg m^–3) and lockdown periods (37% of 25 µg m^–3). At all sites, either no changes or slight increases in the secondary aerosol fraction was found, from zero percentage points (pp) at Pinheiros in São Paulo to 6 pp in eastern Santiago. In all sites studied, increased O₃ mixing ratios were observed over a broad range (6 pp at São Paulo and 69 pp at central Bogotá during lockdown), which suggests an intensification of photochemical activity consistent with the VOC-limited regimes identified earlier, as well as secondary aerosol formation. However, this growth was counteracted by the decrease in primary emissions that resulted in lower concentrations of daily average PM_2.5, NO, and NO₂. The only sites where we inferred an increase in the absolute concentrations of secondary aerosols were in western and central Santiago (approximately 8 pp); thus, at these sites, secondary aerosol production must have surpassed the effect of lower primary emissions.

Concerning changes in AOD, in Santiago, the AOD daylight cycle had a small shift in the maximum from 11:00 to 12:00 (local time). Additionally, a second maximum developed at 17:00 (Figure S4). Before the noon maximum, the lockdown average had a reduction in AOD with respect to the baseline period, while for the secondary maximum, a slightly larger value of AOD during 2020 was found than in previous years at the same hour. Regarding the AE, daylight cycles showed a similar shape to previous years, except for a notable reduction in the value of the AE for all spectral ranges considered (Figure S5 shows one such spectral range), with values lower than 1 standard deviation from the distribution of previous years.

In São Paulo, the daylight cycle of AOD and AE between 9:00 and 17:00 (local time) showed a decrease during the lockdown period compared with the baseline period (Figure S4). However, the maximum at 15:00 remained during the lockdown period and then remained relatively high. As in Santiago, changes in AOD remained within the variability of previous years, with a less substantial reduction in the AE than in Santiago, with values within the variability of previous years.

The AOD depends not only on the total aerosol burden but also on the size distribution. In general, small particles scatter light more efficiently than large particles; if 2 air masses contain the same aerosol burden, a larger AOD will result from the air mass with the larger concentration of smaller particles (e.g., Schuster et al., 2006, Huneeus et al., 2011). Therefore, changes in the combined AOD and AE can suggest relative changes in the size distribution. Based on this, the reduction in the AE could be produced either by a decrease in fine mode particles or an increase in the coarse mode fraction. The reduction in the AE with a decrease in AOD suggests a decrease in the fine mode particles, consistent with emission decreases caused by the significant decrease in mobile sources during the lockdown period. However, comparable AE and AOD values and even larger AOD values suggest an increase in the coarse mode fraction. The latter could be a result of the arrival of aged air parcels, intrusion of coastal air with marine aerosols, resuspension of dust particles, and/or a combination of these.

In the reference periods, the GAMs showed good error statistics for all cities (Table 4), with Pearson correlations for O₃ (NO₂) of approximately or above 80% (60%) and root mean square errors below 9 (11) ppbv for O₃ (NO₂). These error statistics were worse during the lockdown period, which was expected since there were fewer data and because the GAM estimates did not account for changes due to lockdown, that is, vehicular circulation or activity patterns.

Regarding the physical interpretation of the O₃ GAM, we found a linear positive dependence on daily maximum temperatures for Santiago and São Paulo, whereas in Bogotá, it was slightly nonlinear (Figure 9). These results are based on in situ measurements and consistent with other descriptions found in the literature (e.g., Porter and Heald, 2019). Positive linear dependences on solar radiation were found in Santiago and São Paulo, whereas over Bogotá, the relationship was linear. The former is consistent with intensified photochemical activity triggered by radiation. The weak dependence of ozone on solar radiation in Bogotá is somewhat surprising to us, and we hypothesize that this weak dependence results from the competition between increased solar radiation (closer to the equator) and the relatively high abundance of ultraviolet radiation due to its high altitude, leading to O₃ photolysis.

Only in São Paulo, there was an inverse relationship between O₃ and specific humidity (and relative humidity), as documented in other studies (e.g., Camalier et al., 2007). In Bogotá and Santiago, the relationship was slightly nonlinear but positive. The reasons for the different dependences in these 3 cities are unclear and require further investigation, including the role of deposition processes (e.g., Kavassalis and Murphy, 2017). The dependence on the maximum wind speed was quite different among the selected stations. In the case of Bogotá and Santiago, the stations considered were receptor sites. In Bogotá, the relationship was weak but positive (mostly during calm conditions). In Santiago, the relationship was nonlinear, with a positive dependence under nearly calm conditions (as in Bogotá) and an inverse relationship above 2 m/s, indicating the effect of dilution. In São Paulo, the relationship was clearly inversed, also suggesting dilution for winds below 3 m/s. With respect to the day of the week, Santiago had a clear weekend effect. In São Paulo, O₃ mixing ratios also steadily increased over the week from Monday to Sunday. Bogotá had no variations over the week for O₃. However, the weekend effect was clearer as NO₂ declined (Figure S7). Finally, regarding seasonality (Julian day) dependence, we found a progressive decline in O₃ as the cold season approached in Santiago. On the other hand, in Bogotá, the impact of savannah fires early in the year was clear and led to a progressive O₃ decline, with a small break in the trend in early May (JD ≍ 125). São Paulo had an increase in O₃ in early March (JD ≍ 100). Hence, the partial dependences of the O₃ GAM appeared to be physically justified. Something similar occurred in the case of the NO₂ GAM (see Supplementary Material, Figure S7).

In all 3 cities, the GAM outputs (Figure 10) indicate that observed increases in O₃—with respect to GAM-adjusted values—were often attributable to factors other than local meteorology and typical—as captured by the 2015–2019 variability—weekly and seasonal changes. During the lockdown, the GAM results indicate median increases in MDA8 O₃ of 9 ppbv (26%), 15 ppbv (35%), and 8 ppbv (19%) in Bogotá, Santiago, and São Paulo, respectively, when compared with the corresponding multiyear baseline period. However, over the approximately 1-month-long lockdown periods, there was also a significant portion of days when meteorological, weekly, and seasonal changes explain the observed variability.

In São Paulo, the O₃ GAM captured most of the variability found in the observations. This indicates that changes in mobility, precursor emissions, and ozone were relatively minor. Only a few days in the second half of April showed a distinct behavior, with increased ozone values. Overall, these features are consistent with less stringent mobility reductions and the modest decreases in NO₂ values, as shown in Figure S8, as well as in estimated NO_x emissions from mobile sources. This may be explained by the fact that lockdown was applied only to downtown areas of the vast metropolitan area of São Paulo. Interestingly, the Easter weekend resulted in a minor O₃ increase due to less mobility. Something similar occurred in Bogotá and Santiago, except that in Santiago, a sharp O₃ decrease was seen in connection with cloudy conditions and light precipitation in Santiago (not shown).

In Bogotá, one can distinguish 3 periods with O₃ increases with respect to the GAM-adjusted values. However, only the first episode of O₃ increase in late March was readily attributable to NO_x decreases. The remaining O₃ episodes did not show strong decreases in NO₂ as estimated by the GAM method. Thus, we hypothesize that in addition to the reduction in NO_x emissions, the increase in O₃ may have been due to long-range transport of O₃ from areas upwind of Bogotá.

The strongest simultaneous NO₂ decrease and O₃ increase occurred in Santiago, concurrently with the lockdown enforced over the wealthiest areas (March 23–April 10), highlighting the disproportionate role played by these communes in driving Santiago’s mobility. A secondary O₃ increase episode occurred around May 22, which also coincided with a decrease in GAM-adjusted NO₂. The third largest O₃ increase occurred on approximately May 11, that is, the International Labor Day, when decreased mobility was still accompanied by clear skies providing appropriate conditions for photochemistry. We also found more modest O₃ increases and NO₂ decreases on approximately May 11 in Bogotá and São Paulo.

Emissions were calculated using a bottom-up methodology (Tolvett Caro et al., 2016). This approach considered vehicle activity from several sources, as presented in Table S1. While both the approach and the data contain measurement errors and uncertainties, we deem the quality and consistency sufficient to assess the changes in emissions that occurred in connection with mobility restrictions in early 2020.

Instruments and methods used for measuring pollutants in governmental agencies follow international standards and are subject to quality assurance and control, which provides a robust dataset in all cities. Nevertheless, the data used in our study were subject to further quality checking by the authors; thus, we ascertained that we used a reliable dataset of comparable quality to those considered in other studies performed elsewhere in the world and in the region. The characteristics of each measurement station correspond to location, and their representativity may vary according to the pollutant, time of the year, and so on (e.g., Osses et al., 2013). These factors may result in uncertainties in estimating the air quality changes in cities that are heterogeneous.

The lack of time- and space-resolved VOC measurements in ambient air is an important limitation for characterizing the photochemical regimes in all 3 cities. However, the high agreement between the different observation techniques described in this article points toward the validity of our findings.

The consistency in observations is shown in Table 5, using NO₂ as an example. Both in situ and remote sensing methodologies indicated consistent fractional declines in NO₂. However, as these methodologies are different, they are not quantitatively comparable, but both show consistency regarding the changes observed.

In this study, we limited the length of lockdown to the active photochemical period. In these periods, we did not quantitatively examine the relationships between lockdown measures and public adherence.

The GAM calculations applied a Bayesian approach (Wood, 2017), where a covariance matrix was calculated to find the confidence intervals of the model estimate. Thus, the shaded area in Figure 10 corresponds to the confidence intervals between the 0.25 and 0.75 quantiles.

In Bogotá, Santiago, and São Paulo, the improvement in air quality due to emission reductions of primary pollutants at the start of the 2020 pandemic was lessened to some extent by the formation of secondary pollution, particularly in Santiago. Thus, air quality improved in terms of ground-level PM_2.5 at all stations, especially in industrialized areas of Bogotá. However, O₃ and secondary pollution worsened, mainly but not solely at the receptor sites in Bogotá, Santiago, and São Paulo.

The change in NO₂ vehicle emissions was nearly proportional to the in situ measurement of NO₂ mixing ratios and satellite-based tropospheric columns in Santiago and São Paulo, thus showing its direct impact on air quality and the importance of vehicle emissions in these cities. In all 3 cities, urban inequity results in differentiated emission patterns and pollution burdens, especially in Santiago, where the mobility of the wealthiest areas considerably impacts the rest of the city.

The vehicle emission rates of VOCs and NO_x calculated using available inventories for Bogotá, Santiago, and São Paulo revealed the different technological stages that these cities currently have for abating these pollutants.

The AOD decreased in the central locations of Santiago and São Paulo during the early lockdown. The AE indicated that slight reductions in particle size occurred during daylight hours in São Paulo and in the mornings in Santiago. The slight increase in the AE during the afternoon hours in Santiago, on the other hand, suggests an increase in coarser particles.

During lockdown periods, the O₃ sensitivity to daily maximum temperatures provides evidence of the expected air quality changes under a scenario of NO_x emission reduction, highlighting photochemistry’s key role in these cities. This relationship is key in estimating the O₃ climate penalty; however, it must be further investigated in line with other meteorological variables. Nevertheless, these sensitivities must be considered when developing air quality and climate mitigation strategies.

Because of the lack of time- and space-resolved VOC measurements in these megacities, the quantitative role of VOCs in O₃ formation and source attribution could not be addressed. However, the significant and consistent O₃ increases suggest that reactive VOC sources other than fossil fuel-powered vehicles played a key role in the photochemical production of NO₂ and subsequent formation of O₃ during the lockdown in these megacities.

Even though all cities showed substantial declines in mobility and vehicle emissions, particularly during the early lockdown, only Santiago exhibited sustained increases in oxidants (O_x = O₃ + NO₂) and secondary aerosols. In the case of São Paulo, the lack of observed sensitivity may be explained, in part, by less adherence over the whole and extended metropolitan area, which mitigated the changes that were only strictly enforced downtown. In Bogotá, despite large mobility and emission declines, the increase in oxidants was possibly mitigated by intense radiation. Hence, all cities are photochemically sensitive to changes in emissions, but this sensitivity is strongly modulated by local meteorology. In Bogotá, high altitude and intense ultraviolet radiation can have significant effects on driving photochemistry.

The GAM methodology provided further insights into the physical explanation of the observed changes. Interestingly, all cities showed a strong dependence on local meteorology once seasonality and weekly urban activity patterns were accounted for. This opens opportunities for forecasting secondary pollution and for addressing climate penalties. Beyond this statistical analysis, physical models and more detailed emission estimates are required to better understand the underlying physical and chemical processes and to better define mitigation options. This study presents the differing sensitivities in Bogotá, Santiago, and São Paulo to substantial but unevenly distributed mobility changes.

Finally, this unintended experiment highlighted that mobility plays a key role in all cities and the pivotal role of photochemical processes. However, their intertwined nature stresses that environmental policies must consider not only technological solutions but also behavioral changes and social, economic, and environmental inequity in our cities.

All data used in this article are publicly available in the repositories indicated in the methodology or supplemental material. Surface data of Bogotá, Santiago, and São Paulo were compiled at Pangaea (https://doi.pangaea.de/10.1594/PANGAEA.940302). Data on mobile source emissions for São Paulo can be found at https://github.com/atmoschem/vein/tree/master/projects/MASP_2020 and for Santiago and Bogotá at https://data.mendeley.com/datasets/y8drryy2yc/1.

The supplemental files for this article can be found as follows:

Figures S1–S8. Tables S1–S3. Docx

Traffic data for Santiago, March–April 2020, were kindly provided by Unidad Operativa de Control de Tránsito. We are grateful to the CETESB (www.cetesb.sp.gov.br) in São Paulo for providing access to air pollutant data. The open-source R programming language, the integrated development environment RStudio (https://www.r-project.org), and its package Openair (http://www.openair-project.org) were used for statistical computing and plotting pollutant hourly data. We thank Axel Osses for advising on the applicability of the generalized additive model method to model time series.

This work was supported by ANID/FONDAP/15110009; Prediction of Air Pollution in Latin America and the Caribbean project (ID: 777544, H2020-EU.1.3.3). MFA, EL, and TN acknowledge FAPESP (project 2016/18438-0). NH was also supported by the Research and Innovation programs under grant agreement 870301 (AQ-WATCH) and by the MAP-AQ, which is an International Global Atmospheric Chemistry (IGAC)- and World Meteorological Organization (WMO)-sponsored activity. MO acknowledges support from ANID PIA/APOYO AFB180002.

The authors declare no competing interests.

• Contributed to conception and design: RS, LG, MO, NR, TN.

• Contributed to the acquisition of data: RS, LG, MO, NR, TN, CM, PC, HE, SI, EL, ML, SM, FM, GM, NP, SP, JR, RR, ISA, RT, AY.

• Contributed to the analysis and interpretation of data: RS, LG, MO, NR, TN, CM, LB, NH, SI, EL, SM, NP, SP, RR.

• Drafted and/or revised this article: RS, LG, MO, NR, TN, CM, MFA, ZF, SM.

How to cite this article: Seguel, RJ, Gallardo, L, Osses, M, Rojas, NY, Nogueira, T, Menares, C, de Fatima Andrade, M, Belalcázar, LC, Carrasco, P, Eskes, H, Fleming, ZL, Huneeus, N, Ibarra-Espinosa, S, Landulfo, E, Leiva, M, Mangones, SC, Morais, FG, Moreira, GA, Pantoja, N, Parraguez, S, Rojas, JP, Rondanelli, R, da Silva Andrade, I, Toro, R, Yoshida, AC. 2022. Photochemical sensitivity to emissions and local meteorology in Bogotá, Santiago, and São Paulo: An analysis of the initial COVID-19 lockdowns. Elementa: Science of the Anthropocene 10(1). DOI: https://doi.org/10.1525/elementa.2021.00044

Domain Editor-in-Chief: Detlev Helmig, Boulder AIR LLC, Boulder, CO, USA

Associate Editor: Jochen Stutz, Department of Atmospheric and Oceanic Sciences, University of California, Los Angeles, CA, USA

Knowledge Domain: Atmospheric Science