- Email: [email protected]
Using geographical semi-variogram method to quantify the difference between NO2 and PM2.5 spatial distribution characteristics in urban areas
Science of the Total Environment 631–632 (2018) 688–694
Contents lists available at ScienceDirect
Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv
Using geographical semi-variogram method to quantify the difference between NO2 and PM2.5 spatial distribution characteristics in urban areas Weize Song a, Haifeng Jia a,⁎, Zhilin Li b,⁎, Deliang Tang c a b c
School of Environment, Tsinghua University, Beijing 100084, China Department of Land Surveying and Geo-Informatics, Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong Mailman school of Public Health, Columbia University in the City of New York, New York, USA
H I G H L I G H T S
G R A P H I C A L
A B S T R A C T
• Spatial variations in NO2 and PM2.5 in Foshan were assessed by semi-variogram. • The local-scale spatial variance of PM2.5 is smaller than that of NO2. • The spatial range of NO2 autocorrelation is larger than that of PM2.5 • The NO2 and PM2.5 influencing factors have different spatial scale dependence. • The study provides scientific evidence for buffering selection of LUR predictors.
a r t i c l e
i n f o
Article history: Received 11 December 2017 Received in revised form 4 March 2018 Accepted 5 March 2018 Available online xxxx Editor: Jianmin Chen Keywords: Air pollution Semi-variogram Spatial variation Seasonal difference Spatial autocorrelation Spatial scale dependence
a b s t r a c t Urban air pollutant distribution is a concern in environmental and health studies. Particularly, the spatial distribution of NO2 and PM2.5, which represent photochemical smog and haze pollution in urban areas, is of concern. This paper presents a study quantifying the seasonal differences between urban NO2 and PM2.5 distributions in Foshan, China. A geographical semi-variogram analysis was conducted to delineate the spatial variation in daily NO2 and PM2.5 concentrations. The data were collected from 38 sites in the government-operated monitoring network. The results showed that the total spatial variance of NO2 is 38.5% higher than that of PM2.5. The random spatial variance of NO2 was 1.6 times than that of PM2.5. The nugget effect (i.e., random to total spatial variance ratio) values of NO2 and PM2.5 were 29.7 and 20.9%, respectively. This indicates that urban NO2 distribution was affected by both local and regional influencing factors, while urban PM2.5 distribution was dominated by regional influencing factors. NO2 had a larger seasonally averaged spatial autocorrelation distance (48 km) than that of PM2.5 (33 km). The spatial range of NO2 autocorrelation was larger in winter than the other seasons, and PM2.5 has a smaller range of spatial autocorrelation in winter than the other seasons. Overall, the geographical semi-variogram analysis is a very effective method to enrich the understanding of NO2 and PM2.5 distributions. It can provide scientific evidences for the buffering radius selection of spatial predictors for land use regression models. It will also be beneficial for developing the targeted policies and measures to reduce NO2 and PM2.5 pollution levels. © 2018 Elsevier B.V. All rights reserved.
⁎ Corresponding authors. E-mail addresses: [email protected] (W. Song), [email protected] (H. Jia), [email protected] (Z. Li), [email protected] (D. Tang).
https://doi.org/10.1016/j.scitotenv.2018.03.040 0048-9697/© 2018 Elsevier B.V. All rights reserved.
W. Song et al. / Science of the Total Environment 631–632 (2018) 688–694
689
1. Introduction Urban air pollutant distribution is a concern in environmental and health studies. Particularly, the spatial distribution of NO2 and PM2.5, which represent photochemical smog and haze pollution in urban areas, is of concern. NO2 and PM2.5 experience totally different atmospheric processes from emissions, transport, chemical reactions, and deposition (Chen et al., 2018; Jiang and Christakos, 2018; Liu et al., 2016; Strak et al., 2017; Wang et al., 2014b). There have been a number of studies on NO2 and PM2.5 distribution. These studies mainly focus on source apportionment, concentration mapping, formation and transport mechanisms, the effects of meteorological elements on distribution, and interaction analysis between PM2.5, NO2 and other air pollutants (Callen et al., 2014; Hoek et al., 2008; Liao et al., 2017; Song et al., 2014; Wu et al., 2017b). However, whether NO2 and PM2.5 have different spatial distribution characteristics has not been addressed in the previous literature. On the other hand, the geographical semi-variogram has been widely used for the development of kriging interpolation (KI) model (Badaro-Saliba et al., 2014; Cao et al., 2017; Qu et al., 2010). The geographical semi-variogram has been used for assessing the spatial variation of heavy metal pollution in soils, groundwater level, as well as optimization of air quality monitoring network (Ahmadi and Sedghamiz, 2007; Guo et al., 2001; Lin, 2002; Liu et al., 2013; Pahlavani et al., 2017; Wang et al., 2014a). However, it has not been used to quantify the difference between NO2 and PM2.5 spatial variations, or to analyze the spatial scale-dependence of influencing factors. Here, we considered that the spatial variation in air pollutant concentrations could be split into different components related to multiple spatial scales (Alary and Demougeot-Renard, 2010; Lv et al., 2014; Spokas et al., 2003). Then, the geographical semi-variogram was developed based on daily measurements. This may be a simple and effective way to analyze the spatial distribution characteristics of NO2 and PM2.5 in a complex urban environment. The geographical semi-variogram parameters can also quantify the multi-scale spatial variation, spatial autocorrelation distance, as well as the spatial scale dependence of influencing factors in NO2 and PM2.5 concentrations. Overall, the aim of this study is to compare the differences between NO2 and PM2.5 spatial distribution characteristics. The specific objectives are (1) identify the differences between NO2 and PM2.5 variation; (2) identify the difference in the spatial scale-dependence of NO2 and PM2.5 dominant influencing factors; (3) identify the differences in spatial range of NO2 and PM2.5 autocorrelation. 2. Materials and methods 2.1. Study area characteristics The study was conducted in Foshan, China. The geographical extent is 22°38′–23°34′N, and 112°22′–113°23′E. The total area is 3848.48 km2. The study area is located south of the Tropic of Cancer. The annual mean temperature, relative humidity, and wind speed are 22.5 °C, 76%, and 2 m/s, respectively. The rainy season is from April to September during which 80% of the total annual rainfall is received. Winds are mainly northerly in winter and spring, and mainly southerly in summer. The southwest and northwest areas are mountainous (up to 785 m), while the remaining area is generally flat. Fig. 1 shows the locations of government-operated NO2 and PM2.5 monitoring stations. 2.2. Air quality measurements Daily NO2 and PM2.5 measurements were collected from the Foshan air quality monitoring network (http://www.foshanepb.gov.cn/). The network consists of 38 fixed government-operated monitoring stations with simultaneous measurements, which have strict quality assurance and control procedures. The minimum and maximum distance between monitoring stations are 484.2 and 21,993.7 m, respectively. The
Fig. 1. Location of NO2 and PM2.5 monitoring stations.
averaged distance between all monitoring stations is 8117.5 m. The lowest and highest altitude of the monitoring stations is 28 and 58 m respectively. NO2 and PM2.5 were measured using the Ogawa badges method and the tapered element oscillating microbalance (TEOM) method, respectively (Li et al., 2017; Liu et al., 2017; Tessum et al., 2018; Wang et al., 2013). The measurements met the corresponding Ogawa and TEOM analysis protocol, and passed the criteria of the quality assurance and controls according to the environmental protection standard of China (HJ 618-2011; MEPCN) (Xie et al., 2015). 2.3. Geographical semi-variogram analysis The geographical semi-variogram model is a geostatistical analysis method, and is the most common measure for characterizing spatial variability of a regionalized variable (Lin et al., 2018). It is used here to understand the multi-scale spatial variation of urban air pollutants. The semi-variogram model can describe air pollutant variations as in graphical form, and a schematic diagram is presented in Fig. 2. Three major parameters, sill, partial sill, and nugget, are used to quantify the spatial variability of an air pollutant. The nugget parameter (C0) represents the random spatial variance of the air pollutant. The partial sill parameter (C1) represents the structural spatial variance of the air pollutant. A high partial sill means that a large proportion of the spatial variation is caused by regional-scale influencing factors. The sill parameter (C0 + C1) represents the total degree of spatial variation of the air pollutant. Additionally, h indicates the spatial distance between monitoring stations. The range parameter (α) represents the maximum spatial distance of air pollutant autocorrelation, because air pollutant is spatially autocorrelated. Samples separated by distances less than the range are spatially related, whereas those separated by a distance greater than the range are considered not to be spatially related. In short, the semi-variogram parameters effectively depict the spatial structure of air pollutant variations and enrich the understanding of air pollutant distribution. A full discussion of the semi-variogram parameters can be found in the literature (e.g. Burgos et al., 2006; Hu and Xu, 2018; Lin et al., 2018; Ye et al., 2018). The nugget effect (the ratio of random to total spatial variance) indicates whether regional or local-scale factors are more important for air pollutant distribution. The value of nugget effect ranges from 0 to 100%,
690
W. Song et al. / Science of the Total Environment 631–632 (2018) 688–694
Fig. 2. Schematic diagram of the geographical semi-variogram analysis.
representing strong to weak spatial dependence of the air pollutant. If the nugget effect value is less than 25%, regional-scale influencing factors are more important for the air pollutant distribution; if the ratio is between 25% and 75%, then both regional- and local-scale influencing factors are important for the air pollutant distribution; and if the ratio is greater than 75%, then the local-scale influencing factors are more important for the air pollutant distribution (Liu et al., 2004). However, the nugget effect cannot identify the sources affecting air pollutant distribution. In this case study, we developed the NO2 and PM2.5 semi-variogram models based on daily measurements. The specific parameters of the semi-variogram models were calculated using the geostatistical analyst tool in ArcGIS 10.3. Then, we averaged the daily parameters for each season. Finally, the seasonal fluctuations in the geographical semi-variogram parameters were analysed, to compare the differences between NO2 and PM2.5 spatial distribution characteristics within the study domain. 3. Results 3.1. Descriptive statistics of NO2 and PM2.5 concentrations The spatial difference of NO2 and PM2.5 concentrations was presented in Fig. 3. For the case study city, the annual mean NO2 and
PM2.5 concentrations were 49.5 and 51.8 μg/m3 in 2014, respectively. On average, they exceeded the national standards by more than 65 and 48%, respectively. The station-based annual mean concentrations were highly variable ranging from 31.3 to 64.7 μg/m3 for NO2 and from 40.0 μg/m3 to 63.3 μg/m3 for PM2.5. There were 37 and 38 stations whose annual mean values exceeded the respective national standard. The NO2 and PM2.5 measurements showed a large difference across the study area, and their spatial distributions were quite different. For NO2, the smallest annual mean value appeared in the Gaoming district at 31.3 μg/m3. The largest annual mean value appeared in the Chancheng district at 64.7 μg/m3. For PM2.5, the largest and smallest annual mean values appeared in the Nanhai (63.3 μg/m3) and Shunde (40.0 μg/m3) districts. These results indicat that the study area is suffering from serious air pollution. They also show that the central monitoring stations have higher NO2 concentration levels than the stations on either side of the study area. In contrast, PM2.5 has higher concentration levels in the north-western part of the study area. Given the elevation difference (shown in Fig. 1), we can infer that the terrain has a significant impact on the PM2.5 distribution. Descriptive statistics of the NO2 and PM2.5 measurements are provided in Table 1. The averaged NO2, and PM2.5 concentrations in winter were significantly higher than in summer, showing a significant seasonal variation. The results indicated that the spatial distribution of NO2 and PM2.5 pollution also have a large seasonal difference. Fig. 4 shows the seasonal difference in NO2 and PM2.5 measurements. There were larger differences in site-based mean values in spring and winter. Specifically, the spatial variation in PM2.5 concentrations was largest in winter, ranging from 54.8 to 92.8 μg/m3. Significant increases in NO2 and PM2.5 pollution levels were also observed in winter.
3.2. Geographical semi-variogram models of NO2 and PM2.5 The results of the NO2 and PM2.5 semi-variogram models are summarised in Fig. 5, and the annual and seasonal mean values of these parameters were presented in Table 2. The results show the seasonal differences in the structural, random, and total spatial variance in NO2 and PM2.5, respectively. Visually, it is easy to see the difference between the NO2 and PM2.5 spatial distributions and seasonal mean variances. The degree of spatial variation for NO2 was larger than that of PM2.5. The annual mean nugget value of NO2 was also larger than that
Fig. 3. Spatial differences between NO2 and PM2.5 concentrations.
W. Song et al. / Science of the Total Environment 631–632 (2018) 688–694 Table 1 Descriptive statistics for NO2 and PM2.5 measurements. Air pollutants Indicators
Spring Summer Autumn Winter Annual
NO2
36.4 55.6 81.4 9.3
19.3 41.7 57.3 7.6
32.3 50.0 62.2 6.5
42.4 64.8 85.1 10.2
31.3 52.0 79.4 7.8
0.3 0.6 33.1 43.7 58.3 6.2
−0.2 1.2 22.7 31.5 45.7 6.5
−0.2 0.3 39.8 53.6 66.3 7.1
−0.4 0.2 54.8 71.4 106.4 10.9
0.8 4 40 51.1 106.4 11.3
0.7 −0.2
0.6 −0.5
−0.2 −1.1
1.1 1.7
3.5 16.8
PM2.5
Minimum Mean Maximum Standard deviation Skewness Kurtosis Minimum Mean Maximum Standard deviation Skewness Kurtosis
of PM2.5, reflecting that the spatial variance was larger for NO2 than PM2.5 at the local scale. The sill value of NO2 was largest and smallest in winter and summer, meaning the spatial variance of NO2 was larger in winter than in summer. The seasonal averaged sill value of PM2.5 in winter was 2.1 times higher than that in summer. Overall, the spatial variation for both NO2 and PM2.5 was large in winter and small in summer. The maximum spatial distances of NO2 autocorrelation in spring, summer, autumn and winter were 42, 43, 54 and 53 km, respectively. Several authors have reported spatial distances of 37–58 km for NO2 in Beijing from November to December 2012 (Li et al., 2015). The spatial range of NO2 autocorrelation is smaller in summer and larger in winter. The possible reason for this difference is that NO2 has a longer lifespan in winter, and can achieve longer transport distances. The maximum spatial distance of NO2 autocorrelation in autumn was 28.5% higher than in spring. Spring and summer had similar spatial autocorrelation ranges, and autumn and winter had similar spatial autocorrelation range values. The maximum spatial distances of PM2.5 autocorrelation in spring, summer, autumn and winter were 35, 36, 33, and 20 km, respectively. In contrast with the NO2 results, the range values of PM2.5 spatial autocorrelation were larger in summer than in winter. The maximum spatial distance of PM2.5 autocorrelation reached its peak value (48.1 km) in May, and the smallest spatial distance was only 19.9 km in November. The spatial range of PM2.5 autocorrelation in spring and summer is larger than that in autumn and winter. These results mean that the spatial distribution of PM2.5 is more complex in spring and summer. It can easily be observed that NO2 and PM2.5 has a different spatial autocorrelation range in different seasons. The spatial range of NO2 autocorrelation was larger than that of PM2.5. In summer and autumn, the maximum spatial distance of NO2 autocorrelation was 40 km, whereas it was approximately 33 km for PM2.5. These results indicate that
691
PM2.5 has a smaller spatial overflow effect than NO2 at the local scale. NO2 shows a larger variation in spatial autocorrelation range than PM2.5 in autumn and winter, indicating that the spatiotemporal distribution of NO2 is more complex than that of PM2.5. In contrast, the spatiotemporal distribution of PM2.5 is more complex in summer. In spring, the proportion of NO2 and PM2.5 variance caused by small-small influencing factors were smaller than in other seasons. In winter, the spatial range of NO2 autocorrelation was the largest, and the spatial range of PM2.5 autocorrelation was the smallest. As shown in Fig. 5, the seasonal fluctuations in the nugget effect were also significant for NO2 and PM2.5. For NO2, the nugget effect (i.e. random to total spatial variance ratio) values were 18.4, 33.4, 34.9, and 32.2% in spring, summer, autumn, and winter, respectively. These results show that the spatial dependence of NO2 concentrations was moderate in summer, autumn, and winter, with strong spatial dependence in spring. Thus, the spatial dependence of NO2 was larger in spring than in other seasons. For PM2.5, the nugget effect values were 18.5, 20.3, 21.6, and 23.2% in spring, summer, autumn, and winter, respectively. These results mean that the spatial dependence of PM2.5 was strong in all seasons. The annual mean nugget effect values of NO2 and PM2.5 were 29.7 and 20.9%, respectively. These results indicate that the spatial distribution of NO2 concentrations was affected by both regional and local influencing factors, while the spatial distribution of PM2.5 concentrations is mainly influenced by regional influencing factors such as weather and terrain. For PM2.5, there were 245 days (i.e. 67.1% of the days in 2014) when the nugget effect values were less than 25%. In contrast, there were 178 days (48.8%) when the nugget effect values of NO2 were less than 25%. Overall, the spatial distribution of PM2.5 concentrations is dominated by large-scale influencing factors on most days, and the spatial distribution of NO2 had less continuity and more variance than that of PM2.5. 4. Discussion 4.1. Spatial variation of NO2 and PM2.5 concentrations The total, structural, and random spatial variance of NO2 in 2014 was larger than that of PM2.5. This result indicates that the degree of spatial variation of NO2 is larger than that of PM2.5 in the study area. The structural variances of pollutants were caused by regional influencing factors, including meteorological elements, long-range transport and regionalrelated contributions. The random spatial variance of NO2 and PM2.5 were caused by short-range variation and sampling errors. The spatial variance of NO2 in winter is larger than in summer. This result was expected due to its emission sources and intensity; more emissions occur in winter in the study region. The main sources of NO2 include the combustion of fossil fuels and biofuels, biomass burning, lightning, and bacterially mediated decomposition of nitrogen processed through the agriculture/animal/human food chain (Wang et al., 2007). The
Fig. 4. Seasonal variation in NO2 and PM2.5 measurements.
692
W. Song et al. / Science of the Total Environment 631–632 (2018) 688–694
Fig. 5. Seasonal differences of NO2 and PM2.5 distribution.
structural spatial variance of NO2 reflects the spatial variation at the regional scale and is caused by large-scale influencing factors, such as meteorology, external transport, light-catalysed reactions, and terrain. The random variances in NO2 are caused by local-scale influencing factors, such as industry point emissions within 485 m (the shortest distance to any monitoring station). The spatial variance of PM2.5 was larger in winter than in summer. The larger value for random spatial variance indicates that there are more local sources of PM2.5 in winter. The larger regional-scale spatial variances show the effects of regional influencing factors on background
concentrations. The regional influencing factors include long-range transport and photochemical formation of particle organic compounds. Thus, the values can be attributed to inherent PM2.5 sources (such as soil 2− − dust, NO− 3 , SO4 , NH4 ) and to meteorological factors, long-distance transport and photocatalytic reactions. 4.2. Spatial scale dependence of NO2 dominant factors For NO2, higher nugget effect values were observed in summer, autumn, and winter, whereas lower values were observed in spring. This
W. Song et al. / Science of the Total Environment 631–632 (2018) 688–694 Table 2 Seasonally-averaged results of the NO2 and PM2.5 semi-variogram models. Air pollutants
Parameters
Spring
Summer
Autumn
Winter
Annual
NO2
Sill Partial sill Nugget Range Nugget effect Sill Partial sill Nugget Range Nugget effect
278.1 230.2 47.9 41.9 18.4% 167.2 148.7 18.6 34.5 18.5%
191.5 144.8 46.8 42.8 33.4% 93.2 79.2 14.1 35.8 20.3%
227.5 176.3 51.2 53.7 34.9% 170.0 143.7 26.3 32.0 21.6%
359.4 276.7 82.6 53.8 32.2% 240.1 212.3 27.7 28.8 23.2%
264.1 207.0 57.1 48.0 29.7% 167.6 146.0 21.7 32.8 20.9%
PM2.5
result means that the regional influencing factors had the largest impacts on the spatial variation of NO2 in spring. In addition, the local influencing factors explained the largest fraction of the total spatial variance in autumn. This result means that local emissions (e.g. traffic and power plants) had a strong influence on the NO2 distribution. Overall, the spatial distribution of NO2 is affected by both regional and local factors, and more variance is explained by the regional factors. These results agree with previous land-use regression (LUR) studies (Beelen et al., 2013; Cordioli et al., 2017; Lee et al., 2014; Liu et al., 2015; Tang et al., 2013; Weissert et al., 2018). For example, Hoek et al. (2008) summarised the spatial predictors often selected for final NO2 LUR models and found that the road length with 350 m buffers, traffic volume with 300 m buffers, traffic intensity with 50–250 m buffers, and distance to the nearest roads often appeared as the local-scale spatial predictors. Altitude, land cover, and populations with 5000 m buffers appeared as the regional-scale spatial predictors. Thus, we can infer that traffic, roads, power plants, and land cover are the local-scale influencing factors of NO2 distribution. The altitude (as shown in Fig. 1) and meteorological conditions are the regional-scale influencing factors on NO2 concentrations. The seasonal difference in NO2 variations was caused by the varying effects of the dominant influencing factors. The combination of geographical semi-variation method and Pearson correlation method may be benefit for the potential air pollution source identification or apportionment. 4.3. Spatial scale dependence of PM2.5 dominant factors The nugget effect values of PM2.5 were smaller than that of NO2. This means that the spatial dependence of PM2.5 was stronger than that of NO2. This difference in spatial dependence can be attributed to the NO2 and PM2.5 properties. The former has a shorter life cycle, and the latter has a longer life cycle. This means that regional influencing factors, both natural and anthropogenic, have a larger influence on PM2.5. The spatial predictors that are often selected for the final PM2.5 LUR models include building density with 2500 m buffers, industrial land use with 5000 m buffers, populations with 1000–5000 m buffers, and length of expressway with 1000 m buffers. The regional influencing factors also include the contribution of long-distance transport to PM2.5 variation. Although the nugget effect values can reflect whether regional or local influencing factors dominate the spatial distribution of PM2.5 concentrations, they do not differentiate between natural and anthropogenic influencing factors. The small nugget effect shows that the spatial distribution of PM2.5 concentrations is controlled by regional influencing factors, not local influencing factors. Therefore, soil dust has limited impact on the spatial distribution of PM2.5 in the case study city. This result agrees with previous land-use regression (LUR) studies (Eeftens et al., 2012; Kloog et al., 2012; Lee et al., 2016; Wu et al., 2017a). NO2 is a gaseous pollutant, whose lifespan is short and often varies at the local scale. PM2.5 is a particulate pollutant, with a longer lifespan, and can often be transported at a larger regional scale. Thus, largescale meteorological elements play a more important role in urban PM2.5 distribution, and their effects must be considered when we
693
consider air quality and its influencing factors. For example, light winds, elevated temperatures and surface inversions, and low-mixing heights can significant enhance secondary PM2.5 formation (Wang et al., 2014b). 4.4. Spatial range of NO2 and PM2.5 autocorrelation The seasonal difference in NO2 spatial autocorrelation range can be attributed to the combined effects of climate and emission sources. The large fluctuations in PM2.5 spatial autocorrelation range occurred in summer due to the impact of rainfall and photocatalytic reactions. When rainfall occurs, the PM2.5 spatial autocorrelation range is small; when the meteorological conditions are beneficial to photocatalytic reactions, the PM2.5 spatial autocorrelation range is large. PM2.5 has a smaller spatial autocorrelation range than NO 2 , reflecting different spatial autocorrelation properties. The primary reason is that NO2 and PM2.5 are gaseous and particulate pollutants, respectively. The values of spatial autocorrelation range can act as a guide to design denser sampling plans for the case study city in future. The sampling interval should be smaller than the range of spatial autocorrelation. For instance, it could be a quarter or half of the spatial autocorrelation range (Kerry and Oliver, 2003). To better characterize the urban NO2 and PM2.5 distributions, the sampling interval of NO2 should be 10.5–21 km, 10.8–21.5 km, and 13.5–27 km, 13.3–26.5 km in spring, summer, autumn, and winter, respectively. The PM2.5 sampling interval should be 8.8–17.5 km, 9–18 km, 8.3–16.5 km, and 7.3–14.5 km in spring, summer, autumn, and winter, respectively. In brief, the NO2 sampling interval should be 10.5 to 27 km. The PM2.5 sampling interval should be 7.25 to 18 km. The NO2 sampling interval in spring and summer should be smaller than in autumn and winter. The PM2.5 sampling interval in autumn and winter should be closer than in spring and summer. 5. Conclusion In conclusion, these results provide a better insight into the urban NO2 and PM2.5 distribution. The spatial variance of NO2 was larger than that of PM2.5. There were significant seasonal differences in both NO2 and PM2.5 variations. The spatial variance of NO2 and PM2.5 concentrations were larger in winter and smaller in summer. NO2 showed a moderate spatial dependence and PM2.5 a large spatial dependence. The spatial distribution of NO2 is affected by both local and regional influencing factors. In contrast, the spatial distribution of PM2.5 is dominated by regional influencing factors. The maximum spatial distance of NO2 autocorrelation was largest in winter. The spatial autocorrelation of PM2.5 was largest in summer. As a result, the sampling interval of NO2 should be smaller in summer than in winter, and for PM2.5 the sampling interval should be larger in summer than in winter. In short, the sampling interval of NO2 and PM2.5 should be different, and varying depending on the season. Overall, the geographical semi-variogram method is a very effective approach to enrich the understanding of the differences between NO2 and PM2.5 spatial distribution. It can provide scientific evidence for the buffering radius selection of spatial predictors for LUR models. It will also be beneficial for developing targeted policies and measures to reduce ambient NO2 and PM2.5 pollution levels. In future, the method should be used applied in more cities and for additional air pollutants to verify its reliability. Acknowledgements This study was carried out at the School of Environment, Tsinghua University. We thank the Research Institute for Sustainable Urban Development, at the Hong Kong Polytechnic University for its fellowship support.
694
W. Song et al. / Science of the Total Environment 631–632 (2018) 688–694
References Ahmadi, S.H., Sedghamiz, A., 2007. Geostatistical analysis of spatial and temporal variations of groundwater level. Environ. Monit. Assess. 129, 277–294. Alary, C., Demougeot-Renard, H., 2010. Factorial kriging analysis as a tool for explaining the complex spatial distribution of metals in sediments. Environ. Sci. Technol. 44, 593–599. Badaro-Saliba, N., Adjizian-Gerard, J., Zaarour, R., Abboud, M., Farah, W., Saliba, A.N., et al., 2014. A geostatistical approach for assessing population exposure to NO2 in a complex urban area (Beirut, Lebanon). Stoch. Env. Res. Risk A. 28, 467–474. Beelen, R., Hoek, G., Vienneau, D., Eeftens, M., Dimakopoulou, K., Pedeli, X., et al., 2013. Development of NO2 and NOX land use regression models for estimating air pollution exposure in 36 study areas in Europe – the ESCAPE project. Atmos. Environ. 72, 10–23. Burgos, P., Madejon, E., Perez-De-Mora, A., Cabrera, F., 2006. Spatial variability of the chemical characteristics of a trace-element-contaminated soil before and after remediation. Geoderma 130, 157–175. Callen, M.S., Iturmendi, A., Lopez, J.M., 2014. Source apportionment of atmospheric PM2.5bound polycyclic aromatic hydrocarbons by a PMF receptor model. Assessment of potential risk for human health. Environ. Pollut. 195, 167–177. Cao, S., Lu, A., Wang, J., Huo, L., 2017. Modeling and mapping of cadmium in soils based on qualitative and quantitative auxiliary variables in a cadmium contaminated area. Sci. Total Environ. 580, 430–439. Chen, C., Zhu, P.F., Lan, L., Zhou, L., Liu, R.C., Sun, Q.H., et al., 2018. Short-term exposures to PM2.5 and cause-specific mortality of cardiovascular health in China. Environ. Res. 161, 188–194. Cordioli, M., Pironi, C., De Munari, E., Marmiroli, N., Lauriola, P., Ranzi, A., 2017. Combining land use regression models and fixed site monitoring to reconstruct spatiotemporal variability of NO2 concentrations over a wide geographical area. Sci. Total Environ. 574, 1075–1084. Eeftens, M., Beelen, R., de Hoogh, K., Bellander, T., Cesaroni, G., Cirach, M., et al., 2012. Development of land use regression models for PM2.5, PM2.5 absorbance, PM10 and PM coarse in 20 European study areas; results of the ESCAPE project. Environ. Sci. Technol. 46, 11195–11205. Guo, X.D., Fu, B.J., Ma, K.M., Chen, L.D., 2001. Utility of semivariogram for spatial variation of soil nutrients and the robust analysis of semivariogram. J. Environ. Sci. (China) 13, 453–458. Hoek, G., Beelen, R., de Hoogh, K., Vienneau, D., Gulliver, J., Fischer, P., et al., 2008. A review of land-use regression models to assess spatial variation of outdoor air pollution. Atmos. Environ. 42, 7561–7578. Hu, X., Xu, H., 2018. A new remote sensing index for assessing the spatial heterogeneity in urban ecological quality: a case from Fuzhou city, China. Ecol. Indic. 89, 11–21. Jiang, Q.T., Christakos, G., 2018. Space-time mapping of ground-level PM2.5 and NO2 concentrations in heavily polluted northern China during winter using the Bayesian maximum entropy technique with satellite data. Air Qual. Atmos. Health 11, 23–33. Kerry, R., Oliver, M.A., 2003. Variograms of ancillary data to aid sampling for soil surveys. Precis. Agric. 4, 261–278. Kloog, I., Nordio, F., Coull, B.A., Schwartz, J., 2012. Incorporating local land use regression and satellite aerosol optical depth in a hybrid model of spatiotemporal PM2.5 exposures in the mid-Atlantic states. Environ. Sci. Technol. 46, 11913–11921. Lee, J.-H., Wu, C.-F., Hoek, G., de Hoogh, K., Beelen, R., Brunekreef, B., et al., 2014. Land use regression models for estimating individual NOX and NO2 exposures in a metropolis with a high density of traffic roads and population. Sci. Total Environ. 472, 1163–1171. Lee, H.J., Chatfield, R.B., Strawa, A.W., 2016. Enhancing the applicability of satellite remote sensing for PM2.5 estimation using modis deep blue aod and land use regression in California, United States. Environ. Sci. Technol. 50, 6546–6555. Li, Z., Yang, H., Jiang, Y., Yin, Y., Wang, H., Wang, B., et al., 2015. The study on spatial distribution of atmospheric contaminant concentrations and optimization of air quality monitoring sites in Beijing city. Environ. Monit. China 31, 74–78. Li, L., Lurmann, F., Habre, R., Urman, R., Rappaport, E., Ritz, B., et al., 2017. Constrained mixed-effect models with ensemble learning for prediction of nitrogen oxides concentrations at high spatiotemporal resolution. Environ. Sci. Technol. 51, 9920–9929. Liao, H.T., Chou, C.C.K., Huang, S.H., Lu, C.J., Chen, C.C., Hopke, P.K., et al., 2017. Source apportionment of PM2.5 size distribution and composition data from multiple stationary sites using a mobile platform. Atmos. Res. 190, 21–28. Lin, Y.-P., 2002. Multivariate geostatistical methods to identify and map spatial variations of soil heavy metals. Environ. Geol. 42, 1–10. Lin, Z., Chao, L., Wu, C., Hong, W., Hong, T., Hu, X., 2018. Spatial analysis of carbon storage density of mid-subtropical forests using geostatistics: a case study in Jiangle county, Southeast China. Acta Geochim. 37, 90–101.
Liu, X.M., Xu, J.M., Zhang, M.K., Huang, J.H., Shi, J.C., Yu, X.F., 2004. Application of geostatistics and GIS technique to characterize spatial variabilities of bioavailable micronutrients in paddy soils. Environ. Geol. 46, 189–194. Liu, Y., Lv, J., Zhang, B., Bi, J., 2013. Spatial multi-scale variability of soil nutrients in relation to environmental factors in a typical agricultural region, eastern China. Sci. Total Environ. 450-451, 108–119. Liu, W., Li, X., Chen, Z., Zeng, G., León, T., Liang, J., et al., 2015. Land use regression models coupled with meteorology to model spatial and temporal variability of NO2 and PM10 in Changsha, China. Atmos. Environ. 116, 272–280. Liu, C., Henderson, B.H., Wang, D.F., Yang, X.Y., Peng, Z.R., 2016. A land use regression application into assessing spatial variation of intra-urban fine particulate matter (PM2.5) and nitrogen dioxide (NO2) concentrations in city of Shanghai, China. Sci. Total Environ. 565, 607–615. Liu, M., Bi, J., Ma, Z., 2017. Visibility-based PM2.5 concentrations in China: 1957–1964 and 1973–2014. Environ. Sci. Technol. 51, 13161–13169. Lv, J., Liu, Y., Zhang, Z., Dai, B., 2014. Multivariate geostatistical analyses of heavy metals in soils: spatial multi-scale variations in Wulian, eastern China. Ecotoxicol. Environ. Saf. 107, 140–147. Pahlavani, P., Sheikhian, H., Bigdeli, B., 2017. Assessment of an air pollution monitoring network to generate urban air pollution maps using Shannon information index, fuzzy overlay, and Dempster-Shafer theory, a case study: Tehran, Iran. Atmos. Environ. 167, 254–269. Qu, W.J., Arimoto, R., Zhang, X.Y., Zhao, C.H., Wang, Y.Q., Sheng, L.F., et al., 2010. Spatial distribution and interannual variation of surface PM10 concentrations over eightysix Chinese cities. Atmos. Chem. Phys. 10, 5641–5662. Song, W., Jia, H., Huang, J., Zhang, Y., 2014. A satellite-based geographically weighted regression model for regional PM2.5 estimation over the Pearl River Delta region in China. Remote Sens. Environ. 154, 1–7. Spokas, K., Graff, C., Morcet, M., Aran, C., 2003. Implications of the spatial variability of landfill emission rates on geospatial analyses. Waste Manag. 23, 599–607. Strak, M., Janssen, N., Beelen, R., Schmitz, O., Vaartjes, I., Karssenberg, D., et al., 2017. Longterm exposure to particulate matter, NO2 and the oxidative potential of particulates and diabetes prevalence in a large national health survey. Environ. Int. 108, 228–236. Tang, R., Blangiardo, M., Gulliver, J., 2013. Using building heights and street configuration to enhance intraurban PM10, NOX, and NO2 land use regression models. Environ. Sci. Technol. 47, 11643–11650. Tessum, M.W., Larson, T., Gould, T.R., Simpson, C.D., Yost, M.G., Vedal, S., 2018. Mobile and fixed-site measurements to identify spatial distributions of traffic-related pollution sources in Los Angeles. Environ. Sci. Technol. 52 (5), 2844–2853. Wang, Y.X., McElroy, M.B., Martin, R.V., Streets, D.G., Zhang, Q., Fu, T.M., 2007. Seasonal variability of NOX emissions over east China constrained by satellite observations: implications for combustion and microbial sources. J. Geophys. Res.-Atmos. 112. Wang, M., Beelen, R., Basagana, X., Becker, T., Cesaroni, G., de Hoogh, K., et al., 2013. Evaluation of land use regression models for NO2 and particulate matter in 20 European study areas: the ESCAPE project. Environ. Sci. Technol. 47, 4357–4364. Wang, L., Coles, N.A., Wu, C.F., Wu, J.P., 2014a. Spatial variability of heavy metals in the coastal soils under long-term reclamation. Estuar. Coast. Shelf Sci. 151, 310–317. Wang, Y.G., Ying, Q., Hu, J.L., Zhang, H.L., 2014b. Spatial and temporal variations of six criteria air pollutants in 31 provincial capital cities in China during 2013–2014. Environ. Int. 73, 413–422. Weissert, L.F., Salmond, J.A., Miskell, G., Alavi-Shoshtari, M., Williams, D.E., 2018. Development of a microscale land use regression model for predicting NO2 concentrations at a heavy trafficked suburban area in Auckland, NZ. Sci. Total Environ. 619-620, 112–119. Wu, C.D., Chen, Y.C., Pan, W.C., Zeng, Y.T., Chen, M.J., Guo, Y.L., et al., 2017a. Land-use regression with long-term satellite-based greenness index and culture-specific sources to model PM2.5 spatial-temporal variability. Environ. Pollut. 224, 148–157. Wu, J.R., Li, G.H., Cao, J.J., Bei, N.F., Wang, Y.C., Feng, T., et al., 2017b. Contributions of transboundary transport to summertime air quality in Beijing, China. Atmos. Chem. Phys. 17, 2035–2051. Xie, Y., Wang, Y., Zhang, K., Dong, W., Lv, B., Bai, Y., 2015. Daily estimation of ground-level PM2.5 concentrations over Beijing using 3 km resolution MODIS AOD. Environ. Sci. Technol. 49, 12280–12288. Ye, L., Tan, W., Fang, L., Ji, L., Deng, H., 2018. Spatial analysis of soil aggregate stability in a small catchment of the loess plateau, China: I. Spatial variability. Soil Tillage Res. 179, 71–81.
Contents lists available at ScienceDirect
Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv
Using geographical semi-variogram method to quantify the difference between NO2 and PM2.5 spatial distribution characteristics in urban areas Weize Song a, Haifeng Jia a,⁎, Zhilin Li b,⁎, Deliang Tang c a b c
School of Environment, Tsinghua University, Beijing 100084, China Department of Land Surveying and Geo-Informatics, Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong Mailman school of Public Health, Columbia University in the City of New York, New York, USA
H I G H L I G H T S
G R A P H I C A L
A B S T R A C T
• Spatial variations in NO2 and PM2.5 in Foshan were assessed by semi-variogram. • The local-scale spatial variance of PM2.5 is smaller than that of NO2. • The spatial range of NO2 autocorrelation is larger than that of PM2.5 • The NO2 and PM2.5 influencing factors have different spatial scale dependence. • The study provides scientific evidence for buffering selection of LUR predictors.
a r t i c l e
i n f o
Article history: Received 11 December 2017 Received in revised form 4 March 2018 Accepted 5 March 2018 Available online xxxx Editor: Jianmin Chen Keywords: Air pollution Semi-variogram Spatial variation Seasonal difference Spatial autocorrelation Spatial scale dependence
a b s t r a c t Urban air pollutant distribution is a concern in environmental and health studies. Particularly, the spatial distribution of NO2 and PM2.5, which represent photochemical smog and haze pollution in urban areas, is of concern. This paper presents a study quantifying the seasonal differences between urban NO2 and PM2.5 distributions in Foshan, China. A geographical semi-variogram analysis was conducted to delineate the spatial variation in daily NO2 and PM2.5 concentrations. The data were collected from 38 sites in the government-operated monitoring network. The results showed that the total spatial variance of NO2 is 38.5% higher than that of PM2.5. The random spatial variance of NO2 was 1.6 times than that of PM2.5. The nugget effect (i.e., random to total spatial variance ratio) values of NO2 and PM2.5 were 29.7 and 20.9%, respectively. This indicates that urban NO2 distribution was affected by both local and regional influencing factors, while urban PM2.5 distribution was dominated by regional influencing factors. NO2 had a larger seasonally averaged spatial autocorrelation distance (48 km) than that of PM2.5 (33 km). The spatial range of NO2 autocorrelation was larger in winter than the other seasons, and PM2.5 has a smaller range of spatial autocorrelation in winter than the other seasons. Overall, the geographical semi-variogram analysis is a very effective method to enrich the understanding of NO2 and PM2.5 distributions. It can provide scientific evidences for the buffering radius selection of spatial predictors for land use regression models. It will also be beneficial for developing the targeted policies and measures to reduce NO2 and PM2.5 pollution levels. © 2018 Elsevier B.V. All rights reserved.
⁎ Corresponding authors. E-mail addresses: [email protected] (W. Song), [email protected] (H. Jia), [email protected] (Z. Li), [email protected] (D. Tang).
https://doi.org/10.1016/j.scitotenv.2018.03.040 0048-9697/© 2018 Elsevier B.V. All rights reserved.
W. Song et al. / Science of the Total Environment 631–632 (2018) 688–694
689
1. Introduction Urban air pollutant distribution is a concern in environmental and health studies. Particularly, the spatial distribution of NO2 and PM2.5, which represent photochemical smog and haze pollution in urban areas, is of concern. NO2 and PM2.5 experience totally different atmospheric processes from emissions, transport, chemical reactions, and deposition (Chen et al., 2018; Jiang and Christakos, 2018; Liu et al., 2016; Strak et al., 2017; Wang et al., 2014b). There have been a number of studies on NO2 and PM2.5 distribution. These studies mainly focus on source apportionment, concentration mapping, formation and transport mechanisms, the effects of meteorological elements on distribution, and interaction analysis between PM2.5, NO2 and other air pollutants (Callen et al., 2014; Hoek et al., 2008; Liao et al., 2017; Song et al., 2014; Wu et al., 2017b). However, whether NO2 and PM2.5 have different spatial distribution characteristics has not been addressed in the previous literature. On the other hand, the geographical semi-variogram has been widely used for the development of kriging interpolation (KI) model (Badaro-Saliba et al., 2014; Cao et al., 2017; Qu et al., 2010). The geographical semi-variogram has been used for assessing the spatial variation of heavy metal pollution in soils, groundwater level, as well as optimization of air quality monitoring network (Ahmadi and Sedghamiz, 2007; Guo et al., 2001; Lin, 2002; Liu et al., 2013; Pahlavani et al., 2017; Wang et al., 2014a). However, it has not been used to quantify the difference between NO2 and PM2.5 spatial variations, or to analyze the spatial scale-dependence of influencing factors. Here, we considered that the spatial variation in air pollutant concentrations could be split into different components related to multiple spatial scales (Alary and Demougeot-Renard, 2010; Lv et al., 2014; Spokas et al., 2003). Then, the geographical semi-variogram was developed based on daily measurements. This may be a simple and effective way to analyze the spatial distribution characteristics of NO2 and PM2.5 in a complex urban environment. The geographical semi-variogram parameters can also quantify the multi-scale spatial variation, spatial autocorrelation distance, as well as the spatial scale dependence of influencing factors in NO2 and PM2.5 concentrations. Overall, the aim of this study is to compare the differences between NO2 and PM2.5 spatial distribution characteristics. The specific objectives are (1) identify the differences between NO2 and PM2.5 variation; (2) identify the difference in the spatial scale-dependence of NO2 and PM2.5 dominant influencing factors; (3) identify the differences in spatial range of NO2 and PM2.5 autocorrelation. 2. Materials and methods 2.1. Study area characteristics The study was conducted in Foshan, China. The geographical extent is 22°38′–23°34′N, and 112°22′–113°23′E. The total area is 3848.48 km2. The study area is located south of the Tropic of Cancer. The annual mean temperature, relative humidity, and wind speed are 22.5 °C, 76%, and 2 m/s, respectively. The rainy season is from April to September during which 80% of the total annual rainfall is received. Winds are mainly northerly in winter and spring, and mainly southerly in summer. The southwest and northwest areas are mountainous (up to 785 m), while the remaining area is generally flat. Fig. 1 shows the locations of government-operated NO2 and PM2.5 monitoring stations. 2.2. Air quality measurements Daily NO2 and PM2.5 measurements were collected from the Foshan air quality monitoring network (http://www.foshanepb.gov.cn/). The network consists of 38 fixed government-operated monitoring stations with simultaneous measurements, which have strict quality assurance and control procedures. The minimum and maximum distance between monitoring stations are 484.2 and 21,993.7 m, respectively. The
Fig. 1. Location of NO2 and PM2.5 monitoring stations.
averaged distance between all monitoring stations is 8117.5 m. The lowest and highest altitude of the monitoring stations is 28 and 58 m respectively. NO2 and PM2.5 were measured using the Ogawa badges method and the tapered element oscillating microbalance (TEOM) method, respectively (Li et al., 2017; Liu et al., 2017; Tessum et al., 2018; Wang et al., 2013). The measurements met the corresponding Ogawa and TEOM analysis protocol, and passed the criteria of the quality assurance and controls according to the environmental protection standard of China (HJ 618-2011; MEPCN) (Xie et al., 2015). 2.3. Geographical semi-variogram analysis The geographical semi-variogram model is a geostatistical analysis method, and is the most common measure for characterizing spatial variability of a regionalized variable (Lin et al., 2018). It is used here to understand the multi-scale spatial variation of urban air pollutants. The semi-variogram model can describe air pollutant variations as in graphical form, and a schematic diagram is presented in Fig. 2. Three major parameters, sill, partial sill, and nugget, are used to quantify the spatial variability of an air pollutant. The nugget parameter (C0) represents the random spatial variance of the air pollutant. The partial sill parameter (C1) represents the structural spatial variance of the air pollutant. A high partial sill means that a large proportion of the spatial variation is caused by regional-scale influencing factors. The sill parameter (C0 + C1) represents the total degree of spatial variation of the air pollutant. Additionally, h indicates the spatial distance between monitoring stations. The range parameter (α) represents the maximum spatial distance of air pollutant autocorrelation, because air pollutant is spatially autocorrelated. Samples separated by distances less than the range are spatially related, whereas those separated by a distance greater than the range are considered not to be spatially related. In short, the semi-variogram parameters effectively depict the spatial structure of air pollutant variations and enrich the understanding of air pollutant distribution. A full discussion of the semi-variogram parameters can be found in the literature (e.g. Burgos et al., 2006; Hu and Xu, 2018; Lin et al., 2018; Ye et al., 2018). The nugget effect (the ratio of random to total spatial variance) indicates whether regional or local-scale factors are more important for air pollutant distribution. The value of nugget effect ranges from 0 to 100%,
690
W. Song et al. / Science of the Total Environment 631–632 (2018) 688–694
Fig. 2. Schematic diagram of the geographical semi-variogram analysis.
representing strong to weak spatial dependence of the air pollutant. If the nugget effect value is less than 25%, regional-scale influencing factors are more important for the air pollutant distribution; if the ratio is between 25% and 75%, then both regional- and local-scale influencing factors are important for the air pollutant distribution; and if the ratio is greater than 75%, then the local-scale influencing factors are more important for the air pollutant distribution (Liu et al., 2004). However, the nugget effect cannot identify the sources affecting air pollutant distribution. In this case study, we developed the NO2 and PM2.5 semi-variogram models based on daily measurements. The specific parameters of the semi-variogram models were calculated using the geostatistical analyst tool in ArcGIS 10.3. Then, we averaged the daily parameters for each season. Finally, the seasonal fluctuations in the geographical semi-variogram parameters were analysed, to compare the differences between NO2 and PM2.5 spatial distribution characteristics within the study domain. 3. Results 3.1. Descriptive statistics of NO2 and PM2.5 concentrations The spatial difference of NO2 and PM2.5 concentrations was presented in Fig. 3. For the case study city, the annual mean NO2 and
PM2.5 concentrations were 49.5 and 51.8 μg/m3 in 2014, respectively. On average, they exceeded the national standards by more than 65 and 48%, respectively. The station-based annual mean concentrations were highly variable ranging from 31.3 to 64.7 μg/m3 for NO2 and from 40.0 μg/m3 to 63.3 μg/m3 for PM2.5. There were 37 and 38 stations whose annual mean values exceeded the respective national standard. The NO2 and PM2.5 measurements showed a large difference across the study area, and their spatial distributions were quite different. For NO2, the smallest annual mean value appeared in the Gaoming district at 31.3 μg/m3. The largest annual mean value appeared in the Chancheng district at 64.7 μg/m3. For PM2.5, the largest and smallest annual mean values appeared in the Nanhai (63.3 μg/m3) and Shunde (40.0 μg/m3) districts. These results indicat that the study area is suffering from serious air pollution. They also show that the central monitoring stations have higher NO2 concentration levels than the stations on either side of the study area. In contrast, PM2.5 has higher concentration levels in the north-western part of the study area. Given the elevation difference (shown in Fig. 1), we can infer that the terrain has a significant impact on the PM2.5 distribution. Descriptive statistics of the NO2 and PM2.5 measurements are provided in Table 1. The averaged NO2, and PM2.5 concentrations in winter were significantly higher than in summer, showing a significant seasonal variation. The results indicated that the spatial distribution of NO2 and PM2.5 pollution also have a large seasonal difference. Fig. 4 shows the seasonal difference in NO2 and PM2.5 measurements. There were larger differences in site-based mean values in spring and winter. Specifically, the spatial variation in PM2.5 concentrations was largest in winter, ranging from 54.8 to 92.8 μg/m3. Significant increases in NO2 and PM2.5 pollution levels were also observed in winter.
3.2. Geographical semi-variogram models of NO2 and PM2.5 The results of the NO2 and PM2.5 semi-variogram models are summarised in Fig. 5, and the annual and seasonal mean values of these parameters were presented in Table 2. The results show the seasonal differences in the structural, random, and total spatial variance in NO2 and PM2.5, respectively. Visually, it is easy to see the difference between the NO2 and PM2.5 spatial distributions and seasonal mean variances. The degree of spatial variation for NO2 was larger than that of PM2.5. The annual mean nugget value of NO2 was also larger than that
Fig. 3. Spatial differences between NO2 and PM2.5 concentrations.
W. Song et al. / Science of the Total Environment 631–632 (2018) 688–694 Table 1 Descriptive statistics for NO2 and PM2.5 measurements. Air pollutants Indicators
Spring Summer Autumn Winter Annual
NO2
36.4 55.6 81.4 9.3
19.3 41.7 57.3 7.6
32.3 50.0 62.2 6.5
42.4 64.8 85.1 10.2
31.3 52.0 79.4 7.8
0.3 0.6 33.1 43.7 58.3 6.2
−0.2 1.2 22.7 31.5 45.7 6.5
−0.2 0.3 39.8 53.6 66.3 7.1
−0.4 0.2 54.8 71.4 106.4 10.9
0.8 4 40 51.1 106.4 11.3
0.7 −0.2
0.6 −0.5
−0.2 −1.1
1.1 1.7
3.5 16.8
PM2.5
Minimum Mean Maximum Standard deviation Skewness Kurtosis Minimum Mean Maximum Standard deviation Skewness Kurtosis
of PM2.5, reflecting that the spatial variance was larger for NO2 than PM2.5 at the local scale. The sill value of NO2 was largest and smallest in winter and summer, meaning the spatial variance of NO2 was larger in winter than in summer. The seasonal averaged sill value of PM2.5 in winter was 2.1 times higher than that in summer. Overall, the spatial variation for both NO2 and PM2.5 was large in winter and small in summer. The maximum spatial distances of NO2 autocorrelation in spring, summer, autumn and winter were 42, 43, 54 and 53 km, respectively. Several authors have reported spatial distances of 37–58 km for NO2 in Beijing from November to December 2012 (Li et al., 2015). The spatial range of NO2 autocorrelation is smaller in summer and larger in winter. The possible reason for this difference is that NO2 has a longer lifespan in winter, and can achieve longer transport distances. The maximum spatial distance of NO2 autocorrelation in autumn was 28.5% higher than in spring. Spring and summer had similar spatial autocorrelation ranges, and autumn and winter had similar spatial autocorrelation range values. The maximum spatial distances of PM2.5 autocorrelation in spring, summer, autumn and winter were 35, 36, 33, and 20 km, respectively. In contrast with the NO2 results, the range values of PM2.5 spatial autocorrelation were larger in summer than in winter. The maximum spatial distance of PM2.5 autocorrelation reached its peak value (48.1 km) in May, and the smallest spatial distance was only 19.9 km in November. The spatial range of PM2.5 autocorrelation in spring and summer is larger than that in autumn and winter. These results mean that the spatial distribution of PM2.5 is more complex in spring and summer. It can easily be observed that NO2 and PM2.5 has a different spatial autocorrelation range in different seasons. The spatial range of NO2 autocorrelation was larger than that of PM2.5. In summer and autumn, the maximum spatial distance of NO2 autocorrelation was 40 km, whereas it was approximately 33 km for PM2.5. These results indicate that
691
PM2.5 has a smaller spatial overflow effect than NO2 at the local scale. NO2 shows a larger variation in spatial autocorrelation range than PM2.5 in autumn and winter, indicating that the spatiotemporal distribution of NO2 is more complex than that of PM2.5. In contrast, the spatiotemporal distribution of PM2.5 is more complex in summer. In spring, the proportion of NO2 and PM2.5 variance caused by small-small influencing factors were smaller than in other seasons. In winter, the spatial range of NO2 autocorrelation was the largest, and the spatial range of PM2.5 autocorrelation was the smallest. As shown in Fig. 5, the seasonal fluctuations in the nugget effect were also significant for NO2 and PM2.5. For NO2, the nugget effect (i.e. random to total spatial variance ratio) values were 18.4, 33.4, 34.9, and 32.2% in spring, summer, autumn, and winter, respectively. These results show that the spatial dependence of NO2 concentrations was moderate in summer, autumn, and winter, with strong spatial dependence in spring. Thus, the spatial dependence of NO2 was larger in spring than in other seasons. For PM2.5, the nugget effect values were 18.5, 20.3, 21.6, and 23.2% in spring, summer, autumn, and winter, respectively. These results mean that the spatial dependence of PM2.5 was strong in all seasons. The annual mean nugget effect values of NO2 and PM2.5 were 29.7 and 20.9%, respectively. These results indicate that the spatial distribution of NO2 concentrations was affected by both regional and local influencing factors, while the spatial distribution of PM2.5 concentrations is mainly influenced by regional influencing factors such as weather and terrain. For PM2.5, there were 245 days (i.e. 67.1% of the days in 2014) when the nugget effect values were less than 25%. In contrast, there were 178 days (48.8%) when the nugget effect values of NO2 were less than 25%. Overall, the spatial distribution of PM2.5 concentrations is dominated by large-scale influencing factors on most days, and the spatial distribution of NO2 had less continuity and more variance than that of PM2.5. 4. Discussion 4.1. Spatial variation of NO2 and PM2.5 concentrations The total, structural, and random spatial variance of NO2 in 2014 was larger than that of PM2.5. This result indicates that the degree of spatial variation of NO2 is larger than that of PM2.5 in the study area. The structural variances of pollutants were caused by regional influencing factors, including meteorological elements, long-range transport and regionalrelated contributions. The random spatial variance of NO2 and PM2.5 were caused by short-range variation and sampling errors. The spatial variance of NO2 in winter is larger than in summer. This result was expected due to its emission sources and intensity; more emissions occur in winter in the study region. The main sources of NO2 include the combustion of fossil fuels and biofuels, biomass burning, lightning, and bacterially mediated decomposition of nitrogen processed through the agriculture/animal/human food chain (Wang et al., 2007). The
Fig. 4. Seasonal variation in NO2 and PM2.5 measurements.
692
W. Song et al. / Science of the Total Environment 631–632 (2018) 688–694
Fig. 5. Seasonal differences of NO2 and PM2.5 distribution.
structural spatial variance of NO2 reflects the spatial variation at the regional scale and is caused by large-scale influencing factors, such as meteorology, external transport, light-catalysed reactions, and terrain. The random variances in NO2 are caused by local-scale influencing factors, such as industry point emissions within 485 m (the shortest distance to any monitoring station). The spatial variance of PM2.5 was larger in winter than in summer. The larger value for random spatial variance indicates that there are more local sources of PM2.5 in winter. The larger regional-scale spatial variances show the effects of regional influencing factors on background
concentrations. The regional influencing factors include long-range transport and photochemical formation of particle organic compounds. Thus, the values can be attributed to inherent PM2.5 sources (such as soil 2− − dust, NO− 3 , SO4 , NH4 ) and to meteorological factors, long-distance transport and photocatalytic reactions. 4.2. Spatial scale dependence of NO2 dominant factors For NO2, higher nugget effect values were observed in summer, autumn, and winter, whereas lower values were observed in spring. This
W. Song et al. / Science of the Total Environment 631–632 (2018) 688–694 Table 2 Seasonally-averaged results of the NO2 and PM2.5 semi-variogram models. Air pollutants
Parameters
Spring
Summer
Autumn
Winter
Annual
NO2
Sill Partial sill Nugget Range Nugget effect Sill Partial sill Nugget Range Nugget effect
278.1 230.2 47.9 41.9 18.4% 167.2 148.7 18.6 34.5 18.5%
191.5 144.8 46.8 42.8 33.4% 93.2 79.2 14.1 35.8 20.3%
227.5 176.3 51.2 53.7 34.9% 170.0 143.7 26.3 32.0 21.6%
359.4 276.7 82.6 53.8 32.2% 240.1 212.3 27.7 28.8 23.2%
264.1 207.0 57.1 48.0 29.7% 167.6 146.0 21.7 32.8 20.9%
PM2.5
result means that the regional influencing factors had the largest impacts on the spatial variation of NO2 in spring. In addition, the local influencing factors explained the largest fraction of the total spatial variance in autumn. This result means that local emissions (e.g. traffic and power plants) had a strong influence on the NO2 distribution. Overall, the spatial distribution of NO2 is affected by both regional and local factors, and more variance is explained by the regional factors. These results agree with previous land-use regression (LUR) studies (Beelen et al., 2013; Cordioli et al., 2017; Lee et al., 2014; Liu et al., 2015; Tang et al., 2013; Weissert et al., 2018). For example, Hoek et al. (2008) summarised the spatial predictors often selected for final NO2 LUR models and found that the road length with 350 m buffers, traffic volume with 300 m buffers, traffic intensity with 50–250 m buffers, and distance to the nearest roads often appeared as the local-scale spatial predictors. Altitude, land cover, and populations with 5000 m buffers appeared as the regional-scale spatial predictors. Thus, we can infer that traffic, roads, power plants, and land cover are the local-scale influencing factors of NO2 distribution. The altitude (as shown in Fig. 1) and meteorological conditions are the regional-scale influencing factors on NO2 concentrations. The seasonal difference in NO2 variations was caused by the varying effects of the dominant influencing factors. The combination of geographical semi-variation method and Pearson correlation method may be benefit for the potential air pollution source identification or apportionment. 4.3. Spatial scale dependence of PM2.5 dominant factors The nugget effect values of PM2.5 were smaller than that of NO2. This means that the spatial dependence of PM2.5 was stronger than that of NO2. This difference in spatial dependence can be attributed to the NO2 and PM2.5 properties. The former has a shorter life cycle, and the latter has a longer life cycle. This means that regional influencing factors, both natural and anthropogenic, have a larger influence on PM2.5. The spatial predictors that are often selected for the final PM2.5 LUR models include building density with 2500 m buffers, industrial land use with 5000 m buffers, populations with 1000–5000 m buffers, and length of expressway with 1000 m buffers. The regional influencing factors also include the contribution of long-distance transport to PM2.5 variation. Although the nugget effect values can reflect whether regional or local influencing factors dominate the spatial distribution of PM2.5 concentrations, they do not differentiate between natural and anthropogenic influencing factors. The small nugget effect shows that the spatial distribution of PM2.5 concentrations is controlled by regional influencing factors, not local influencing factors. Therefore, soil dust has limited impact on the spatial distribution of PM2.5 in the case study city. This result agrees with previous land-use regression (LUR) studies (Eeftens et al., 2012; Kloog et al., 2012; Lee et al., 2016; Wu et al., 2017a). NO2 is a gaseous pollutant, whose lifespan is short and often varies at the local scale. PM2.5 is a particulate pollutant, with a longer lifespan, and can often be transported at a larger regional scale. Thus, largescale meteorological elements play a more important role in urban PM2.5 distribution, and their effects must be considered when we
693
consider air quality and its influencing factors. For example, light winds, elevated temperatures and surface inversions, and low-mixing heights can significant enhance secondary PM2.5 formation (Wang et al., 2014b). 4.4. Spatial range of NO2 and PM2.5 autocorrelation The seasonal difference in NO2 spatial autocorrelation range can be attributed to the combined effects of climate and emission sources. The large fluctuations in PM2.5 spatial autocorrelation range occurred in summer due to the impact of rainfall and photocatalytic reactions. When rainfall occurs, the PM2.5 spatial autocorrelation range is small; when the meteorological conditions are beneficial to photocatalytic reactions, the PM2.5 spatial autocorrelation range is large. PM2.5 has a smaller spatial autocorrelation range than NO 2 , reflecting different spatial autocorrelation properties. The primary reason is that NO2 and PM2.5 are gaseous and particulate pollutants, respectively. The values of spatial autocorrelation range can act as a guide to design denser sampling plans for the case study city in future. The sampling interval should be smaller than the range of spatial autocorrelation. For instance, it could be a quarter or half of the spatial autocorrelation range (Kerry and Oliver, 2003). To better characterize the urban NO2 and PM2.5 distributions, the sampling interval of NO2 should be 10.5–21 km, 10.8–21.5 km, and 13.5–27 km, 13.3–26.5 km in spring, summer, autumn, and winter, respectively. The PM2.5 sampling interval should be 8.8–17.5 km, 9–18 km, 8.3–16.5 km, and 7.3–14.5 km in spring, summer, autumn, and winter, respectively. In brief, the NO2 sampling interval should be 10.5 to 27 km. The PM2.5 sampling interval should be 7.25 to 18 km. The NO2 sampling interval in spring and summer should be smaller than in autumn and winter. The PM2.5 sampling interval in autumn and winter should be closer than in spring and summer. 5. Conclusion In conclusion, these results provide a better insight into the urban NO2 and PM2.5 distribution. The spatial variance of NO2 was larger than that of PM2.5. There were significant seasonal differences in both NO2 and PM2.5 variations. The spatial variance of NO2 and PM2.5 concentrations were larger in winter and smaller in summer. NO2 showed a moderate spatial dependence and PM2.5 a large spatial dependence. The spatial distribution of NO2 is affected by both local and regional influencing factors. In contrast, the spatial distribution of PM2.5 is dominated by regional influencing factors. The maximum spatial distance of NO2 autocorrelation was largest in winter. The spatial autocorrelation of PM2.5 was largest in summer. As a result, the sampling interval of NO2 should be smaller in summer than in winter, and for PM2.5 the sampling interval should be larger in summer than in winter. In short, the sampling interval of NO2 and PM2.5 should be different, and varying depending on the season. Overall, the geographical semi-variogram method is a very effective approach to enrich the understanding of the differences between NO2 and PM2.5 spatial distribution. It can provide scientific evidence for the buffering radius selection of spatial predictors for LUR models. It will also be beneficial for developing targeted policies and measures to reduce ambient NO2 and PM2.5 pollution levels. In future, the method should be used applied in more cities and for additional air pollutants to verify its reliability. Acknowledgements This study was carried out at the School of Environment, Tsinghua University. We thank the Research Institute for Sustainable Urban Development, at the Hong Kong Polytechnic University for its fellowship support.
694
W. Song et al. / Science of the Total Environment 631–632 (2018) 688–694
References Ahmadi, S.H., Sedghamiz, A., 2007. Geostatistical analysis of spatial and temporal variations of groundwater level. Environ. Monit. Assess. 129, 277–294. Alary, C., Demougeot-Renard, H., 2010. Factorial kriging analysis as a tool for explaining the complex spatial distribution of metals in sediments. Environ. Sci. Technol. 44, 593–599. Badaro-Saliba, N., Adjizian-Gerard, J., Zaarour, R., Abboud, M., Farah, W., Saliba, A.N., et al., 2014. A geostatistical approach for assessing population exposure to NO2 in a complex urban area (Beirut, Lebanon). Stoch. Env. Res. Risk A. 28, 467–474. Beelen, R., Hoek, G., Vienneau, D., Eeftens, M., Dimakopoulou, K., Pedeli, X., et al., 2013. Development of NO2 and NOX land use regression models for estimating air pollution exposure in 36 study areas in Europe – the ESCAPE project. Atmos. Environ. 72, 10–23. Burgos, P., Madejon, E., Perez-De-Mora, A., Cabrera, F., 2006. Spatial variability of the chemical characteristics of a trace-element-contaminated soil before and after remediation. Geoderma 130, 157–175. Callen, M.S., Iturmendi, A., Lopez, J.M., 2014. Source apportionment of atmospheric PM2.5bound polycyclic aromatic hydrocarbons by a PMF receptor model. Assessment of potential risk for human health. Environ. Pollut. 195, 167–177. Cao, S., Lu, A., Wang, J., Huo, L., 2017. Modeling and mapping of cadmium in soils based on qualitative and quantitative auxiliary variables in a cadmium contaminated area. Sci. Total Environ. 580, 430–439. Chen, C., Zhu, P.F., Lan, L., Zhou, L., Liu, R.C., Sun, Q.H., et al., 2018. Short-term exposures to PM2.5 and cause-specific mortality of cardiovascular health in China. Environ. Res. 161, 188–194. Cordioli, M., Pironi, C., De Munari, E., Marmiroli, N., Lauriola, P., Ranzi, A., 2017. Combining land use regression models and fixed site monitoring to reconstruct spatiotemporal variability of NO2 concentrations over a wide geographical area. Sci. Total Environ. 574, 1075–1084. Eeftens, M., Beelen, R., de Hoogh, K., Bellander, T., Cesaroni, G., Cirach, M., et al., 2012. Development of land use regression models for PM2.5, PM2.5 absorbance, PM10 and PM coarse in 20 European study areas; results of the ESCAPE project. Environ. Sci. Technol. 46, 11195–11205. Guo, X.D., Fu, B.J., Ma, K.M., Chen, L.D., 2001. Utility of semivariogram for spatial variation of soil nutrients and the robust analysis of semivariogram. J. Environ. Sci. (China) 13, 453–458. Hoek, G., Beelen, R., de Hoogh, K., Vienneau, D., Gulliver, J., Fischer, P., et al., 2008. A review of land-use regression models to assess spatial variation of outdoor air pollution. Atmos. Environ. 42, 7561–7578. Hu, X., Xu, H., 2018. A new remote sensing index for assessing the spatial heterogeneity in urban ecological quality: a case from Fuzhou city, China. Ecol. Indic. 89, 11–21. Jiang, Q.T., Christakos, G., 2018. Space-time mapping of ground-level PM2.5 and NO2 concentrations in heavily polluted northern China during winter using the Bayesian maximum entropy technique with satellite data. Air Qual. Atmos. Health 11, 23–33. Kerry, R., Oliver, M.A., 2003. Variograms of ancillary data to aid sampling for soil surveys. Precis. Agric. 4, 261–278. Kloog, I., Nordio, F., Coull, B.A., Schwartz, J., 2012. Incorporating local land use regression and satellite aerosol optical depth in a hybrid model of spatiotemporal PM2.5 exposures in the mid-Atlantic states. Environ. Sci. Technol. 46, 11913–11921. Lee, J.-H., Wu, C.-F., Hoek, G., de Hoogh, K., Beelen, R., Brunekreef, B., et al., 2014. Land use regression models for estimating individual NOX and NO2 exposures in a metropolis with a high density of traffic roads and population. Sci. Total Environ. 472, 1163–1171. Lee, H.J., Chatfield, R.B., Strawa, A.W., 2016. Enhancing the applicability of satellite remote sensing for PM2.5 estimation using modis deep blue aod and land use regression in California, United States. Environ. Sci. Technol. 50, 6546–6555. Li, Z., Yang, H., Jiang, Y., Yin, Y., Wang, H., Wang, B., et al., 2015. The study on spatial distribution of atmospheric contaminant concentrations and optimization of air quality monitoring sites in Beijing city. Environ. Monit. China 31, 74–78. Li, L., Lurmann, F., Habre, R., Urman, R., Rappaport, E., Ritz, B., et al., 2017. Constrained mixed-effect models with ensemble learning for prediction of nitrogen oxides concentrations at high spatiotemporal resolution. Environ. Sci. Technol. 51, 9920–9929. Liao, H.T., Chou, C.C.K., Huang, S.H., Lu, C.J., Chen, C.C., Hopke, P.K., et al., 2017. Source apportionment of PM2.5 size distribution and composition data from multiple stationary sites using a mobile platform. Atmos. Res. 190, 21–28. Lin, Y.-P., 2002. Multivariate geostatistical methods to identify and map spatial variations of soil heavy metals. Environ. Geol. 42, 1–10. Lin, Z., Chao, L., Wu, C., Hong, W., Hong, T., Hu, X., 2018. Spatial analysis of carbon storage density of mid-subtropical forests using geostatistics: a case study in Jiangle county, Southeast China. Acta Geochim. 37, 90–101.
Liu, X.M., Xu, J.M., Zhang, M.K., Huang, J.H., Shi, J.C., Yu, X.F., 2004. Application of geostatistics and GIS technique to characterize spatial variabilities of bioavailable micronutrients in paddy soils. Environ. Geol. 46, 189–194. Liu, Y., Lv, J., Zhang, B., Bi, J., 2013. Spatial multi-scale variability of soil nutrients in relation to environmental factors in a typical agricultural region, eastern China. Sci. Total Environ. 450-451, 108–119. Liu, W., Li, X., Chen, Z., Zeng, G., León, T., Liang, J., et al., 2015. Land use regression models coupled with meteorology to model spatial and temporal variability of NO2 and PM10 in Changsha, China. Atmos. Environ. 116, 272–280. Liu, C., Henderson, B.H., Wang, D.F., Yang, X.Y., Peng, Z.R., 2016. A land use regression application into assessing spatial variation of intra-urban fine particulate matter (PM2.5) and nitrogen dioxide (NO2) concentrations in city of Shanghai, China. Sci. Total Environ. 565, 607–615. Liu, M., Bi, J., Ma, Z., 2017. Visibility-based PM2.5 concentrations in China: 1957–1964 and 1973–2014. Environ. Sci. Technol. 51, 13161–13169. Lv, J., Liu, Y., Zhang, Z., Dai, B., 2014. Multivariate geostatistical analyses of heavy metals in soils: spatial multi-scale variations in Wulian, eastern China. Ecotoxicol. Environ. Saf. 107, 140–147. Pahlavani, P., Sheikhian, H., Bigdeli, B., 2017. Assessment of an air pollution monitoring network to generate urban air pollution maps using Shannon information index, fuzzy overlay, and Dempster-Shafer theory, a case study: Tehran, Iran. Atmos. Environ. 167, 254–269. Qu, W.J., Arimoto, R., Zhang, X.Y., Zhao, C.H., Wang, Y.Q., Sheng, L.F., et al., 2010. Spatial distribution and interannual variation of surface PM10 concentrations over eightysix Chinese cities. Atmos. Chem. Phys. 10, 5641–5662. Song, W., Jia, H., Huang, J., Zhang, Y., 2014. A satellite-based geographically weighted regression model for regional PM2.5 estimation over the Pearl River Delta region in China. Remote Sens. Environ. 154, 1–7. Spokas, K., Graff, C., Morcet, M., Aran, C., 2003. Implications of the spatial variability of landfill emission rates on geospatial analyses. Waste Manag. 23, 599–607. Strak, M., Janssen, N., Beelen, R., Schmitz, O., Vaartjes, I., Karssenberg, D., et al., 2017. Longterm exposure to particulate matter, NO2 and the oxidative potential of particulates and diabetes prevalence in a large national health survey. Environ. Int. 108, 228–236. Tang, R., Blangiardo, M., Gulliver, J., 2013. Using building heights and street configuration to enhance intraurban PM10, NOX, and NO2 land use regression models. Environ. Sci. Technol. 47, 11643–11650. Tessum, M.W., Larson, T., Gould, T.R., Simpson, C.D., Yost, M.G., Vedal, S., 2018. Mobile and fixed-site measurements to identify spatial distributions of traffic-related pollution sources in Los Angeles. Environ. Sci. Technol. 52 (5), 2844–2853. Wang, Y.X., McElroy, M.B., Martin, R.V., Streets, D.G., Zhang, Q., Fu, T.M., 2007. Seasonal variability of NOX emissions over east China constrained by satellite observations: implications for combustion and microbial sources. J. Geophys. Res.-Atmos. 112. Wang, M., Beelen, R., Basagana, X., Becker, T., Cesaroni, G., de Hoogh, K., et al., 2013. Evaluation of land use regression models for NO2 and particulate matter in 20 European study areas: the ESCAPE project. Environ. Sci. Technol. 47, 4357–4364. Wang, L., Coles, N.A., Wu, C.F., Wu, J.P., 2014a. Spatial variability of heavy metals in the coastal soils under long-term reclamation. Estuar. Coast. Shelf Sci. 151, 310–317. Wang, Y.G., Ying, Q., Hu, J.L., Zhang, H.L., 2014b. Spatial and temporal variations of six criteria air pollutants in 31 provincial capital cities in China during 2013–2014. Environ. Int. 73, 413–422. Weissert, L.F., Salmond, J.A., Miskell, G., Alavi-Shoshtari, M., Williams, D.E., 2018. Development of a microscale land use regression model for predicting NO2 concentrations at a heavy trafficked suburban area in Auckland, NZ. Sci. Total Environ. 619-620, 112–119. Wu, C.D., Chen, Y.C., Pan, W.C., Zeng, Y.T., Chen, M.J., Guo, Y.L., et al., 2017a. Land-use regression with long-term satellite-based greenness index and culture-specific sources to model PM2.5 spatial-temporal variability. Environ. Pollut. 224, 148–157. Wu, J.R., Li, G.H., Cao, J.J., Bei, N.F., Wang, Y.C., Feng, T., et al., 2017b. Contributions of transboundary transport to summertime air quality in Beijing, China. Atmos. Chem. Phys. 17, 2035–2051. Xie, Y., Wang, Y., Zhang, K., Dong, W., Lv, B., Bai, Y., 2015. Daily estimation of ground-level PM2.5 concentrations over Beijing using 3 km resolution MODIS AOD. Environ. Sci. Technol. 49, 12280–12288. Ye, L., Tan, W., Fang, L., Ji, L., Deng, H., 2018. Spatial analysis of soil aggregate stability in a small catchment of the loess plateau, China: I. Spatial variability. Soil Tillage Res. 179, 71–81.