Analysis of the Impact of Road Trac on the Urban Air Quality

With a focus on the hot topics of trac congestion and smog (air quality), in this study, the impact of road trac on urban air quality was the rst quantitatively analyzed based on the aerosol optical thickness (AOD) and geographical weighted regression (GWR) models, including the road network density, road area occupancy, intersection number, and bus network density. The main research conclusions are as follows. There is a strong positive correlation between the peak congestion delay index (PCDI) and air quality. Based on the GWR model, AOD has high correlations with four road network trac characteristic parameters, and these correlations are much higher than ordinary linear regression, that is, GWR renes the local spatial changes in the AOD and the road parameters. The correlation was mainly positive. The correlation between AOD and the road area occupancy was the highest, and the correlations between PM2.5 and the density of the bus network and the number of intersections were higher than those with the road network density. Thus, bus route planning, bus emission reduction, road network planning, and signal timing (intersection) have a greater impact on air quality, especially in areas with trac jams. This study has a certain guiding signicance for trac planning and trac control, and it provides support and a basis for trac planning and control.


Introduction
As the largest developing country, China's air quality has always been a focus of research. Air pollution and the processes that generate air pollution exhibit spatial heterogeneity. According to the pollutant analysis results released by Beijing, Jinan, Hangzhou, vehicle exhaust has surpassed coal as the main source of urban air pollution (especially PM2.5). By June 2020, the number of motor vehicles in China had reached 360 million, and tra c congestion had become the norm in many Chinese cities. As a result, the contribution rate of vehicle exhaust pollution to air quality will continuously increase. Its contribution to PM2. 5 . Few studies on the impacts of road network tra c characteristics (e.g., road density, intersection, and bus network density) on air quality have considered the spatial heterogeneity.
Air pollution and the process of producing air pollution exhibit spatial heterogeneity. The geographically weighted regression (GWR) model considering spatial heterogeneity was used to analyze urban air pollutants, and the results revealed that the GWR is better than regression model ( 2019). The retrieval of the air pollution status based on remote sensing data not only makes up for the lack of observation data, but also re ects the spatial distribution characteristics of the air pollution, so remote sensing inversion has become an important method for studying air pollution. Previous studies have shown that there is a certain correlation between the aerosol optical thickness (AOD) and the concentration of near surface particles, and the AOD product of MODIS (Moderate Resolution Imaging Spectroradiometer) is the most widely used in air pollution research (Chu et al, 2003; Slater et al, 2004; Wang et al, 2003;Gupta et al,2006;Tao, et al,2013). The objective of this study is to quantitatively analyze the impacts of road tra c characteristics on urban air quality based on their spatial heterogeneity.

Study area
The main urban area in Jinan (non-administrative boundaries, Fig. 1) was taken as the study area for two reasons: severe haze and tra c congestion. Jinan has been ranked among the cities in China with the (2) Weather: it was sunny with northeast winds of < 3 m/s, which is favorable for remote sensing data. (3) Air quality: the air in Jinan was mildly polluted, with a mean PM2.5 value of 85 µg/m 3 . The tra c, weather, and air quality in the study area on December 26, 2017, were the most consistent with the study requirements.

Data
In total, monitoring data from 11 air quality monitoring stations in the study area (https://www. aqistudy.cn) and two stations that were deployed by two research teams were utilized. The distribution of the monitoring stations was shown in Fig. 2.
Two types of remote sensing data were used in this study: MCD19A2 AOD products and landsat8 OLI images. MCD19A2 is the o cial 1 km resolution AOD products of MODIS, and it is produced using the aerosol algorithm in MAIAC (the multi-angle atmospheric correction algorithm). Compared with the 10 km and 3 km resolution MOD04 aerosol products, the MCD19A2 has a higher resolution (study area located in H27V05). The column number of the Landsat8 image of Jinan was 12235. Both datasets were collected on December 26, 2017.
Roads less than 3 m wide were excluded from the road network in the study area. The network of public transport routes in the study area is shown in Fig. 2.
As a measure of the number of hours of peak congestion delay, the Peak Congestion Delay Index (PCDI) is an indicator used to evaluate the degree of urban congestion, which typically has peaks at 7:00-9:00 in the morning and at 17:00-19:00 in the evening. It is a visual representation of a city's tra c operations, and it re ects the exhaust emissions. The PCDI was obtained from China's Major Urban Transport Report released by AMap.

Air quality distribution based on MCD19A2
The MCD19A2AOD products used in this study were taken on December 26, 2017. The AOD data for the different orbits (there are 4 tracks in the daily AOD data) were combined into the daily AOD, and the AOD in the study area is shown in Fig. 3.

Gridding
Using the gridding tool, the study area was divided into 1 km x 1 km grids (Fig. 4), with less than 1 km remaining on the northern and eastern sides of the study area being excluded from the gridding). The AOD, road network density, road area ratio, and number of intersections in each grid were calculated.

GWR-based analysis of the impact of the road network tra c characteristics on the urban air quality
Geographically weighted regression (GWR) is a spatial analysis technique that is widely used in geography and related disciplines involving the analysis of spatial patterns, and it can be used to quantify spatial heterogeneity. It has a higher accuracy because the local effects of spatial objects (i.e., the spatial heterogeneity) are taken into account.
The GWR analysis was based on the spatial correlation test of PM2.5 distribution in the study area. The spatial autocorrelation analysis tool in ARCGIS (the Moran's Index I) was used for the analysis. Moran's index I is mainly used to measure the spatial distribution characteristics of the data throughout the entire region.
The Moran's index I distributions of PM2.5was 0.49 in the study area, the distributions of the PM2.5 was closely correlated, had obvious spatial clustering characteristics, and was heterogeneous, which makes them suitable for GWR analysis.

Correlation analysis
The correlation between the PCDI in each quarter for the past ve years (2016 to Q1 2020) and the PM2.5 (the average values of each quarter in Jinan) was analyzed, and the tting diagram is shown in Fig. 5.
The correlations between PM2.5 and the gridded density road network, the road area occupancy, the number of intersections, and the network density of the bus lines were analyzed (linear regression), and the results revealed that their correlations were all low.

Tra c and air quality
As can be seen from Fig. 5 the correlation coe cients between the PCDI and the PM2.5 has R 2 values of up to 0.4962 (R 0.70). The data were paired separately and the F Test was conducted in the two groups of data when p < 0.05, i.e., the samples was variance congruent. Then the t-Test was conducted using the two-sample equal variance hypothesis. With a = 0.05, all p < 0.05, that was, the tted linear equation passed the signi cance test. The results demonstrate PM2.5 is positively correlated with PCDI. According to the source analysis of PM2.5 in Beijing, Jinan, and Hangzhou, exhaust gases surpassed coal combustion as the main source of PM2.5 pollution in cities, that is, tra c (exhaust) has a large impact on urban air quality, especially in tra c jams. However, the correlations between AOD and the gridded density road network, the road area occupancy, the number of intersections, and the network density of the bus lines (linear regression) were all low, so the GWR model considering spatial heterogeneity was used.

Single-parameter impact analysis
The AOD model with road network density, road area occupancy, number of intersections, and bus network density were constructed separately based GWR. The contrast between the input parameters in the GWR output tool is shown in Table 1. As can be seen from Table 1, the PM2.5 based on the GWR model analysis was closely correlated with the road network density, road area occupancy, and the number of intersections. That is, PM2.5 has strong heterogeneities, so the GWR model was used to re ne the characterization of the AOD distribution and the local spatial variation in the road network in the study area. The regression parameters for each variable were positive or negative, i.e., each factor had a facilitating or inhibiting impact on air quality in the different regions. However, based on the median of the regression coe cients for each variable, it is clear that the road network density, road area occupancy, and number of intersections had signi cant impacts on the regional AOD, and they were positively correlated.
The road area occupancy was exhibited the highest correlations with AOD, i.e., during tra c congestion, the road occupancy area had the greatest impact on the air quality in the region. AOD was better correlated with the density of the bus route network than with the density of the road network, which means that buses had a greater impact on the urban AOD, and reducing pollution from buses is important for reducing urban AOD concentrations. AOD was more signi cantly correlated with the number of intersections than with the density of the road network, i.e., the setting of the tra c lights (tra c signals) within the regional road network. The number of intersections and the network of bus routes had signi cant impacts on the exhaust emission (i.e., the regional air quality).

Parameter autocorrelation analysis
In this study, four parameters were selected to analyze the impacts of the road network density, the road area occupancy, the number of intersections, and the bus route network density on urban air quality in a congested environment, and all four parameters had signi cant impacts on the air quality in the corresponding areas. To avoid multicollinearity among the variables, the correlations between the four parameters themselves were analyzed before constructing the multi-parameter GWR model (Table 2). As can be seen from Table 2, the three parameters, including the road network density, the road area occupancy, and the number of intersections, were signi cantly correlated. In particular, the correlation coe cient between network density and road area occupancy was 0.924, but the correlation between road area occupancy and intersections was weaker (0.7891). The correlations between the bus route network density and the other three parameters were less signi cant (less than 0.3). Thus, road area occupancy and bus route network density were used to construct the GWR model, and the variance in ation factor (VIF) of the independent variables test was determined for the two variables (i.e., road area occupancy and bus route network density). VIF = 1.33, i.e., the collinearity between the independent variables was small, but meets the requirements of the regression analysis, and thus, the GWR modelbased results are credible.

Error analysis based on the GWR model
The standard deviations of the AOD models constructed based on the GWR in the study area is shown in Fig. 6.
As can be seen from Fig. 6, fewer grids had standard deviations of greater than 2.5 or less than − 2.5. The AOD model constructed based on GWR has three grids with large errors, Fig. 7.
As can be seen from Fig. 7, the types of ground cover within the seven error-prone grids are complex, i.e., a large error exists in the model constructed based on GWR under complex surface conditions. The reasons for this are as follows. The complex surface conditions were prone to lower remote sensing inversion AOD accuracies, leading to large estimation errors. Second, for the complex surface conditions, the calculation accuracy of the road network characteristic parameters was lower.

Spatial-temporal deduction of the remote sensing data
Although remote sensing data can retrieve the spatial distribution characteristics of the AOD (air pollution) well, it is instantaneous data, that is, it can only represent the spatial distribution of the satellite transit time. However, the AOD and tra c exhaust emissions are always changing, so the AOD retrieved by remote sensing has the problem of time scale deduction. To ensure the correspondence of the time scale, the observation data corresponding to the response time of the satellite transit time are generally used for research. In order to reduce the time scale error, in this study, the correlation analysis between the PCDI and the air quality was based on the average value. Therefore, the GWR analysis was conducted using the daily mean value of the AOD and the road network density, road area occupancy rate, intersection number, and bus network density.
The AOD data from MODIS generally contain data for more than three orbits each day (each orbit time is different). The AOD data are commonly missing on days with heavy pollution or very little pollution, e.g., from January 1 to 10, 2017, the air pollution in study area was serious, and the MCD19A2AOD in Jinan is all no data. Thus, the algorithm of the AOD needs to be further improved.
The correlations between the PCDI and the AOD were positive, that is, the more congested the tra c is, the stronger the relationship between the PCDI and air pollution is. Thus, the results of this study provide a signi cant reference for tra c planning and air quality control in congested areas.

Scale problem
In addition to spatial heterogeneity, spatial scale is also an important feature of geographical phenomena. Fotheringham et al. developed the multi-scale geographically weighted regression (MGWR) model to study the in uence of different scales on air quality. To reduce the impact of the different scales on the analysis, in this study, the gridded study area (i.e., the 1 km grid) was selected to match the 1 km resolution the AOD product. The difference index of the 30 m resolution can better retrieve the relationships between the roads, road network, and air quality, but the difference index of the 30 m resolution lacks a strict theoretical basis, that is, a higher classi cation of the remote sensing products is a direction for future research.

Conclusions
Based on the AOD retrieved from remote the sensing data and GWR models, in this study, the impacts of four road network tra c characteristic parameters on air quality were the rst quantitatively analyzed, including the road network density, road area occupancy, intersection number, and bus network density.
The main research conclusions are as follows. There is a strong positive correlation between the PCDI and air quality. Based on the GWR model, AOD has high correlations with the road network density, road area occupancy, intersection number, and bus network density, and these correlations are much higher than ordinary linear regression, that is, GWR re nes the local spatial changes in AOD and the road tra c parameters. The regression parameters of the four variables have positive and negative effects, that is, the in uence of each factor on air quality in the different regions can promote or inhibit, but positive correlations were dominant. The correlation between AOD and road area occupancy was the highest, and the correlations between the AOD and the density of the bus network and the number of intersections were higher than that between the AOD and the road network density, that is, bus route planning, bus emission reduction, road network planning, and signal timing (intersection) have greater impacts on air quality. The accuracy of the AOD models based on GWR was low under the conditions presented in the table. This study has certain guiding signi cance for tra c planning and tra c control, and provides support and basis for tra c planning and control, especially in areas with tra c jam.

Declarations
Author Contributions: Haixia Feng was involved in the research design and nalized the paper; Erwei Ning was involved in software and validation, Jian Li and Haiying Feng analyzed the data; Qi Wang gave useful comments which improved the paper.
Funding: Gridding of the study area