Next Article in Journal
Australian Food Safety Policy Changes from a “Command and Control” to an “Outcomes-Based” Approach: Reflection on the Effectiveness of Its Implementation
Previous Article in Journal
Knowledge, Attitudes and Behaviours of Healthcare Workers in the Kingdom of Saudi Arabia to MERS Coronavirus and Other Emerging Infectious Diseases
Previous Article in Special Issue
Spatial Analysis of Severe Fever with Thrombocytopenia Syndrome Virus in China Using a Geographically Weighted Logistic Regression Model
Article Menu

Export Article

Int. J. Environ. Res. Public Health 2016, 13(12), 1215; https://doi.org/10.3390/ijerph13121215

Article
Ground Level PM2.5 Estimates over China Using Satellite-Based Geographically Weighted Regression (GWR) Models Are Improved by Including NO2 and Enhanced Vegetation Index (EVI)
1
State Key Laboratory of Information Engineering in Surveying, Mapping, and Remote Sensing, Wuhan University, Wuhan 430079, China
2
Collaborative Innovation Center for Geospatial Technology, Wuhan 430079, China
3
College of Information Science and Engineering, Wuchang Shouyi University, Wuhan 430064, China
*
Authors to whom correspondence should be addressed.
Academic Editor: Jamal Jokar Arsanjani
Received: 21 September 2016 / Accepted: 5 December 2016 / Published: 7 December 2016

Abstract

:
Highly accurate data on the spatial distribution of ambient fine particulate matter (<2.5 μm: PM2.5) is currently quite limited in China. By introducing NO2 and Enhanced Vegetation Index (EVI) into the Geographically Weighted Regression (GWR) model, a newly developed GWR model combined with a fused Aerosol Optical Depth (AOD) product and meteorological parameters could explain approximately 87% of the variability in the corresponding PM2.5 mass concentrations. There existed obvious increase in the estimation accuracy against the original GWR model without NO2 and EVI, where cross-validation R2 increased from 0.77 to 0.87. Both models tended to overestimate when measurement is low and underestimate when high, where the exact boundary value depended greatly on the dependent variable. There was still severe PM2.5 pollution in many residential areas until 2015; however, policy-driven energy conservation and emission reduction not only reduced the severity of PM2.5 pollution but also its spatial range, to a certain extent, from 2014 to 2015. The accuracy of satellite-derived PM2.5 still has limitations for regions with insufficient ground monitoring stations and desert areas. Generally, the use of NO2 and EVI in GWR models could more effectively estimate PM2.5 at the national scale than previous GWR models. The results in this study could provide a reasonable reference for assessing health impacts, and could be used to examine the effectiveness of emission control strategies under implementation in China.
Keywords:
nationwide ambient PM2.5; MODIS (Moderate Resolution Imaging Spectroradiometer) AOD; satellite-derived NO2 column density; enhanced vegetation index; geographically weighted regression

1. Introduction

Numerous previous studies reported that atmospheric particulate matter emitted from both anthropogenic and natural sources exert influences on climate change and environmental deterioration [1,2]. Many epidemiological studies have shown that exposure to fine suspended particles with aerodynamic diameter less than 2.5 μm (PM2.5) are linked with cardiovascular and respiratory diseases [3,4,5,6]. With vast consumption of energy and rapid economic development, China has suffered from severe PM2.5 pollution and the related social problems have caused wide concerns [7,8]. Although an air quality monitoring network has been established in China since 2013, large-scale estimation of PM2.5 is not practical due to the limited spatial coverage of the monitoring stations [9]. Consequently, there exists an urgent need to acquire a spatially resolved characterization of human exposure to PM2.5 at the national scale in China [10].
In order to make up for the blanks in ground measurements, methods using remote sensing by satellite have been adopted to estimate ground-level PM2.5 mass concentrations [11,12]. There exists a direct relationship between the atmospheric particles and satellite-derived Aerosol Optical Depth (AOD) because AOD represents the quantity of light removed from a beam by the role of aerosol scattering or absorption during its path [13,14,15,16]. Previous studies proposed establishing empirical models to correlate ground-level PM2.5 and satellite derived AOD (e.g., linear, nonlinear, and logarithmic models) [13,17,18,19]. Superior models have been established for better predicting the PM2.5 concentrations combined with meteorological parameters or based on atmospheric transport models, such as the mixed effects model (MEM) [20], the artificial neural network model (ANN) [17,21,22], and the chemical transport model (CTM) [23]. Since the correlation between AOD and PM2.5 should vary along with the spatial context due to different geographical areas possessing different aerosol types [9,24,25], the Geographically Weighted Regression (GWR) model, which better constrains the spatial variability in a large-region regression, has been adopted to estimate geographical elements in large regions [26,27]. The particulate matter vertical distribution has been additionally taken into consideration from a physics perspective, which could improve the correlation between PM2.5 and AOD [28]. Moreover, previous study of the GWR model combined with physical corrections indicated that the atmospheric vertical feature could be embedded in the GWR model via Planetary Boundary Layer Height (PBLH) for better estimation accuracy of atmospheric particulates at the national scale [29]. Nevertheless, these statistical models rely greatly on observation from satellites and meteorological parameters, and anthropogenic emissions need additionally consideration, such as traffic density [30].
In this study, the molar concentration of OMI-NO2 (Ozone Measuring Instrument) and MODIS-EVI (Moderate Resolution Imaging Spectroradiometer) were introduced into the GWR model, considering that these two variables can be measured via satellites at a large spatial scale. A different GWR model was then established to predict PM2.5 concentrations at the national scale using fused MODIS-AOD based on a dark target and deep blue algorithm, meteorological parameters, daily tropospheric NO2 molar concentration, and fixed 16-day composite EVI. For quantitatively evaluating the model performance, a leave-one-out cross-validation was adopted to demonstrate the relationship between measured PM2.5 and estimated PM2.5, and results from the previous GWR model without NO2 and EVI were also provided for contrast. Moreover, the two-year annual spatial distribution of the satellite-derived PM2.5 at the national scale was demonstrated, analyzed, and discussed.

2. Materials and Methods

2.1. Ground PM2.5 Measurements

Hourly ground-level PM2.5 measurements in China from 1 January 2014 to 31 December 2015 were collected primarily from the official website of the China Environmental Monitoring Center (CEMC) [31] As demonstrated in Figure 1, more than 1300 air quality monitoring stations have been built up covering residential cities in all provinces of China by the end of 2014. According to the officially released documents of the Chinese Ministry of Environmental Protection (MEP), the PM2.5 data were measured using the tapered element oscillating microbalance method (TEOM) or the beta-attenuation method, combined with periodic calibration [32].

2.2. Moderate Resolution Imaging Spectroradiometer (MODIS) Aerosol Products

The MODIS aboard the NASA Earth Observing System (EOS) satellite, Aqua, was shown to provide aerosol products with assured quality when compared with other satellite sensors [33,34,35,36]. In this study, we use the Collection 6 (C6) aerosol products because these products have generally been proven to attain fine accuracy by validation against ground monitoring observations of sun photometers from the Aerosol Robotic Network (AERONET) in China [37,38]. The Aqua AOD datasets used in this study, which were retrieved based on the relationship between the radiation value of specific bands over land and the aerosol optical depth, were distributed in Hierarchical Data Format (HDF) format from the NASA Goddard Space Flight Center [39] at the national scale (longitude (73°40’–135°2.5’ E), latitude (3°52’–53°33’ N)). The second generation Deep Blue (DB) was expanded to cover brighter desert/urban areas and vegetated land surfaces, which preferably makes up for deficiencies in the Dark Target (DT) algorithms [40,41]. However, the spatial resolution of the Aqua Level 2 AOD using DB algorithms is 10 × 10 km, while the spatial resolution of the Aqua Level 2 AOD using DT algorithms reaches 3 × 3 km. Thus, the AOD datasets were fused by 3 km DT AOD and 10 km DB AOD, based on the concept of complementary advantages, in order to make merging AOD possess advantages on both spatial resolution and spatial coverage [42]. Moreover, only retrievals reaching the required quality assurance (QA) were used (corresponding to flag QA = 3 for DT; flag QA = 2 or QA = 3 for DB), ensuring the accuracy of the fused AOD.

2.3. Aerological and Surface Meteorological Parameters

Aerological parameters, including PBLH, and surface meteorological parameters, consisting of surface relative humidity, u-components, and v-components of surface winds, surface temperature, and surface atmospheric pressure, were collected from the National Centers for Environmental Prediction (NCEP) reanalysis datasets with 1° spatial resolution. Atmospheric product and land surface data taken every six hours are available in the NCEP datasets, which are available on its website [43].

2.4. Satellite-Derived EVI and NO2 Data

The Enhanced Vegetation Index (EVI), which is calculated by surface reflectance in near-infrared, red, and blue bands, is an optimized vegetation index to represent vegetation biomass. The EVI, instead of the normalized difference vegetation index (NDVI) data, was adopted in this study to simply characterize the reduction effect by vegetation on PM2.5 concentrations because the EVI is more sensitive to variations in regions having high biomass, which is an obvious advantage over NDVI [44]. Although the calculation of EVI is similar to NDVI, the distortions caused by atmospheric particles in the reflected light have been corrected, and the EVI product barely becomes saturated when scanning regions containing large amounts of chlorophyll [45]. EVI data with 1 km spatial resolution and 16-day temporal resolution was collected from the NASA LAADS website [39].
The daily tropospheric column density of NO2 were collected from the NASA OMI level 2 nitrogen dioxide dataset with 0.25° spatial resolution [46]. The OMI NO2 algorithm could compute accurate vertical column densities from NO2 slant column densities, retrieved by spectral fitting. Since NO2 is a short-life trace gas and has a tight relation to anthropogenic emissions and energy consumption [47,48,49,50], we utilized the satellite-derived NO2 molar concentration as a simplified proxy of anthropogenic emissions.

2.5. Data Integration

From the perspective of time, the Aqua MODIS and Aura-OMI passes the equator at approximately 1:30 p.m. local time, thus all the aerological and surface meteorological variables were correspondingly selected around 2:00 p.m. local time (corresponding to UTC = 6:00 a.m.). Moreover, the daily NO2 dataset and 16-day EVI dataset were respectively utilized to represent the daily extent of anthropogenic emissions and vegetation coverage for 16 days. In addition, a Kriging resampling approach was applied to ensure spatial consistency of all independent variables, which were demonstrated in Table 1 in detail. Moreover, the meteorological parameters and AOD values were both selected from the pixel in which the ground-level monitor is located.

2.6. Model Development, Comparison, and Validation

Because previous research has shown that the relationship between AOD and PM2.5 obviously varies according to the spatial context, and the correlation coefficients could lead to poor accuracy of estimation when using global parameters [36]. To solve this problem, a GWR model has been established in this study, and the adaptive Gaussian bandwidth search method was utilized to accommodate the uneven distribution of ground-monitoring stations. The structure of the GWR model developed in this study is expressed in the following equation:
P M 2.5 , l , d = β 0 , l , d + β 1 , l , d R e v i s e d A O D l , d + β 2 , l , d S T l , d + β 3 , l , d R H l , d + β 4 , l , d P S l , d + β 5 , l , d W S l , d + β 6 , l , d N O 2 , l , d + β 7 , l , d E V I l , d
where the PM2.5,l,d (μg/m3) is the ground-level PM2.5 concentration at location l on day d; β0,l,d denotes the intercept at location l on day d; β0,l,d to β7,l,d represent location-specific slopes; Revised_AODl,d (no unit) is the Aqua-MODIS AOD fused products revised by PBLH (m) at location l on day d. As demonstrated in previous studies, physics corrections based on vertical distribution could make remarkable improvement in the relationship between AOD and PM [29,51]. Because the ground monitoring stations measured mass concentration data of near ground PM2.5, optical parameters for the ambient atmosphere should be measured instead of satellite-derived AOD that characterizes the entire atmospheric column. STl,d, RHl,d, PSl,d and WSl,d are the surface temperature (K), surface relative humidity (%), atmospheric pressure (Pa), and surface wind speed (m/s), respectively, at location l on day d. NO2,l,d represents the column molar concentration of NO2 at location l on day d, and EVIl,d represents the enhanced vegetation index at location l on 16-day d.
Moreover, to examine the effectiveness of introducing NO2 and EVI into the GWR model as representatives of the anthropogenic emissions and the reduction effect by vegetation on PM2.5. As a contrast, a GWR model without parameters of NO2 and EVI was established using the same meteorological factors:
P M 2.5 , l , d = β 0 , l , d + β 1 , l , d R e v i s e d _ A O D l , d + β 2 , l , d S T l , d + β 3 , l , d R H l , d + β 4 , l , d P S l , d + β 5 , l , d W S l , d
In addition, a 10-fold cross validation analysis [52] was conducted to validate the quality of the model by comparing the estimated PM2.5 against the monitoring values. The entire dataset was randomly split into 10 folds, with around ten percent of the total data in each subset, and the model was then fitted by nine folds with one fold set for validation in each cross-validation circle. This process was completely repeated 10 times so as to validate every fold. Furthermore, the estimation equation, decision coefficient R2, and mean absolute error (MAE, μg/m3) were calculated to evaluate the model performance.

3. Results and Discussion

3.1. Descriptive Statistics

The histograms and descriptive statistics of all the variables in the GWR model are illustrated in Figure 2, including the dependent variable and independent variables. It demonstrates that, apart from the surface air pressure displaying a bimodal distribution, which is possibly caused by the obvious elevation difference in southwest China, the remaining variables were approximately log-normal distributed. Overall, the mass concentrations of PM2.5 ranged from 1 to 864.5 μg/m3 with an annual average of 50.38 μg/m3 and standard deviation (Std. Dev.) of 45.33 μg/m3. The AOD frequency histograms have a shape similar to the measured PM2.5 with an annual average AOD value of 497.49 and Std. Dev. of 470.72. Considering the Chinese standard of ambient air quality [53], the averaged mass concentration of PM2.5 ranging from 2014 to 2015 exceeded the level 2 standard (35 μg/m3).

3.2. Model Fitting, Validation, and Comparison

The mass concentrations of PM2.5 and corresponding estimations were obtained across China from 1 January 2014 to 31 December 2015. The original GWR model, the compared GWR model only adding NO2, the compared GWR model only adding EVI, and newly developed GWR model were tested using the same datasets. There existed 119,885 pieces of data added to the model, and the cross-validation (CV) results for the GWR model are shown in Figure 3. The R2 of the CV validation increased from 0.77 to 0.78 when only adding NO2, with MAE decreased from 15.81 μg/m3 to 15.08 μg/m3. In addition, the R2 of the CV validation increased from 0.77 to 0.82 when only adding EVI, with MAE decreased from 15.81 μg/m3 to 13.83 μg/m3. It showed that the introduction of NO2 into the GWR model slightly improved the performance of model, while the introduction of EVI into the GWR model could substantially improve the performance of model. The R2 of the CV validation increased from 0.77 to 0.87 when adding two parameters, which means that both parameters are relevant to the new developed GWR model. This two-year estimation result could account for approximately 87% of the variability in the corresponding PM2.5 mass concentrations, which was relatively great at the national scale when comparing to other studies in China (cross validation R2 achieved 0.64 and 0.79 in annual average, respectively) [38,54]. Moreover, the MAE calculated from the new model decrease from 15.81 μg/m3 to 11.84 μg/m3 against the original model, indicating that PM2.5 estimated using the GWR model with NO2 and EVI agreed better with the measurements. Moreover, according to the linear regression results, the regression line (solid black) of the new model better fit the actual line (dotted black), which illustrates that the introduction of NO2 and EVI into GWR model could improve the estimation of PM2.5, since it could achieve higher R2 and better consistency with measurements.
However, the slope of the linear fitting equation was less than 1.0 and the intercept was positive, which suggests that the model is inclined to overestimate when the measurements of PM2.5 are less than around 60 μg/m3, and to underestimate when the measurements of PM2.5 are larger than approximately 60 μg/m3. Furthermore, the demarcation point, which was calculated by the point of intersection of linear fitting line and y = x reference line, was roughly equal in both GWR models. This phenomenon was probably caused by a mechanism of the GWR model that the coefficients of variables tend to be similar when they are geographically close to each other; therefore, the estimated values seem to be slightly averaged in a small region. Therefore, the particular value at an intersection point depends on the measured value of the dependent variable (PM2.5) to a large extent.

3.3. Annual Estimation of PM2.5 Mass Concentration

The estimated annual averaged PM2.5 mass concentrations in 2014 and 2015 are illustrated in Figure 4. From a macro perspective, the spatial distributions of PM2.5 are generally consistent from the year 2014 to 2015. The highest values of PM2.5 were observed in the Jing-Jin-Ji Region (including Beijing, Tianjin, and Hebei), followed by Central China (including Hunan, Hubei, and Henan), and the Xinjiang Autonomous Region. In the Jing-Jin-Ji Region, the annual averaged PM2.5 mass concentrations were generally higher than 80 µg/m3 in 2014, and higher than 60 µg/m3 in 2015. High levels of industrialization and urbanization, as well as the activities of dense human populations, have led to severe PM2.5 pollution in these areas [55,56]. Based on the World Health Organization (WHO) Air Quality Interim Target (IT) levels, WHO IT-1 set the PM2.5 mass concentration standard of 35 μg/m3 [57], which means that most of regions in the North China Plain still suffer from severe fine particulate pollution. The situation is similar in central China, where most regions slightly exceeded the WHO IT-1 standard until 2015. The cleanest regions are in Tibet, Yunnan, and Hainan, where the annual average PM2.5 values are generally lower than 20 µg/m3.
Nevertheless, the policy-driven efforts in energy conservation and emission reduction should not be ignored. Although there still exists PM2.5 pollution in many residential areas, the apparent decrease in the overall trend is worth recognition. Not only the severity of PM2.5 pollution but also the spatial coverage of heavy PM2.5 pollution was alleviated, to a certain extent, from 2014 to 2015, especially in the North China Plain, Central Plain, scattered regions in the northeastern area and along the southern coast.
Although the model was demonstrated to estimate the spatial distribution of PM2.5 effectively in most of areas in China, there still exist estimation errors in several regions. For instance, the mass concentrations of PM2.5 in the Xinjiang Autonomous Region should be relatively high because the Tarim Basin, which is located in the southern section of the Xinjiang Autonomous Region, is mostly covered by the Taklimakan Desert, where dust aerosols come mostly from primal generation and entrained effects across eastern Asia [58]. However, the prediction results demonstrated that the annual averaged mass concentration of PM2.5 in the Xinjiang Autonomous Region were less than 70 µg/m3, which was slightly below the measurements. The possible reason for this phenomenon in the model prediction is the introduction of NO2. As widely acknowledged, the Xinjiang Autonomous Region is mostly covered by desert, which leads to relatively lower density of NO2 measured by satellite. Thus, due to the positive fitted coefficients of NO2, the prediction results would be less than measurements when there was no ground air-quality monitoring station in the desert. Moreover, the predictions of PM2.5 in 2014 in Tibet, Qinghai, Gansu, northern Inner Mongolia, and western Sichuan were a little bit overestimated. This overestimation phenomenon primarily resulted from uneven distribution of the ground level PM2.5 monitoring sites in 2014. There was primary coverage in large urban centers and sparse coverage in rural areas, especially in the western regions of the country. Less coverage of ground level PM2.5 monitoring stations in these regions would directly result in the increase of bandwidth in the GWR model, further leading to the consequence that regional modeling would be influenced primarily by the farther surrounding areas. Since the GWR model adopts an adaptive bandwidth searching method, the bandwidth will increase when the ground stations are sparse, further decreasing the effects of model performance. However, along with construction of the Chinese air-quality monitoring network, the situation was alleviated in 2015, and this issue will gradually be resolved. In the current data conditions, the model could only achieve anticipative efficacy of estimation in eastern China.

4. Conclusions

The estimation of mass concentrations of PM2.5 at the national scale was conducted using a GWR model, by introducing NO2 and EVI, combined with fused Aqua MODIS AOD and meteorological parameters. According to the results of 10-fold cross-validation, the introduction of NO2 and EVI as independent variables in the GWR model obviously improved the quality of estimation, where the decision coefficient R2 increased from 0.77 to 0.87, and MAE decreased from 15.81 to 11.84 μg/m3 against the results from the original model. Moreover, the newly developed model and original model both tended to overestimate when the measured value of PM2.5 was less than around 60 μg/m3, but tended to underestimate when the measured value was larger than approximately 60 μg/m3. In addition, this particular value that marks this boundary, which was calculated from the intersection point of the linear fitting line and y = x as the reference line, depends on the measurements of dependent variable (PM2.5) to a large extent.
According to the spatial distribution of annual average PM2.5 mass concentrations in 2014 and 2015, the highest values of PM2.5 arose in the Jing-Jin-Ji Region, followed by Central China and the Xinjiang Autonomous Region. Furthermore, the NO2 embedded GWR model would be inclined to underestimate in the desert regions that possess relatively low density of NO2 measured by satellite. In addition, regional modeling would be influenced more by surrounding areas when the coverage of ground level monitoring stations is insufficient. Thus, in current data conditions, it is hard to achieve anticipative performance of estimation in western China. Nevertheless, this situation was alleviated in 2015 and will gradually be resolved following construction of the air-quality monitoring network in China.
In general, the introduction of NO2 and EVI into the GWR model could more effectively estimate PM2.5 at the national scale compared to the original GWR model using satellite-derived AOD and meteorological parameters. Further research will be conducted on the modeling algorithm and source apportionment of PM2.5 to achieve better performance. The estimation of PM2.5 mass concentrations at the national scale could provide a reasonable reference for assessing health impacts in China, and for examining the effectiveness of the emission control strategies under implementation.

Acknowledgments

Our study was supported by the Natural Science Foundation of Hubei (No. 2015CFA002), the National Natural Science Foundation of China (No. 41101334, No. 41127901, No. 41571344), and the China Postdoctoral Science Foundation (No. 2015M572198).

Author Contributions

The research topic was designed by Tianhao Zhang. Tianhao Zhang and Yuxi Ji conducted the experiment, and the paper was written by Wei Gong and Wei Wang. Tianhao Zhang, Zhongmin Zhu, and Yusi Huang checked the experimental data and examined the experimental results. All authors agreed to submission of the manuscript.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Kaufman, Y.J.; Tanré, D.; Boucher, O. A satellite view of aerosols in the climate system. Nature 2002, 419, 215–223. [Google Scholar] [CrossRef] [PubMed]
  2. Peters, A.; Dockery, D.W.; Muller, J.E.; Mittleman, M.A. Increased particulate air pollution and the triggering of myocardial infarction. Circulation 2001, 103, 2810–2815. [Google Scholar] [CrossRef] [PubMed]
  3. See, S.W.; Balasubramanian, R. Chemical characteristics of fine particles emitted from different gas cooking methods. Atmos. Environ. 2008, 42, 8852–8862. [Google Scholar] [CrossRef]
  4. Van Donkelaar, A.; Martin, R.V.; Brauer, M.; Kahn, R.; Levy, R.; Verduzco, C.; Villeneuve, P.J. Global Estimates of Ambient Fine Particulate Matter Concentrations from Satellite-Based Aerosol Optical Depth: Development and Application; University of British Columbia: Vancouver, BC, Canada, 2015. [Google Scholar]
  5. Brunekreef, B.; Forsberg, B. Epidemiological evidence of effects of coarse airborne particles on health. Eur. Respir. J. 2005, 26, 309–318. [Google Scholar] [CrossRef] [PubMed]
  6. Künzli, N.; Jerrett, M.; Mack, W.J.; Beckerman, B.; LaBree, L.; Gilliland, F.; Thomas, D.; Peters, J.; Hodis, H.N. Ambient air pollution and atherosclerosis in Los Angeles. Environ. Health Perspect. 2005, 113, 201–206. [Google Scholar] [CrossRef] [PubMed]
  7. Zhang, Q.; He, K.; Huo, H. Policy: Cleaning China’s air. Nature 2012, 484, 161–162. [Google Scholar] [PubMed]
  8. Chan, C.K.; Yao, X. Air pollution in mega cities in China. Atmos. Environ. 2008, 42, 1–42. [Google Scholar] [CrossRef]
  9. Hu, X.; Waller, L.A.; Al-Hamdan, M.Z.; Crosson, W.L.; Estes, M.G.; Estes, S.M.; Quattrochi, D.A.; Sarnat, J.A.; Liu, Y. Estimating ground-level PM2.5 concentrations in the southeastern U.S. using geographically weighted regression. Environ. Res. 2013, 121, 1–10. [Google Scholar] [CrossRef] [PubMed]
  10. Hu, X.; Waller, L.A.; Lyapustin, A.; Wang, Y.; Al-Hamdan, M.Z.; Crosson, W.L.; Estes, M.G.; Estes, S.M.; Quattrochi, D.A.; Puttaswamy, S.J. Estimating ground-level PM2.5 concentrations in the southeastern United States using MAIAC AOD retrievals and a two-stage model. Remote Sens. Environ. 2014, 140, 220–232. [Google Scholar] [CrossRef]
  11. Van Donkelaar, A.; Martin, R.V.; Brauer, M.; Kahn, R.; Levy, R.; Verduzco, C.; Villeneuve, P.J. Global estimates of ambient fine particulate matter concentrations from satellite-based aerosol optical depth: Development and application. Environ. Health Perspect. 2010, 118, 847. [Google Scholar] [CrossRef] [PubMed]
  12. Che, H.; Xia, X.; Zhu, J.; Wang, H.; Wang, Y.; Sun, J.; Zhang, X.; Shi, G. Aerosol optical properties under the condition of heavy haze over an urban site of Beijing, China. Environ. Sci. Pollut. Res. 2015, 22, 1043–1053. [Google Scholar] [CrossRef] [PubMed]
  13. Chu, D.A.; Kaufman, Y.; Zibordi, G.; Chern, J.; Mao, J.; Li, C.; Holben, B. Global monitoring of air pollution over land from the Earth Observing System-Terra Moderate Resolution Imaging Spectroradiometer (MODIS). J. Geophys. Res. Atmos. 2003, 108. [Google Scholar] [CrossRef]
  14. Koelemeijer, R.; Homan, C.; Matthijsen, J. Comparison of spatial and temporal variations of aerosol optical thickness and particulate matter over Europe. Atmos. Environ. 2006, 40, 5304–5315. [Google Scholar] [CrossRef]
  15. Gupta, P.; Christopher, S.A.; Wang, J.; Gehrig, R.; Lee, Y.; Kumar, N. Satellite remote sensing of particulate matter and air quality assessment over global cities. Atmos. Environ. 2006, 40, 5880–5892. [Google Scholar] [CrossRef]
  16. Che, H.; Xia, X.; Zhu, J.; Li, Z.; Dubovik, O.; Holben, B.; Goloub, P.; Chen, H.; Estelles, V.; Cuevas-Agulló, E. Column aerosol optical properties and aerosol radiative forcing during a serious haze-fog month over North China Plain in 2013 based on ground-based sunphotometer measurements. Atmos. Chem. Phys. 2014, 14, 2125–2138. [Google Scholar] [CrossRef]
  17. Gupta, P.; Christopher, S.A. Particulate matter air quality assessment using integrated surface, satellite, and meteorological products: Multiple regression approach. J. Geophys. Res. Atmos. 2009, 114. [Google Scholar] [CrossRef]
  18. Liu, Y.; Sarnat, J.A.; Kilaru, V.; Jacob, D.J.; Koutrakis, P. Estimating ground-level PM2.5 in the eastern United States using satellite remote sensing. Environ. Sci. Technol. 2005, 39, 3269–3278. [Google Scholar] [CrossRef] [PubMed]
  19. Wallace, J.; Kanaroglou, P. An investigation of air pollution in southern Ontario, Canada, with MODIS and MISR aerosol data. In Proceedings of the Geoscience and Remote Sensing Symposium, Barcelona, Spain, 23–28 July 2007; pp. 4311–4314.
  20. Lee, H.; Liu, Y.; Coull, B.; Schwartz, J.; Koutrakis, P. A novel calibration approach of MODIS AOD data to predict PM2.5 concentrations. Atmos. Chem. Phys. Discuss. 2011, 11, 9769–9795. [Google Scholar] [CrossRef]
  21. Yap, X.; Hashim, M. A robust calibration approach for PM10 prediction from MODIS aerosol optical depth. Atmos. Chem. Phys. Discuss. 2012, 12, 31483–31505. [Google Scholar] [CrossRef]
  22. Wu, Y.; Guo, J.; Zhang, X.; Tian, X.; Zhang, J.; Wang, Y.; Duan, J.; Li, X. Synergy of satellite and ground based observations in estimation of particulate matter in eastern China. Sci. Total Environ. 2012, 433, 20–30. [Google Scholar] [CrossRef] [PubMed]
  23. Geng, G.; Zhang, Q.; Martin, R.V.; van Donkelaar, A.; Huo, H.; Che, H.; Lin, J.; He, K. Estimating long-term PM2.5 concentrations in China using satellite-based aerosol optical depth and a chemical transport model. Remote Sens. Environ. 2015, 166, 262–270. [Google Scholar] [CrossRef]
  24. Zhang, Y.; Li, Z. Remote sensing of atmospheric fine particulate matter (PM2.5) mass concentration near the ground from satellite observation. Remote Sens. Environ. 2015, 160, 252–262. [Google Scholar] [CrossRef]
  25. Li, Z.; Zhang, Y.; Shao, J.; Li, B.; Hong, J.; Liu, D.; Li, D.; Wei, P.; Li, W.; Li, L. Remote sensing of atmospheric particulate mass of dry PM2.5 near the ground: Method validation using ground-based measurements. Remote Sens. Environ. 2016, 173, 59–68. [Google Scholar] [CrossRef]
  26. Stewart Fotheringham, A.; Charlton, M.; Brunsdon, C. The geography of parameter space: An investigation of spatial non-stationarity. Int. J. Geogr. Inform. Syst. 1996, 10, 605–627. [Google Scholar] [CrossRef]
  27. Zhao, N.; Yang, Y.; Zhou, X. Application of geographically weighted regression in estimating the effect of climate and site conditions on vegetation distribution in Haihe catchment, China. Plant Ecol. 2010, 209, 349–359. [Google Scholar] [CrossRef]
  28. Wang, Z.; Chen, L.; Tao, J.; Zhang, Y.; Su, L. Satellite-based estimation of regional particulate matter (PM) in Beijing using vertical-and-RH correcting method. Remote Sens. Environ. 2010, 114, 50–63. [Google Scholar] [CrossRef]
  29. Zhang, T.; Gong, W.; Zhu, Z.; Sun, K.; Huang, Y.; Ji, Y. Semi-physical estimates of national-scale PM10 concentrations in China using a satellite-based geographically weighted regression model. Atmosphere 2016, 7, 88. [Google Scholar] [CrossRef]
  30. Kloog, I.; Melly, S.J.; Ridgway, W.L.; Coull, B.A.; Schwartz, J. Using new satellite based exposure methods to study the association between pregnancy PM2.5 exposure, premature birth and birth weight in massachusetts. Environ. Health 2012, 11, 40. [Google Scholar] [CrossRef] [PubMed]
  31. China Environmental Monitoring Center. Available online: http://113.108.142.147:20035/emcpublish/ (accessed on 6 December 2016).
  32. Determination of Atmospheric Articles PM10 and PM2.5 in Ambient Air by Gravimetric Method. Available online: http://english.mep.gov.cn/standards_reports/standards/Air_Environment/air_method/201111/t20111101_219390.htm (accessed on 6 December 2016).
  33. Remer, L.A.; Kaufman, Y.; Tanré, D.; Mattoo, S.; Chu, D.; Martins, J.V.; Li, R.-R.; Ichoku, C.; Levy, R.; Kleidman, R. The MODIS aerosol algorithm, products, and validation. J. Atmos. Sci. 2005, 62, 947–973. [Google Scholar] [CrossRef]
  34. Levy, R.C.; Remer, L.A.; Mattoo, S.; Vermote, E.F.; Kaufman, Y.J. Second-generation operational algorithm: Retrieval of aerosol properties over land from inversion of moderate resolution imaging spectroradiometer spectral reflectance. J. Geophys. Res. Atmos. 2007, 112. [Google Scholar] [CrossRef]
  35. Chu, D.; Kaufman, Y.; Ichoku, C.; Remer, L.; Tanré, D.; Holben, B. Validation of MODIS aerosol optical depth retrieval over land. Geophys. Res. Lett. 2002, 29. [Google Scholar] [CrossRef]
  36. Engel-Cox, J.A.; Holloman, C.H.; Coutant, B.W.; Hoff, R.M. Qualitative and quantitative evaluation of MODIS satellite sensor data for regional and urban scale air quality. Atmos. Environ. 2004, 38, 2495–2509. [Google Scholar] [CrossRef]
  37. Ma, Z.; Hu, X.; Sayer, A.M.; Levy, R.; Zhang, Q.; Xue, Y.; Tong, S.; Bi, J.; Huang, L.; Liu, Y. Satellite-based spatiotemporal trends in PM2.5 concentrations: China, 2004–2013. Environ. Health Perspect. 2015, 124. [Google Scholar] [CrossRef] [PubMed]
  38. You, W.; Zang, Z.; Zhang, L.; Li, Y.; Pan, X.; Wang, W. National-scale estimates of ground-level PM2.5 concentration in China using geographically weighted regression based on 3 km resolution MODIS AOD. Remote Sens. 2016, 8, 184. [Google Scholar] [CrossRef]
  39. NASA LAADS MODIS. Available online: http://ladsweb.nascom.nasa.gov/ (accessed on 6 December 2016).
  40. Sayer, A.; Hsu, N.; Bettenhausen, C.; Jeong, M.J. Validation and uncertainty estimates for MODIS Collection 6 “Deep Blue” aerosol data. J. Geophys. Res. Atmos. 2013, 118, 7864–7872. [Google Scholar] [CrossRef]
  41. Sayer, A.; Munchak, L.; Hsu, N.; Levy, R.; Bettenhausen, C.; Jeong, M.J. Modis collection 6 aerosol products: Comparison between aqua’s E-deep blue, dark target, and “merged” data sets, and usage recommendations. J. Geophys. Res. Atmos. 2014, 119, 13965–13989. [Google Scholar] [CrossRef]
  42. Zhang, T.; Liu, G.; Zhu, Z.; Gong, W.; Ji, Y.; Huang, Y. Real-time estimation of satellite-derived PM2.5 based on a semi-physical geographically weighted regression model. Int. J. Environ. Res. Public Health 2016, 13, 974. [Google Scholar] [CrossRef] [PubMed]
  43. CFS NCEP Reanalysis Meteorological Datasource. Available online: http://cfs.ncep.noaa.gov/ (accessed on 6 December 2016).
  44. Wang, Z.; Liu, C.; Huete, A. From AVHRR-NDVI to MODIS-EVI: Advances in vegetation index research. Acta Ecol. Sin. 2002, 23, 979–987. [Google Scholar]
  45. Weier, J.; Herring, D. Measuring Vegetation (NDVI and EVI). Available online: http://earthobservatory.nasa.gov/Features/MeasuringVegetation/ (accessed on 6 December 2016).
  46. NASA Aura OMI. Available online: http://disc.sci.gsfc.nasa.gov/Aura/data-holdings/OMI/omno2_v003.shtml (accessed on 6 December 2016).
  47. Zhang, Q.; Geng, G.; Wang, S.; Richter, A.; He, K. Satellite remote sensing of changes in no X emissions over China during 1996–2010. Chin. Sci. Bull. 2012, 57, 2857–2864. [Google Scholar] [CrossRef]
  48. Zheng, Y.; Zhang, Q.; Liu, Y.; Geng, G.; He, K. Estimating ground-level PM2.5 concentrations over three megalopolises in China using satellite-derived aerosol optical depth measurements. Atmos. Environ. 2016, 124, 232–242. [Google Scholar] [CrossRef]
  49. Zhang, T.; Zhu, Z.; Gong, W.; Xiang, H.; Fang, R. Characteristics of fine particles in an urban atmosphere—Relationships with meteorological parameters and trace gases. Int. J. Environ. Res. Public Health 2016, 13, 807. [Google Scholar] [CrossRef] [PubMed]
  50. Zhang, T.; Zhu, Z.; Gong, W.; Xiang, H.; Li, Y.; Cui, Z. Characteristics of ultrafine particles and their relationships with meteorological factors and trace gases in Wuhan, central China. Atmosphere 2016, 7, 96. [Google Scholar] [CrossRef]
  51. Song, W.; Jia, H.; Huang, J.; Zhang, Y. A satellite-based geographically weighted regression model for regional PM2.5 estimation over the pearl river delta region in China. Remote Sens. Environ. 2014, 154, 1–7. [Google Scholar] [CrossRef]
  52. Rodriguez, J.D.; Perez, A.; Lozano, J.A. Sensitivity analysis of K-fold cross validation in prediction error estimation. IEEE Trans. Pattern Anal. Mach. Intell. 2010, 32, 569–575. [Google Scholar] [CrossRef] [PubMed]
  53. China, M. Ambient Air Quality Standards. GB 3095-2012; China Environmental Science Press: Beijing, China, 2012. [Google Scholar]
  54. Ma, Z.; Hu, X.; Huang, L.; Bi, J.; Liu, Y. Estimating ground-level PM2.5 in China using satellite remote sensing. Environ. Sci. Technol. 2014, 48, 7436–7444. [Google Scholar] [CrossRef] [PubMed]
  55. Quan, J.; Zhang, Q.; He, H.; Liu, J.; Huang, M.; Jin, H. Analysis of the formation of fog and haze in North China Plain (NCP). Atmos. Chem. Phys. 2011, 11, 8205–8214. [Google Scholar] [CrossRef]
  56. Tao, M.; Chen, L.; Su, L.; Tao, J. Satellite observation of regional haze pollution over the North China Plain. J. Geophys. Res. Atmos. 2012, 117. [Google Scholar] [CrossRef]
  57. World Health Organization. Air Quality Guidelines: Global Update 2005: Particulate Matter, Ozone, Nitrogen Dioxide, and Sulfur Dioxide; World Health Organization: Geneva, Switzerland, 2006. [Google Scholar]
  58. Huang, J.; Minnis, P.; Chen, B.; Huang, Z.; Liu, Z.; Zhao, Q.; Yi, Y.; Ayers, J.K. Long-range transport and vertical structure of Asian dust from Calipso and surface measurements during PACDEX. J. Geophys. Res. Atmos. 2008, 113. [Google Scholar] [CrossRef]
Figure 1. Spatial distribution of PM2.5 monitoring stations (solid red dots) from which data were gathered in this study.
Figure 1. Spatial distribution of PM2.5 monitoring stations (solid red dots) from which data were gathered in this study.
Ijerph 13 01215 g001
Figure 2. Histograms and descriptive statistics for PM2.5, AOD, Planetary Boundary Layer Height (PBLH), wind speed, surface relative humidity (RH), surface temperature, surface air pressure, NO2 column density, and EVI in the model fitting.
Figure 2. Histograms and descriptive statistics for PM2.5, AOD, Planetary Boundary Layer Height (PBLH), wind speed, surface relative humidity (RH), surface temperature, surface air pressure, NO2 column density, and EVI in the model fitting.
Ijerph 13 01215 g002
Figure 3. Scatter plot of model fitting and cross-validation for the GWR model: results of classical GWR model with vertical corrected AOD and meteorological parameters (a); results of GWR model only adding NO2 (b); results of GWR model only adding EVI (c); and results of new GWR model adding NO2 and EVI (d). The solid line and dotted line are the regression line, and y = x reference line, respectively.
Figure 3. Scatter plot of model fitting and cross-validation for the GWR model: results of classical GWR model with vertical corrected AOD and meteorological parameters (a); results of GWR model only adding NO2 (b); results of GWR model only adding EVI (c); and results of new GWR model adding NO2 and EVI (d). The solid line and dotted line are the regression line, and y = x reference line, respectively.
Ijerph 13 01215 g003
Figure 4. Estimations of annual averaged AOD-derived PM2.5 in 2014 (a) and 2015 (b) when corresponding dataset values were available.
Figure 4. Estimations of annual averaged AOD-derived PM2.5 in 2014 (a) and 2015 (b) when corresponding dataset values were available.
Ijerph 13 01215 g004
Table 1. Relevant Level 2 SDS titles and contents for Aqua MODIS AOD at 550 nm.
Table 1. Relevant Level 2 SDS titles and contents for Aqua MODIS AOD at 550 nm.
DataSourceTemporal ResolutionSpatial ResolutionSpatial Resolution after Resampling
PM2.5Ground-level Measurement1 h--
DT-AODAqua-MODIS1 day3 km3 km
DB-AODAqua-MODIS1 day10 km3 km
Meteorological ParametersNCEP Reanalysis6 h100 km3 km
NO2Aura-OMI1 day25 km3 km
EVIAqua-MODIS16 days1 km3 km
DT: Dark Target; DB: Deep Blue; EVI: Enhanced Vegetation Index; AOD: Aerosol Optical Depth; MODIS: Moderate Resolution Imaging Spectroradiometer; NCEP: National Centers for Environmental Prediction; OMI: Ozone Measuring Instrument; SDS: Safety data sheet.
Int. J. Environ. Res. Public Health EISSN 1660-4601 Published by MDPI AG, Basel, Switzerland RSS E-Mail Table of Contents Alert
Back to Top