The Alerting Effect from Rising Public Awareness of Air Quality on the Outdoor Activities of Megacity Residents

This study investigated how the public awareness of air quality affects people’s decisions to participate in outdoor activities. Given that the keyword search volume of particulate matter (PM) in Seoul, South Korea started to grow dramatically only after November 2013, we defined two periods (low and high public awareness of PM) and conducted a series of comparative analyses to investigate the impact of public awareness of air quality on the relationship between PM level and people’s outdoor activities. In the low public awareness period, people’s outdoor activities measured by the number of daily subway passengers did not significantly vary over PM levels, even in the ’unhealthy’ range (80 < PM10 < = 150 µg/m3). On the contrary, during the high awareness period, people’s activities were significantly affected by the PM level, even in the ’moderate’ range. Specifically, the perceived safety threshold of PM10 level that people use to decide for engaging outdoor activities has decreased from 120 to 70 µg/m3. These results suggest that public awareness of air quality and its harmful ramifications on health is a key determinant of outdoor activities rather than PM10 concentration itself. Thus, this study alludes to a need for more timely and effective dissemination of air quality information to the public as much as for curbing anthropogenic emissions.

In the context of increasingly concentrated air pollution, it is critical to prevent people from exposure to harmful environments. As such, modern governments should protect the public from

Index and Comparison Days
In general, the relationship between air pollution and human behavior can be confounded by other factors that affect the latter, such as the day of the week, seasonal trends, and weather conditions. To control for such confounding effects caused by long-term time series patterns, weekly and annual variations, temperature, and precipitation, we compared an index day group with a symmetrical comparison day group [9,11,15,32].
Paired t-tests were conducted on the numbers of daily subway passengers for the index and comparison day groups to investigate the relationship between PM 10 and public outdoor activities. 'Index' days are defined as the days with a daily mean PM 10 concentration higher than a certain concentration threshold. 'Comparison' days are defined as the days a week before or after the index days if the daily mean PM 10 concentration of those days is lower than the same concentration threshold. Therefore, an index day can be paired with zero, one, or two comparison days. If an index day had zero comparison days, the set was discarded. If two comparison days were available, their average was paired with the index day. With this approach, we controlled for the confounding effects caused by weekly cycles, seasonal variation, annual trends, and other meteorological conditions, such as temperature. Holidays were screened to remove the weekly cycle properly. As this data sampling process normalizes the data, the differences of subways passengers between index days and comparison days represent the none-routine portion of the total subway passengers and are compared with the outdoor activity defined earlier (i.e., non-routine part of total activity).
Controlling for weather conditions should be noted in more detail. Pairing data a week before and after can control for weather conditions with a longer frequency (e.g., annual variation). However, short-term variation (e.g., daily variation) is not completely separated because these short-term weather conditions are strongly tied to PM 10 concentration, as discussed later. Precipitation, however, is too critical for PM 10 concentration. Hence, we utilized a rain flag (i.e., 'rain' or 'no rain') to compare index days with comparison days under the same precipitation conditions to reduce noise effects.

High and Low Public Awareness Periods
While the surface PM 10 concentration level in South Korea was high in the early 2000s, it has drawn little public attention until recent years. People used to consider severe haze cases to be a part of the Asian dust phenomenon, and few paid serious attention to aerosols with anthropogenic sources. The main trigger for the sudden increase in public interest and concern about the PM level was the announcement of the World Health Organization (WHO) in October 2013 that PM is a cancer-causing agent. The International Agency for Research on Cancer (IARC), a part of the WHO, has classified outdoor air quality as a carcinogenic substance, along with previously reported substances including diesel engine exhaust emissions, solvents, metals, and dust. In its evaluation, the IARC concluded that outdoor air pollution may cause lung or bladder cancer. Figure 1 shows the time series of monthly PM 10 concentration in Seoul, along with Google's SVI and the number of articles on PM during 2008-2016. Although PM 10 concentration in Seoul was high in the early years, there are few search volumes or news articles until October 2013. However, both search volume and the number of news articles escalated after November 2013. For further analysis, we defined two periods of public awareness of PM before and after November 2013: The low public awareness (January 2008 to October 2013) and the high public awareness (November 2013 to December 2016).
Sustainability 2020, 12, 820 4 of 13 'Index' days are defined as the days with a daily mean PM10 concentration higher than a certain concentration threshold. 'Comparison' days are defined as the days a week before or after the index days if the daily mean PM10 concentration of those days is lower than the same concentration threshold. Therefore, an index day can be paired with zero, one, or two comparison days. If an index day had zero comparison days, the set was discarded. If two comparison days were available, their average was paired with the index day. With this approach, we controlled for the confounding effects caused by weekly cycles, seasonal variation, annual trends, and other meteorological conditions, such as temperature. Holidays were screened to remove the weekly cycle properly. As this data sampling process normalizes the data, the differences of subways passengers between index days and comparison days represent the none-routine portion of the total subway passengers and are compared with the outdoor activity defined earlier (i.e., non-routine part of total activity). Controlling for weather conditions should be noted in more detail. Pairing data a week before and after can control for weather conditions with a longer frequency (e.g., annual variation). However, short-term variation (e.g., daily variation) is not completely separated because these shortterm weather conditions are strongly tied to PM10 concentration, as discussed later. Precipitation, however, is too critical for PM10 concentration. Hence, we utilized a rain flag (i.e., 'rain' or 'no rain') to compare index days with comparison days under the same precipitation conditions to reduce noise effects.

High and Low Public Awareness Periods
While the surface PM10 concentration level in South Korea was high in the early 2000s, it has drawn little public attention until recent years. People used to consider severe haze cases to be a part of the Asian dust phenomenon, and few paid serious attention to aerosols with anthropogenic sources. The main trigger for the sudden increase in public interest and concern about the PM level was the announcement of the World Health Organization (WHO) in October 2013 that PM is a cancercausing agent. The International Agency for Research on Cancer (IARC), a part of the WHO, has classified outdoor air quality as a carcinogenic substance, along with previously reported substances including diesel engine exhaust emissions, solvents, metals, and dust. In its evaluation, the IARC concluded that outdoor air pollution may cause lung or bladder cancer. Figure 1

Environmental Variables and Outdoor Activity
We first analyzed the change in the number of subway passengers according to the environmental variables (e.g., temperature, precipitation, and PM concentration) and temporal variations (e.g., seasonal and weekly variations). Since the outdoor activity level in Seoul, Korea showed a gradually increasing trend, we removed the annual trend by normalizing with the annual average. The data on subway passengers were transformed into a normalized subway passenger (nSP) value for each of the seasonal and environmental factors to avoid the effect of an increasing trend over time. nSP ij , the index value of factor level i in year j, is defined as follows: where SP ij is the average number of daily subway passengers for factor level i in year j and SP j is the average number in year j. For instance, if we were to inspect the effect of month, the index value for month i in year j would be calculated by dividing the average number of daily subway passengers for month i in year j (SP ij ) by the daily average number in year j (SP j ) for every i = 1, 2, . . . , 12 and j = 2008, 2009, . . . , 2016. The number of subway passengers differed according to the time of year (e.g., month and day of the week). Figure 2a shows the boxplot of the nSP values for each month. Monthly seasonality presents a low level of outdoor activities in January, February, and August, which implies people aim to avoid extreme heat or cold. On the contrary, the level of outdoor activity is high in spring (e.g., March to May) and autumn (e.g., October and November) when fine weather is expected. December, although already in the cold season, also has high activity patterns because it includes the end-of-year holiday season. The weekly variation in outdoor activities is apparent in Figure 2b. nSPs are high during weekdays, with the highest ratios on Friday, and low during weekends, showing the lowest frequencies on Sundays and holidays.
Weather conditions had modest but significant effects on the number of daily subway passengers (Figure 2c,d). On days with a daily mean temperature of 5-15 • C (41-59 • F), the nSPs are the highest. For each 10 • C bin, the outdoor activity nSPs change by around 3%-4% per 10 • C temperature change. As expected, nSPs are high on no rain days and tend to decrease as the amount of precipitation increases. Moreover, nSPs also seem to be spread widely for higher precipitation, implying higher uncertainty in predicting patterns of subway passenger counts in high precipitation conditions. The comparison of PM 10 concentration with nSPs reveals an interesting correlation between PM 10 concentration and activity. Figure 3 shows the nSPs and the detrended number of daily subway passengers according to PM 10 concentration. Four concentration intervals are set to the four stages of the AQI. Unlike initial expectations, the nSPs have a positive correlation with PM 10 concentration. The number of daily subway passengers is the lowest in the 'good' stage and increases as the AQI moves toward the 'moderate' and 'unhealthy' stages. The mean of the nSPs for the 'very unhealthy' condition is slightly lower than that in the 'unhealthy' case, whereas it has a wider spread of values with more extreme cases either higher or lower than the values in the 'unhealthy' stage. This result is intuitive if we consider the high correlation between synoptic weather and pollution level [33][34][35][36][37]. High PM concentration events tend to happen in anticyclonic weather conditions, under which we typically have a warmer temperature, higher surface pressure, calm and stagnant wind velocity, and less or no precipitation. Therefore, we cannot directly conclude that there is a positive correlation between PM 10 concentration and outdoor activities (Figure 3) because the effects of weather and pollution might be combined. People seem to determine their outdoor activities primarily on the basis of the weather conditions as far as the air quality is good or moderate. However, when the air pollution becomes severe, the air quality level becomes a deciding factor for one's outdoor decision. Sustainability 2020, 12, 820 6 of 13 The comparison of PM10 concentration with nSPs reveals an interesting correlation between PM10 concentration and activity. Figure 3 shows the nSPs and the detrended number of daily subway passengers according to PM10 concentration. Four concentration intervals are set to the four stages of the AQI. Unlike initial expectations, the nSPs have a positive correlation with PM10 concentration. The number of daily subway passengers is the lowest in the 'good' stage and increases as the AQI moves toward the 'moderate' and 'unhealthy' stages. The mean of the nSPs for the 'very unhealthy' condition is slightly lower than that in the 'unhealthy' case, whereas it has a wider spread of values with more extreme cases either higher or lower than the values in the 'unhealthy' stage. This result is intuitive if we consider the high correlation between synoptic weather and pollution level [33][34][35][36][37]. High PM concentration events tend to happen in anticyclonic weather conditions, under which we typically have a warmer temperature, higher surface pressure, calm and stagnant wind velocity, and less or no precipitation. Therefore, we cannot directly conclude that there is a positive correlation between PM10 concentration and outdoor activities (Figure 3) because the effects of weather and pollution might be combined. People seem to determine their outdoor activities primarily on the basis of the weather conditions as far as the air quality is good or moderate. However, when the air pollution becomes severe, the air quality level becomes a deciding factor for one's outdoor decision.

Weather and Pollution Effects on Activity
To examine the activity-pollution correlation more carefully, a series of paired t-tests were designed and conducted. As described in the Method Section, the index days and comparison days were selected to examine the sensitivity of PM10 concentration to the number of subway passengers. Conceptually, index days mean relatively polluted conditions compared with comparison days. If the number of subway passengers is lower on index days than on comparison days, this can be interpreted as the pollution level discouraging people from going outside. Figure 4 and Table 1 compare the number of subway passengers according to the index days and comparison days as well as their p-values from the paired t-tests. PM10 concentration on the xaxis denotes the PM10 concentration thresholds used to define each index day. It is notable that the number of subway passengers on index and comparison days tend to diverge as PM10 threshold levels

Weather and Pollution Effects on Activity
To examine the activity-pollution correlation more carefully, a series of paired t-tests were designed and conducted. As described in the Method Section, the index days and comparison days were selected to examine the sensitivity of PM 10 concentration to the number of subway passengers. Conceptually, index days mean relatively polluted conditions compared with comparison days. If the number of subway passengers is lower on index days than on comparison days, this can be interpreted as the pollution level discouraging people from going outside. Figure 4 and Table 1 compare the number of subway passengers according to the index days and comparison days as well as their p-values from the paired t-tests. PM 10 concentration on the x-axis denotes the PM 10 concentration thresholds used to define each index day. It is notable that the number of subway passengers on index and comparison days tend to diverge as PM 10 threshold levels become either extremely low or high. When the PM 10 threshold is lower than 90 µg/m 3 , the subway passenger counts are higher on index days than on comparison days. This pattern is the opposite for the PM 10 threshold higher than 90 µg/m 3 .

Effect of Public Awareness
Public interest in PM can be clearly shown in the monthly trend of Google's SVI on 'particulate matter' and the number of news articles (Figure 1). While the annual level of surface PM10 concentration in South Korea has declined continuously since 2008, the monthly search volume and number of news articles suddenly increased after November 2013. Moreover, since then, the correlation between the monthly SVI and monthly PM10 concentration had increased rapidly. The  This result is consistent with the activity-pollution correlation shown in Figure 3, implying the potential role of weather in promoting outdoor activities in less polluted conditions. As we addressed earlier, pairing data with ±7 days can control for annual variations in meteorology, but this approach Sustainability 2020, 12, 820 8 of 12 still contains signals from daily variations in weather conditions. In most cases, the weather conditions on index days are warmer and less windy than on comparison days (see Supplementary Materials Figure S2). As aforementioned, this result confirms our inference that two main factors (i.e., weather and pollution) determine outdoor activities. Thus, we named the region above PM 10 90 µg/m 3 as the 'pollution-considered zone' and the region below that level as the 'weather-considered zone' (Figure 3).
The p-values in Table 1 also confirm the pattern of two-tailed decision making. Below the 90 µg/m 3 PM 10 threshold, the difference between index days and comparison days is statistically significant (p < 0.05), implying the role of weather conditions in people's decisions to participate in outdoor activities. Over the 130 µg/m 3 PM 10 threshold, the difference between index days and comparison days is statistically significant at the p = 0.1 level, implying people tend to avoid going out when PM pollution is high.

Effect of Public Awareness
Public interest in PM can be clearly shown in the monthly trend of Google's SVI on 'particulate matter' and the number of news articles (Figure 1). While the annual level of surface PM 10 concentration in South Korea has declined continuously since 2008, the monthly search volume and number of news articles suddenly increased after November 2013. Moreover, since then, the correlation between the monthly SVI and monthly PM 10 concentration had increased rapidly. The Pearson correlation between PM 10 concentration and the SVI showed a clear increase, changing from 0.196 before November 2013 to 0.725 thereafter.
We analyzed the impact of this increased public interest in PM on the actual pattern of public outdoor activities. We separated the data into two periods, namely before and after November 2013 when public interest in PM dramatically increased, which we defined as low and high awareness periods. We applied the same paired t-test analysis between index and comparison days, as explained in the analysis Methods Section, and the results are presented in Figure 5 and Table 2.    In the low public awareness period (Figure 5a), the number of subway passengers on index days is significantly higher than the number on comparison days, implying that weather conditions are a factor in people's decisions to participate in outdoor activities. During the low public awareness period, people seem to refer to the pollution level for their outdoor activity decisions only when PM 10 concentration is higher than 120 µg/m 3 , although we cannot confirm its significance.
Interestingly, this pattern changed dramatically after November 2013 (Figure 5b). In the high public awareness period, the threshold for differentiating between the weather-considered zone and pollution-considered zone shifted significantly to the level of PM 10 70 µg/m 3 . The number of subway passengers on index days is almost the same as or slightly higher than the number on comparison days in the weather-considered zone, whereas it is significantly lower in the pollution-considered zone over 70 µg/m 3 . Thus, it is evident that people began to use pollution information as a way to decide on their outdoor activities to avoid potential adverse health effects.
Lastly, Figure 6 shows the percentage change in the number of subway passengers on index days against comparison days in the 'moderate' and 'unhealthy' stages during the low and high awareness periods. People's decision-making pattern for outdoor activities has clearly changed. People used to consider the weather conditions rather than the pollution factor in the 'moderate' PM condition in the low public awareness period. However, in the high awareness period, people began to consider both weather and pollution factors in the 'moderate' pollution condition and consider the pollution factor four to five times more seriously in the high awareness period, deciding not to go outside when they perceived high pollution. This result clearly shows the effectiveness of pollution-related information in the decision making of megacity residents on whether to participate in outdoor activities. Finally, further studies to investigate if this behavior change resulted in the actual exposal reduction and health impact improvement are warranted in the future.
to consider both weather and pollution factors in the 'moderate' pollution condition and consider the pollution factor four to five times more seriously in the high awareness period, deciding not to go outside when they perceived high pollution. This result clearly shows the effectiveness of pollutionrelated information in the decision making of megacity residents on whether to participate in outdoor activities. Finally, further studies to investigate if this behavior change resulted in the actual exposal reduction and health impact improvement are warranted in the future.

Conclusions
Air pollution has become a pressing priority of environmental, economic, and health problems globally. Recent studies have documented the harmful health effects of particle pollution exposure. Countries and regulatory agencies have developed air quality guidelines and proper monitoring systems to mitigate the harmful effects by minimizing public exposure to unhealthy air pollution. However, our understanding of outdoor activity patterns influenced by outdoor air pollution has been limited. The objective of this study was to enhance our understanding by investigating the effect of PM, a major component of air pollution, on public outdoor activities using empirical data on PM10 levels and the daily volume of subway passengers recorded for 2008-2016.
The key findings of this study are as follows. First, the level of PM tends to suppress public outdoor activities in polluted conditions, implying that people react to avoid such a hazardous environment. Combined with the traditional explanation of weather-activity relations, we conclude that people consider weather conditions when deciding on outdoor activities in less polluted conditions, while the pollution level is more important in their decision in more polluted conditions. Second, public awareness of PM has shifted the sensitivity of people's responses for the frequency of going out. Since November 2013, when public awareness became higher, external activities have decreased significantly in a PM10 range over 70 μg/m 3 , which used to be much higher (i.e., 120 μg/m 3 ) in low public awareness. Third, the shift of activity is statistically significant in the 'unhealthy' stage (80 < PM10 < = 150 μg/m 3 ), suggesting that people's reactions may reflect the stages of the AQI

Conclusions
Air pollution has become a pressing priority of environmental, economic, and health problems globally. Recent studies have documented the harmful health effects of particle pollution exposure. Countries and regulatory agencies have developed air quality guidelines and proper monitoring systems to mitigate the harmful effects by minimizing public exposure to unhealthy air pollution. However, our understanding of outdoor activity patterns influenced by outdoor air pollution has been limited. The objective of this study was to enhance our understanding by investigating the effect of PM, a major component of air pollution, on public outdoor activities using empirical data on PM 10 levels and the daily volume of subway passengers recorded for 2008-2016.
The key findings of this study are as follows. First, the level of PM tends to suppress public outdoor activities in polluted conditions, implying that people react to avoid such a hazardous environment. Combined with the traditional explanation of weather-activity relations, we conclude that people consider weather conditions when deciding on outdoor activities in less polluted conditions, while the pollution level is more important in their decision in more polluted conditions. Second, public awareness of PM has shifted the sensitivity of people's responses for the frequency of going out. Since November 2013, when public awareness became higher, external activities have decreased significantly in a PM 10 range over 70 µg/m 3 , which used to be much higher (i.e., 120 µg/m 3 ) in low public awareness. Third, the shift of activity is statistically significant in the 'unhealthy' stage (80 < PM 10 < = 150 µg/m 3 ), suggesting that people's reactions may reflect the stages of the AQI provided by the government. This fact implies that the AQI intervals should be determined carefully based on actual epidemiological research.
To conclude, this study demonstrates that public interest in PM is a major factor affecting people's decisions to participate in outdoor activities rather than PM 10 concentration itself. Hence, timely and accurate information on pollution is as important as emission control efforts in reducing public exposure to harmful environments.
Supplementary Materials: The following are available online at http://www.mdpi.com/2071-1050/12/3/820/s1, Figure S1: Spatial distribution of annual mean surface PM10 concentration over South Korea (left) and Seoul Metropolitan Area. Thick black line indicates the administrative boundary of Seoul. Figure S2