Introduction

Population exposure to air pollution is a concept regarding how the local population can suffer from ambient pollution exposure during a particular length of time causing unfavorable health conditions [2, 6, 21, 36]. Based on the definition of the United Nations Office for Disaster Risk Reduction (UNDRR), exposure is the “situation of people, infrastructure, housing, production capacities and other tangible human assets located in hazard-prone areas”, which implies that population exposure involves environmental quality as well as population distribution and mobility, especially in high-density areas with high-density living, such as China. Specifically, China has experienced high pollution levels due to urbanization and industrial development [14]. Even on a low air pollution day, the pollution level exceeds WHO’s air quality thresholds or China air quality guidelines, inducing severe health risks across cities in China [5, 37]. Thus, previous studies have attempted to combine population datasets and urban mobility data (e.g., mobile phone data, bike trajectories) to estimate exposure to air pollution among various populations (e.g., general population, bike riders) in areas with high-density living [1, 3, 10, 19, 31].

PM2.5 (particulate matter with a diameter < 2.5 µm) is one of the riskiest air pollutants worldwide [9, 30, 34]. The results of Hammer et al. [12] were highly accurate (R2 = 0.81; slope = 0.90), thus, this open dataset could represent the micro-scale spatial variability of PM2.5 in Wuhan.

Fig. 3
figure 3

Annual PM2.5 concentration in Wuhan in 2014

Full size image

Hourly PM2.5 data retrieved from nine monitoring stations (Table 1) covering the study period (May 13–May 31, 2014), were applied to measure the temporal variability of ambient exposure. The following method based on the spatial varying data from the annual PM2.5 map and temporal varying data from monitoring stations was proposed to calculate the spatiotemporal concentration of PM2.5 at a certain location pi,j in Wuhan at different times. This method was adopted from a previous study to measure spatiotemporal exposure to PM2.5 based on satellite-derived data, air quality monitoring stations, and GPS-based wearable devices [43]. Particularly, PM2.5 concentration \({C}_{i,j}^{t}\) at location pi,j and time t can be calculated with the following formula:

$$C_{i,j}^{t} = C_{{{\text{base}}\;{\text{site}}}}^{t} + (C_{i,j} - C_{{{\text{base}}\;{\text{site}}}} ),$$
(1)

where Ctbase site is the temporal varying PM2.5 concentration retrieved from the representative station at time t, Ci,j and Cbase site is the spatial varying PM2.5 concentrations at location pi,j and representative station retrieved from the annual PM2.5 map. The representative station to calculate PM2.5 concentration Cti,j at location pi,j and time t should meet the following requirements: (1) no obvious outliers during the study period; (2) a minimum difference between the real-time concentrations from the monitoring station and the concentration of the corresponding position in the annual PM2.5 map. This Eq. (1) assumes that the spatial distribution of ambient PM2.5 exposure varied depending on multiple factors (e.g., built environment, urban structure), and the overall concentration is affected by temporal varying factors (e.g., weather condition of a day) and estimates the spatiotemporal variability of PM2.5 exposure in high resolution across the study area.

Table 1 PM2.5 stations in the monitoring network in Wuhan
Full size table

Dynamic population exposure to PM2.5 among taxi drivers

The daily trajectories for each taxi i were combined TRIPSi = {tripi1, tripi2, tripi3,…} to measure the dynamic population exposure to PM2.5 among taxi drivers. As depicted in “GPS-based taxi trajectory data”, all trips along the road network were generated after data preprocessing and each trip consisted of the following attributes, the start and end time (timestart and timeend), trip distance, and the turning point coordinates. The dynamic population exposure to PM2.5 among taxi drivers was estimated at the trip level, which can be expressed as:

$${\text{EI}}\left({\text{trip}}_{i}^{n}\right) = \lambda \int_{{{\text{time}}\,_{{{\text{start}}}} }}^{{{\text{time}}\,_{{{\text{end}}}} }} {{\text{Ce(}}t{)}dt,}$$
(2)

where EI(tripin) is the population exposure level of taxi driver i during a trip tripin, which is the integral of PM2.5 exposure where the driver was located with respect to time, λ is the air pollutant filtration coefficient considering a difference in PM2.5 exposure between the indoor environment of the taxi and the external environment. However, it was impossible to consider the internal and external protective measures (e.g., opening and closing windows, whether the driver wears a mask, whether there is air filtration equipment in the taxi). Given that this study assumed that all taxi drivers had similar air pollution filtering conditions in the working environment, λ was a fixed coefficient. Specifically, this study was based on a scenario in May 2014 when taxi drivers often opened windows when driving if the weather conditions were good, thus, this calculation was based on an open window scenario. Ce(t) is PM2.5 exposure in the environment where the taxi was located at time t. Based on the spatiotemporal data of PM2.5 concentration at the grid level estimated according to Eq. (1), the integral formula is transformed as follows:

$${\text{EI}}\left( {{\text{trip}}_{i}^{n} } \right) = \lambda \int_{{{\text{time}}{\kern 1pt} _{{{\text{start}}}} }}^{{{\text{time}}{\kern 1pt} _{{{\text{end}}}} }} {{\text{Ce(}}t)dt = \lambda \sum\limits_{{{\text{grid}}}} {\left( {C_{{{\text{grid}}}}^{t} \times \Delta t} \right)} } ,$$
(3)

where Ctgrid is PM2.5 exposure of the grid at time t, Δt is the elapsed time of taxi in the corresponding grid and each trip is split according to its relationships with grids to calculate the summation of PM2.5 exposure in each segment.

Subgroup analysis was conducted to evaluate the overall effect throughout the study period and weekday/weekend effects. Hotspot analysis was implemented to visualize and analyses the spatial–temporal variability of dynamic PM2.5 exposure among the top 1% of taxi drivers who experienced higher and lower levels of air pollution for each subgroup (e.g., weekdays, and weekends). Specifically, we chose the top 1% of taxi drivers with high and low PM2.5 exposure and analyzed the spatiotemporal variations of their activities, respectively. To guarantee the validity and robustness of the results, only taxi drivers who operated at least ten weekdays or four weekends were analyzed.

Kernel density estimation (KDE) was applied to estimate the hotspots for the above-mentioned two groups of taxi drivers based on their origins and destinations on weekdays and weekends. KDE is a method for smoothing data that converts a set of recorded observations, presented as geographically referenced point data, into a continuous surface [17]. This surface reflects the intensity of individual observations across space and is commonly used for hotspot analysis of air pollution exposure (e.g., [15]). In this study, ArcGIS 10.5 software was used to implement the hotspot analysis with a cell size of 0.001° × 0.001°.

Results

Spatial and temporal variability of ambient PM2.5 exposure

According to the current "Air Quality Standards" adopted in China, a PM2.5 concentration below 35 μg/m3 is “good” air quality, a concentration ranging from 35 μg/m3 to 75 μg/m3 is “moderate”, a concentration ranging from 75 μg/m3 to 115 μg/m3 is “unhealthy for sensitive groups” and a concentration exceeding 115 μg/m3 is “poor” [25]. In addition, WHO’s air quality guideline recommends that the 24-h average PM2.5 exposure should not exceed 25 µg/m3 [40].

Based on the spatial varying information retrieved from the annual PM2.5 map, the average PM2.5 concentration in Wuhan was 45.5 μg/m3, indicating an overall moderate air quality in Wuhan with higher pollution level in some central areas such as Qiaokou, Jianghan, Jiang'an, Wuchang, and Qingshan districts (Fig. 3). There was also high ambient PM2.5 exposure in the southwest of the Caidian and Hannan districts. However, a comparison of the temporal varying information retrieved from nine air pollution monitoring stations in Wuhan (Fig. 4) indicated that 13 days of the study period (May 13–May 31, 2014) had a pollution level above the threshold of “unhealthy for sensitive groups” (> 75 μg/m3), including 9 days with an ambient PM2.5 level exceeding 115 μg/m3. Furthermore, several days had pollution extremes. Ambient PM2.5 levels were almost 300 μg/m3. The peak PM2.5 pollution was between May 21, 2014 and May 27, 2014. In contrast, there were no significant regional differences between the nine monitoring stations in Wuhan, with a relatively consistent pattern of change over time observed. These results indicate that neither spatially varying data from the annual PM2.5 map nor temporal varying data from monitoring stations could comprehensively represent the spatiotemporal variability of PM2.5 exposure in line with our assumption to combine both datasets to better estimate population exposure.

Fig. 4
figure 4

PM2.5 data collected from the monitoring stations in Wuhan. a Donghu Liyuan, b Donghu Gaoxin, c Wujia Mountain, d Wuchang Ziyang, e Hankou Jiangtan, f Hankou Huaqiao, g Hanyang Yuehu, h Chenhu Qihao, and i Qingshan Ganghua

Full size image

Hourly dynamic exposure of PM2.5 among taxi drivers

To identify a representative station for estimating spatiotemporal variability of PM2.5 exposure, the differences in the data from all monitoring stations and the annual PM2.5 map, and the corresponding variances were calculated, as shown in Fig. 5 and Table 2. There was a large difference between May 21 and May 27, which was associated with severe pollution in Wuhan. The data distribution of all monitoring stations in the remaining days of the study period conformed to the annual distribution of PM2.5. The overall variances retrieved from Wuchang Ziyang (183.68 μg/m3) and Hankou Jiangtan (131.97 μg/m3) stations were the lowest, indicating better fitting as the representative monitoring station, thus Wuchang Ziyang station was selected as the representative station.

Fig. 5
figure 5

Concentration difference for each station in Wuhan. a Donghu Liyuan, b Donghu Gaoxin, c Wujia Mountain, d Wuchang Ziyang, e Hankou Jiangtan, f Hankou Huaqiao, g Hanyang Yuehu, h Chenhu Qihao, and i Qingshan Ganghua

Full size image
Table 2 The related variance for each station (μg/m3)
Full size table

Dynamic population PM2.5 exposure among taxi drivers at the trip level was then estimated (Fig. 6a), showing that there was higher PM2.5 exposure at midnight and in the morning (3:00–4:00 and 9:00–10:00 am). During these peaks, taxi drivers could be exposed to 86.61 μg/m3 (CI: 86.36–86.86 μg/m3) and 83.60 μg/m3 (CI 83.33–83.88 μg/m3) within an hour. These pollution levels were ~ 15.48% and 11.47% higher than the threshold for “unhealthy for sensitive groups”. However, the lowest average (11:00–12:00 pm) was only 60.90 μg/m3 (CI 60.71–61.10 μg/m3).

Fig. 6
figure 6

Hourly dynamics of PM2.5 exposure among taxi drivers in all days, weekdays and weekends across Wuhan during the study period. Mean represents the mean of exposure, upper and lower represent the 95% confidence interval of the exposure

Full size image

Additionally, PM2.5 exposure among taxi drivers during weekdays was generally higher than exposure at weekends (Fig. 6b, c). The peaks of weekday PM2.5 exposure were at midnight and morning with drivers exposed to above 95 μg/m3 within an hour between 3 and 9 am. In comparison, the peak hour of weekday PM2.5 exposure among drivers was in the evening and midnight and was ~ 59.79 μg/m3 (CI 59.51–60.06 μg/m3) at 7 pm during the weekend.

The hourly dynamic of PM2.5 exposure was also highly varied by day (Fig. 7). Overall, the representative monitoring station overestimated the dynamic PM2.5 exposures of taxi drivers. For the days with higher pollution levels retrieved from monitoring stations, hourly variations of PM2.5 exposure among taxi drivers were larger than the other days. Specifically, the most polluted days (May 21 and May 27) had large variations, with averages of 156.45 μg/m3 and 101.14 μg/m3 and standard deviations of 11.79 μg/m3 and 7.99 μg/m3.

Fig. 7
figure 7

Hourly dynamic of PM2.5 exposure in each day between May 13–May 31. Mean represents the mean of exposure, upper and lower represent the 95% confidence interval of the exposure, site means the exposure based on the monitoring stations

Full size image

Moreover, the 24-h average PM2.5 exposure among taxi drivers was calculated for each day (Fig. 8). During the weekdays, average 24-h PM2.5 exposure among taxi drivers was 83.60 μg/m3 (SD: 6.65 μg/m3), which was 234.4% higher than the WHO recommendation and 11.5% higher than the cutoff for “unhealthy for sensitive groups” noted in China’s guideline. During the weekend, 24-h average PM2.5 exposure among taxi drivers was 55.62 μg/m3 (SD: 4.72 μg/m3), which was approximately 25.8% lower than the pollution level during the weekdays. Although this pollution level did not reach the cutoff for “unhealthy for sensitive groups” noted in China’s guideline, it was still 2.2 times the WHO recommendation.

Fig. 8
figure 8

The daily 24-h average PM2.5 exposure among taxi drivers

Full size image

Spatio-temporal analysis of taxi drivers’ activities

Before the comparison of the spatiotemporal variations of activities between taxi drivers with high and low PM2.5 exposure, the distribution of the 24-h average PM2.5 exposure on weekdays and weekends was explored (Fig. 9), showing a normal distribution on both weekdays and weekends. On weekdays, the 24-h average PM2.5 exposure ranged from 16.44 μg/m3 to 162.74 μg/m3, whereas the 24-h average PM2.5 exposure ranged from 16.57 μg/m3 to 91.70 μg/m3 on weekends.

Fig. 9
figure 9

Distribution of the 24-h average PM2.5 exposure on weekdays and weekends across Wuhan in the study period

Full size image

We further explored the difference among taxi drivers with different levels of PM2.5 exposure based on their travel patterns using KDE analysis. As shown in Fig. 10, the two groups presented a similar distribution of hotspots on weekdays and weekends respectively, while the patterns of the two groups differed on both weekdays and weekends. Specifically, taxi drivers with high PM2.5 exposure had smaller major activity areas but a longer average trip distance than those with low exposure. For taxi drivers with high PM2.5 exposure, the average trip distances on weekdays and weekends were 12.26 km and 12.12 km, respectively compared to 4.82 km and 4.84 km for the taxi drivers with low PM2.5 exposure.

Fig. 10
figure 10

Hotspots of taxi drivers with high and low PM2.5 exposure on weekdays and weekends. As shown in a–d, where the dark red depicts the high-density areas, the high-density areas could correspond to the pollution hotspots

Full size image

The activity hotspots of the taxi drivers with high PM2.5 exposure were mainly in (1) busy riverside commercial areas within historic and central districts bounded by the “Inner Ring Road”, such as Jianghan Road (the marked red area in Fig. 10a, c) and Xudong, and (2) major transportation interchange/terminals (e.g., Hankou Railway Station). Specifically, Jianghan Road is a famous century-old commercial street in Wuhan located in the center of Hankou District with a total length of 1600 m. Jianghan Road and Xudong are also close to the Wuhan Yangtze River Tunnel and the Second Yangtze River Bridge, respectively, which are major roads with high traffic flow.

For the taxi drivers with low PM2.5 exposure, despite activity hotspots near major transportation interchange/terminals (e.g., Wuchang Railway Station), more hotspots overlapped with the new commercial areas in Wuhan, such as Optics Valley International Plaza (the marked red area in Fig. 10b, d) and Optical Valley Software Park within the “Third Ring Road”. These results suggested that the dynamic population exposure to PM2.5 among taxi drivers was related to their travel behavior.

Discussion

Interpretations of results

This study developed a method to rapidly assess dynamic population exposure among taxi drivers. An empirical study was conducted in Wuhan, China and subgroup analyses were performed to quantify population exposure in different scenarios (e.g., weekday/weekend effect). Our results indicated that neither spatially varying data from the annual PM2.5 map nor temporal varying data from monitoring stations could represent spatiotemporal variability of PM2.5 exposure independently. Based on the established equation to quantify dynamic population exposure among taxi drivers, it was found that taxi drivers are usually exposed to higher PM2.5 during the morning (3:00 am to 4:00 am). Furthermore, the peak hours of weekday PM2.5 exposure were at midnight and in the morning, while the peak hours of weekday PM2.5 exposure were during the evening and midnight. Additionally, taxi drivers with high PM2.5 exposure were typically clustered in commercial areas among central districts bounded by the “Inner Ring Road” and major transportation interchange/terminals, while taxi drivers with low PM2.5 exposure usually drove further in each trip with activity hotspots in new commercial areas bounded by the “Third Rind Road”, indicating a high spatiotemporal variability.

These results demonstrate how daily travel behaviors could affect the air pollution exposure of taxi drivers. For example, commercial areas in historic and central districts usually have high PM2.5 exposure due to the co-effects of road traffic and population mobility. These commercial areas are associated with PM2.5 pollution due to poor ventilation from high-rise, high-density urban morphology [

Fig. 11
figure 11

Two main hotspots: a Jianghan Road (Open-source photo from government site: http://www.jianghan.gov.cn/) and (b) Optics Valley International Plaza (Open-source photo from local newspaper: http://www.cnhubei.com/)

Full size image

In comparison, activity hotspots of taxi drivers with low exposure were in new commercial areas in extended urbanized areas such as Optics Valley International Plaza (Fig. 11b). Although these areas also have frequent road transit that can increase traffic-related PM2.5 exposure, they are better designed with better ventilation and a lower building density. Furthermore, these drivers need to drive further and pass through nearby districts for work which are usually residential areas or suburbs with more greenery and less traffic leading to lower PM2.5 exposure.

For major transportation interchanges/terminals, taxi drivers may be frequently exposed to PM2.5 pollution from both nearby road traffic and railway systems. Specifically, Hankou Railway Station and Wuchang Railway Station are two major train stations in Wuhan responsible for the departure or transit of northbound and southbound passengers every day. Thus, trains frequently pass through the railway systems near these stations daily. More importantly, there are several nearby bus terminals with a large flow of people and vehicles including interurban coaches and local buses as well as cars and taxis. This population dynamic potentially increases taxi drivers' exposure to traffic-related PM2.5 during the peak hour of transit and partially explains why taxi drivers could also be exposed to high PM2.5 exposure at midnight on both weekdays and weekends, as they might serve customs to connect to intracity trains and coaches.

The pollution episode between May 21—27 matched with the local news from China, for example, the People’s Daily reported that Wuhan had the second worst air pollution among 161 Chinese cities (http://env.people.com.cn/n/2014/0521/c1010-25044033.html), possibly due to regional climate but this requires further exploration.

Health and policy implications

Based on the results, taxi drivers who work in urban areas could be at risk, especially those who frequently drive to commercial areas in central districts and major transportation interchanges. Specifically, the average PM2.5 exposure among taxi drivers during weekdays was already 11.5% higher than the WHO’s threshold and cutoff for “unhealthy for sensitive groups” noted in China’s guideline. Those who frequently drive to central districts and major transportation interchanges may be exposed to a higher level of PM2.5, thus taxi drivers in urban China could be at risk of severe health risks due to their high PM2.5 exposure during working hours.

It is important to note that long-term exposure to PM2.5 can induce lung cancer among the urban Chinese population. For example, a study of the Yangtze River Delta (YRD) region [36] found that ~ 12,574–14,504 lung cancer deaths were attributable to PM2.5 in YRD, and the deaths in urban areas could be 7–13 times higher than those in rural areas. Guo et al. [11] also found that long-term exposure to PM2.5 could result in a much higher incidence rate and mortality risk among urban males than rural males across 353 counties in China. Long-term exposure to PM2.5 in China is also associated with cardiovascular and respiratory diseases [20, 42]. Furthermore, frequently driving a taxi might also affect the driver’s physical activity and diet resulting in comorbidities such as obesity, hypertension, diabetes, and related chronic diseases that could further increase the risk of PM2.5 exposure.

Thus, it is important to provide community health training (e.g., program to enhance knowledge, attitude, and practice) to increase the health awareness of taxi drivers [24]. Taxi drivers should also enhance their environmental awareness, such as knowledge to install of air purifiers and attitude regarding frequently checking air filters. Additionally, knowledge to enhance healthy behaviors should also be provided to minimize potential risks from comorbidity. Raising such awareness among taxi drivers about the risks of PM2.5 exposure and providing guidance on protective measures can contribute to their well-being.

Additionally, strategies to improve urban design should also be applied to reduce the risk of long-term PM2.5 exposure. For example, urban greening should be applied along the major roads to remove air pollutants [18] and future major roads could be built along wind corridors to enhance urban ventilation so that less PM2.5 is trapped in the high-rise, high-density environment [7].

Limitations

Several limitations need to be noted in this study for future research: (1) the representative station was near major traffic spots and although these stations were more statistically stable for map** and analysis, the pollution level was much higher than the average dynamic exposure. Therefore, other types of monitoring stations should be used for comparison to quantify the local risk and differentiate different local scenarios (e.g., air pollution level in residential areas). (2) Our base map that represents spatial variability of PM2.5 exposure was from a nationwide model (spatial resolution: 1 km) and although this model is representative of the China scenario, enhanced results for local scale with finer resolution should also be applied to enhance the estimation. Specifically, it is recommended to use real-time traffic volume data and data from various emission sources for local modelling and validation in the future. (3) This study followed previous studies [43] to use the difference between the monitoring data and reference data for the temporal adjustment. However, for temporal adjusting spatially rich data, the ratio between the monitoring data and reference data in the background or the nearest monitoring site may also be appropriate, and different methods may yield the results. Future studies should compare the differences and uncertainties between various types of temporal adjustment. (4) As the data on taxi driver’s behaviors were retrieved from GPS information, we were not able to obtain sociodemographic information on each driver for further analysis, so future studies should consider using multiple sources of big data (e.g., taxi apps with drivers’ information) to enhance the data analysis.