Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Open Access

Peer-reviewed

Research Article

Estimation of ground-level PM2.5 concentration using MODIS AOD and corrected regression model over Beijing, China

Abstract

To establish a new model for estimating ground-level PM2.5 concentration over Beijing, China, the relationship between aerosol optical depth (AOD) and ground-level PM2.5 concentration was derived and analysed firstly. Boundary layer height (BLH) and relative humidity (RH) were shown to be two major factors influencing the relationship between AOD and ground-level PM2.5 concentration. Thus, they are used to correct MODIS AOD to enhance the correlation between MODIS AOD and PM2.5. When using corrected MODIS AOD for modelling, the correlation between MODIS AOD and PM2.5 was improved significantly. Then, normalized difference vegetation index (NDVI), surface temperature (ST) and surface wind speed (SPD) were introduced as auxiliary variables to further improve the performance of the corrected regression model. The seasonal and annual average distribution of PM2.5 concentration over Beijing from 2014 to 2016 were mapped for intuitively analysing. Those can be regarded as important references for monitoring the ground-level PM2.5 concentration distribution. It is obviously that the PM2.5 concentration distribution over Beijing revealed the trend of “southeast high and northwest low”, and showed a significant decrease in annual average PM2.5 concentration from 2014 to 2016.

Citation: Xu X, Zhang C (2020) Estimation of ground-level PM2.5 concentration using MODIS AOD and corrected regression model over Beijing, China. PLoS ONE 15(10): e0240430. https://doi.org/10.1371/journal.pone.0240430

Editor: Zongbo Shi, University of Birmingham, UNITED KINGDOM

Received: August 22, 2019; Accepted: September 25, 2020; Published: October 13, 2020

Copyright: © 2020 Xu, Zhang. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All relevant data are within the manuscript and its Supporting Information files.

Funding: The author(s) received no specific funding for this work.

Competing interests: The authors have declared that no competing interests exist.

Introduction

With vibrant development of economy and rapid improvement of industry, environmental issues have gradually become a focus of attention. In recent years, the problem of haze has become outstanding, and the main reason for this is the absorption and scattering of PM2.5 (particulate matter with aerodynamic diameters less than or equal to 2.5 μm) in visible light intensity. In addition to affecting the human daily life and economical production, PM2.5 also contributes to chronic respiratory, pulmonary, cardiovascular, and other human diseases [1, 2]. The reduction of PM2.5 concentration has been a matter of great urgency all over the world.

The real-time monitoring of PM2.5 concentration plays a vital role in PM2.5 concentration reduction and environmental protection. Monitoring methods of PM2.5 concentration can be generally divided into two categories: ground stations monitoring and satellite remote sensing image estimating [3]. The former obtains environmental pollution information by setting up monitoring stations on the earth surface. This category can provide accurate monitoring of PM2.5 concentration, but the spatial distribution of monitoring stations is uneven and the equipment is expensive, making it difficult to obtain a wide range distribution of PM2.5 concentration. The latter can acquire continuous, large scale atmospheric aerosol information through remote sensing data, which can be converted to the PM2.5 concentration and distribution of the earth surface [4–6]. This category is more comprehensive and can greatly reduce the cost of research. So building a high-precision regression model between satellite based atmospheric aerosol information and ground measured PM2.5 concentration has great practical significance.

Griggs et al. discovered that ERTS-1 satellite data provided the possibility of monitoring atmospheric AOD [7]. Meanwhile, the monotonic linear relationship between the radiation values of infrared and visible channels and AOD was confirmed theoretically. Since Moderate Resolution Imaging Spectroradiometer (MODIS) sensors were put into use, AOD inversion algorithms and products have been greatly developed. It pushed the aerosol inversion research from ocean to land which supports most human activities. Dark target (DB) and Deep blue (DT) algorithms have gradually become mature AOD inversion algorithms over land through continuous improvement and development. Butt et al. adopted the DB algorithm AOD products of the Terra and Aqua satellites from 2000 to 2013 to analyse changes of AOD over Arabia, and concluded that the Aqua satellite’s AOD data were closer to the AOD results measured by ground stations than the Terra satellite’s AOD data [8]. Bilal et al. utilized AOD data from Aerosol Robotic Network (AERONET) level 2.0 cloud screening and quality control data to evaluate the static Normalized Difference Vegetation Index (NDVI) pixel selection criteria of AOD which based on the combination of the DB, DT, and DT and DB (DTB) algorithms. The experimental results revealed that the proposed dynamic NDVI pixel selection criteria could improve the performance of DTB products and reduce uncertainty [9].

With the shrinkage of particulate matter index from PM10 to PM2.5 and new development of AOD products, researchers have exploited AOD data to estimate PM2.5 concentration. The main approaches generally include statistical models and physical-chemical models. Compared to the physical-chemical models which require complex analysis of light transmission process in atmosphere and are difficult to generalize, statistical ones can use a regression model fitted by the relationship between AOD and PM2.5 concentration to approximately estimate the PM2.5 concentration.

Statistical models can be divided into the linear regression model, mixed effect model, geographically weighted regression model (GWR), and some others. On the basis of measured AOD and mixed layer height data, Schäfer et al. derived the extinction coefficient of aerosol, and acquired the extinction efficiency of aerosol through a linear regression model [10]. The experimental results showed that the extinction coefficient of aerosol was significantly related to the surface PM concentration in winter conditions. To study the correlations between PM2.5 concentration and AOD in North China, Xin et al. established linear regression functions between ground-level PM2.5 concentration and ground-observed AOD and MODIS AOD, respectively [11]. The results indicated that the linear regression functions of different seasons are significantly different and this is because the aerosol types were affected by seasonal variations. Tai et al. analysed the relationship between PM2.5 and different meteorological variables using the multivariate regression model (MLR) from 1998 to 2008 in adjacent areas of the United States [12]. The experiments showed that the distribution of PM2.5 are influenced by disturbances of wind patterns due to climate change.

Lee et al. proposed a mixed effect model, which fully exploited remote sensing data to establish the relationships between daily PM2.5 concentration and AOD in the New Zealand region and suggested the daily relations between AOD and PM2.5 concentration through the statistical method [13]. Xie et al. developed a mixed effect model to invert daily PM2.5 concentration in Beijing, which taken the diurnal variation relationship of AOD and PM2.5 concentration into account [14]. This research concluded that high-resolution daily PM2.5 concentration maps can contribute to assessment of short-term and long-term PM2.5 exposure. Zheng et al. put forward a linear mixed effect model that integrated AOD measurement, meteorological parameters, and NO2 column density for the estimation of PM2.5 concentration in Beijing, Tianjin and Hebei, the Yangtze River Delta, and the Pearl River Delta [15]. As shown in the experimental results, the combination of different factors contributed to the accuracy of model, which provided the possibility of assessing PM2.5 concentration in areas with different pollution levels.

You et al. adopted the GWR model to estimate the concentration of PM2.5 concentration in China and introduced meteorological features as auxiliary variables into the model to improve the accuracy [16]. In the experiments, cross validated R2 reached 0.79. The experimental results demonstrated that the proposed approach could evaluate a large area of PM2.5 concentration distribution. Guo et al. proposed a geographic and temporal weighted regression model (GTWR), which took both the temporal and spatial variability into consideration [17]. The AOD measurements, meteorological factors, and land use variables were also employed for fitting different seasons. The experiments indicated that GTWR was superior to MLR and GWR but that it was influenced by the sampling frequency of PM2.5 data [17]. Continuous daily PM2.5 observations could improve the performance of the model.

The aforementioned methods, such as MLR, mixed effects mode, GWR, GTWR, established the empirical linear relationship between PM2.5 concentration and satellite based AOD, and successfully mapped the PM2.5 concentration distribution of the study area. However, adequate analysis of how the relevant factors, such as boundary layer height (BLH) and relative humidity (RH), effect the PM2.5 concentration from a theoretical point of view is also essential for estimation of PM2.5 concentration distribution.

In this work, we first analyse the relationship between ground-level PM2.5 concentration and AOD, BLH, RH, and get the conclusion that BLH and RH can effectively correct the relationship between AOD and PM2.5 concentration. Thereby, three kinds of more theoretical AOD-PM2.5 regression models are established for PM2.5 concentration estimation, which are more interpretable. Then, NDVI, ST and SPD are introduced as auxiliary variables to further improve the performance of the corrected regression model. The modified AOD-PM2.5 concentration regression models were established separately according to the season and used to estimate the seasonal average PM2.5 concentration over Beijing, China during 2014 to 2016. The seasonal and annual PM2.5 concentration distribution over Beijing were mapped using the established models for better analysing the geographic distribution and annual changes of PM2.5 concentration.

Materials and methods

Beijing, located at 39.4°-41.6° N, 115.7°-117.4° E, is an important economic and political center of China. With the rapid development of the economy, Beijing has become one of the cities heavily affected by haze, especially in spring and winter. PM2.5 has a serious impact on urban traffic and people’s life. In this section, MODIS AOD and NDVI products, station-measured PM2.5 data, four kinds of meteorological data are described in this section (seen in Table 1).

thumbnail
Download:
Table 1. Descriptive statistics of PM2.5 measurements, MODIS AOD product, NDVI, and meteorological data during the study period (2014 to 2016).

https://doi.org/10.1371/journal.pone.0240430.t001

PM2.5 measurements

Hourly station-measured PM2.5 concentration data observed at 15 uniformly-distributed air quality monitoring station in Beijing, which were collected from the Beijing Municipal Environmental Monitoring Center (http://zx.bjmemc.com.cn/) [18, 19], were used for the experiments. The location of all the 15 air quality monitoring stations in Beijing is shown in Fig 1. and the geographical coordinates are listed in S1 Table. We averaged the measured PM2.5 concentration between 13:00 pm and 14:00 pm local time to match the overpass time of the MODIS Aqua satellite (about 13:30 pm local time). The study period was from January 1, 2014, to December 31, 2016, spanning a total of 1096 days.

thumbnail
Download:
Fig 1. The location of 15 air quality monitoring stations in Beijing.

https://doi.org/10.1371/journal.pone.0240430.g001

MODIS AOD and NDVI products

MODIS onboard the NASA Aqua satellite has been in operation since 2002. It provides retrieval products of aerosol and cloud properties with nearly daily global coverage. The MODIS data used in this study were the MYD04 Level 2 aerosol products with a resolution of 10 km and MYD13 monthly NDVI products with a resolution of 1 km [9]. The overpass time of the MODIS Aqua satellite is about 13:30 pm local time. The MODIS DT aerosol algorithms produce two separate products with different resolutions, with spatial resolutions of 3km and 10km respectively. However, it performs unsatisfactorily in urban regions. The DB algorithm performs better than the DT algorithm over urban areas, and it can provide better land coverage over both dark and bright surfaces, which is more reasonable for PM2.5 concentration distribution research in Beijing [20–22].

ECMEF ERA interim meteorological data

The ERA-Interim data provided by the European Centre for Medium-Range Weather Forecasts (ECMWF) are real-time updated global atmospheric reanalysis data (https://www.ecmwf.int/en/forecasts/datasets/reanalysis-datasets/era-interim). The meteorological parameters used in this study are BLH and RH at 100m vertical direction, surface temperature and wind speed at 14:00 pm local time [23]. The spatial resolution of the ERA-Interim reanalysis data used in this study was approximately 0.125°.

AOD-PM2.5 corrected regression model

Because some parameters used to describe atmospheric physical conditions, such as air pressure, atmospheric temperature, and atmospheric humidity, change much more in the vertical direction than the horizontal direction [24], it is often assumed that the atmosphere has a structure in which the horizontal direction is uniform and the vertical direction is layered. When considering a single homogeneous atmospheric layer containing spherical aerosol particles, the mass concentration at the surface can be represented as (1) in which ρ denotes the density of aerosol particles (g/m3) and ndry(r) denotes the particle size distribution spectrum under dry conditions. Generally, AOD is the integral of the extinction coefficient of aerosol particles in the vertical direction of the atmosphere with aerosol scale height H [25] (2) where σext,amb denotes the aerosol extinction coefficient under ambient conditions. It can be expressed as , where Qext,amb denotes the particulate matter extinction efficiency under ambient conditions and namb(r) denotes the particle size distribution under ambient conditions. Thus, at a given wavelength λ, the aerosol optical thickness from the ground to the height H can be expressed as [26] (3)

As we all know, different aerosol particles have different hygroscopic properties, e.g., water-soluble particles and organic aerosol particles have distinct hygroscopic properties. Thus, aerosol particles with the same mass concentration but different composition show different extinction characteristics under different humidity conditions. That is to say, the correlation between the extinction coefficient and the mass concentration changes with humidity. Therefore, to reduce the uncertainty introduced by the hygroscopic growth of the particles, it is necessary to convert the AOD obtained by satellite remote sensing into the mass concentration of dry particles.

A hygroscopic growth factor f(RH), which represents the ratio between these (size-distribution integrated) extinction efficiencies, is required to convert particulate matter extinction efficiency under ambient conditions to under dry conditions. Then, Eq (3) can be converted to the following [26, 27] (4)

Furthermore, to rewrite the above formula, the size-distribution integrated extinction efficiency 〈Qext〉 and effective radius reff can be introduced: (5) (6)

Hence, the relationship between AOD and near-surface PM mass concentration is finally derived as follows [25, 28] (7)

As the aerosol scale height H is an assumed equivalent height under ideal conditions, it is difficult to measure directly when the near-surface aerosol extinction coefficient is unknown. In the planetary boundary layer, the obvious turbulent motion causes strong atmospheric mixing, resulting in a uniform vertical distribution of aerosols. In practice, the BLH has a similar physical meaning to the aerosol scale height H and is often used instead of H.

According to the above relationship between AOD and PM, it can be inferred that if the AOD is corrected by the factors BLH and f(RH), the corrected AOD should be expected to obtain a better correlation with PM.

Here, a time-space matched MODIS AOD-PM2.5 data pair and a corresponding BLH and f(RH) were used to construct different linear regression models in different seasons [26]. There were 4 types of models: (8) (9) (10) (11) where a and b are model regression coefficients. Eq (8) represents a simple, original regression model, Eq (9) represents a model corrected by BLH, Eq (10) represents a model corrected by RH, and Eq (11) represents a model jointly corrected by BLH and RH.

In this study, NDVI, ST and SPD are introduced as auxiliary variables to further improve the performance of the corrected regression model for estimating ground-level PM2.5 concentration. 4 types of modified corrected regression models can be described as (12) (13) (14) (15)

Those modified corrected regression models are fitted separately according to the season and used to estimate the seasonal average PM2.5 concentration over Beijing, China during 2014 to 2016. For model evaluation, we assessed the model performance by calculating the coefficient of determination R2 and root mean squared error (RMSE) between estimated and measured PM2.5 concentration in different seasons: (16) (17) where fi and yi are the predicted PM2.5 and the measured PM2.5 concentration of the ith sample, respectively, and N is the total number of AOD-PM2.5 pair samples.

Fig 2 shows the flow diagram of the development process of the modified corrected AOD-PM2.5 model over Beijing. Firstly, BLH, RH, ST and SPD are interpolated to the same resolution as MODIS DB AOD by means of bilinear interpolation. NDVI is resized to the same resolution as MODIS DB AOD by means of nearest neighbor. Then, all variables located at 15 air quality monitoring stations are used to construct a AOD-PM2.5 modified regression model according to the season in every year. Finally, the seasonal and annual PM2.5 concentration distribution map is calculated by the AOD-PM2.5 modified regression model, which can directly reflect the PM2.5 distribution of Beijing.

thumbnail
Download:
Fig 2. Flow diagram for developing the modified corrected AOD-PM2.5 model in Beijing.

https://doi.org/10.1371/journal.pone.0240430.g002

Results and discussions

In our study, the correlation between estimated and measured PM2.5 concentration using aforementioned models is evaluated. R2 between estimated and measured PM2.5 concentration obtained by different models according to the season during 2014-2016 is tabulated in Table 2. f(RH) is defined empirically as f(RH) = 1/(1 − RH/100).

thumbnail
Download:
Table 2. R2 between estimated and measured PM2.5 concentration obtained by different models according to the season during 2014-2016.

https://doi.org/10.1371/journal.pone.0240430.t002

In Table 2, the R2 of model I can be interpreted as the correlation between measured PM2.5 concentration and MODIS DB AOD. Generally, in spring and autumn, the estimated PM2.5 concentration had closer correlation with the measured PM2.5 concentration. This may be a result of more uniform near-surface aerosol composition, as well as lower BLH, which limits aerosol diffusion in Beijing during spring and autumn.

According to Eq (7), BLH and f(RH) corrections could improve the AOD-PM2.5 correlation theoretically. The BLH corrected model was established on the basis of this. In Table 2, The R2 of model II can be interpreted as the correlation between measured PM2.5 concentration and BLH corrected AOD. Model correlations were affected differently in different seasons. The R2 in spring had a large increase after BLH correction, from 0.72 to 0.75 in 2014, from 0.69 to 0.71 in 2015, from 0.76 to 0.82. The more obvious advantages of BLH correction appears in autumn. The R2 has about 0.08-0.09 increase in different years, and reaches 0.81 in 2014 and 2015, 0.79 in 2016, demonstrating the effectiveness of BLH correction in autumn. For winter, the BLH corrected model achieved a slight improvement in 2014 and 2015 and a significant improvement in 2016. However, the R2 of the BLH corrected model had a decrease in summer, which because higher BLH cannot limit aerosol diffusion, as well as the poor quality of MODIS AOD products in summer because of cirrocumulus and rainfall.

The R2 of model III can be interpreted as the correlation between measured PM2.5 concentration and RH corrected AOD. With RH correction separately (Model III), a obvious increase of R2 are appeared in summer, 2016, from 0.62 to 0.73. However, the results change little in other times, and the fluctuation range of R2 is within 0.02. When considering BLH and f(RH) correction simultaneously (as Model IV), the results are very close to Model II.

When introduced NDVI, ST and SPD as auxiliary variables, the R2 of Model V-VIII in all seasons have about 2% increase correspondingly in most situations. Fig 3 show the scatter plots and regression results between estimated and measured PM2.5 concentration in the different seasons with the best results among all kinds of modified corrected models. In total, the correlation between estimated and measured PM2.5 concentration is more higher in spring and autumn. The R2 are 0.77, 0.74 and 0.84 in spring, 0.69, 0.75 and 0.74 in summer, 0.82, 0.82 and 0.80 in autumn, 0.70, 0.73 and 0.82 in winter during 2014 to 2016 respectively.

thumbnail
Download:
Fig 3. Scatter plots between estimated and measured PM2.5 concentration in different seasons during 2014 to 2016.

(a) Spring, 2014, Model VIII. (b) Spring, 2015, Model VIII. (c) Spring, 2016, Model VI. (d) Summer, 2014, Model VIII. (e) Summer, 2015, Model VI. (f) Summer, 2016, Model VII. (g) Autumn, 2014, Model VI. (h) Autumn, 2015, Model VI. (i) Autumn, 2016, Model VI. (j) Winter, 2014, Model V. (k) Winter, 2015, Model V. (l) Winter, 2016, Model VIII.

https://doi.org/10.1371/journal.pone.0240430.g003

In our study, even though the complicated PM2.5 concentration distribution in Beijing, the modified corrected AOD-PM2.5 models achieved higher model fitting R2 values than those in previous studies, e.g., an observation-based algorithm that considers the effect of the main aerosol characteristics applied to the Beijing-Tianjin-Hebei region with R2 of 0.70 [3], the GWR model applied to the whole China mainland with an overall R2 of 0.64 [5], an improved model applied to the Beijing-Tianjin-Hebei region with an R2 of 0.77 [15], satellite-driven PM2.5 models with VIIRS nighttime data applied to the Beijing-Tianjin-Hebei region with R2 of 0.75 [19], and NAQPMS data incorporated to MODIS data applied to the Beijing-Tianjin-Hebei region from January to December 2017 with seasonal R2 values of 0.75, 0.62, 0.80, and 0.78 in the spring, summer, autumn, and winter, respectively [22].

Table 3 is comparison between estimated and measured seasonal average PM2.5 concentration of 15 monitoring stations from 2014 to 2016. Minor difference reflects the good performance of the established models in each season. Table 4. is the comparison of regional average PM2.5 concentration from 2014 to 2016 over Beijing. In term of interannual seasonal change, PM2.5 concentration in summer are more lower the other seasons because of a great deal of rainfall. PM2.5 concentration in spring and winter is general higher than other seasons because of low rainfall and home-heating. In term of annual change of PM2.5 concentration during 2014 to 2016, it witnesses a downward trend in the same season. The annual average PM2.5 concentration decrease from 75.1 μg/m3 to 62.0 μg/m3, about a 17% decrease, which confirms the effectiveness of a series of stringent clean air actions implemented by the Chinese government from 2013 to 2017 [29]. In 2016 winter, the PM2.5 concentration decreased to 59.4 μg/m3, which is inspiring.

thumbnail
Download:
Table 3. The comparison between estimated and measured seasonal average PM2.5 concentration of 15 stations in Beijing from 2014 to 2016.(μg/m3).

https://doi.org/10.1371/journal.pone.0240430.t003

thumbnail
Download:
Table 4. The comparison of average PM2.5 concentration from 2014 to 2016 in Beijing.(μg/m3).

https://doi.org/10.1371/journal.pone.0240430.t004

The spatial distribution of the seasonal and annual average PM2.5 concentration are mapped to investigate its spatial distribution characteristics over Beijing during 2014 to 2017, as shown in Fig 4. The measured average PM2.5 concentration is represented as a circle using the same color scheme. In Beijing, PM2.5 concentration spatial distribution reveals the trend of “southeast high and northwest low”. In spring, autumn and winter of 2014, autumn and winter of 2015, autumn of 2016, this trend is more obvious. PM2.5 concentration of the southeast of Beijing during these seasons are higher than 75 μg/m3. This conforms to the urbanization level and population distribution in Beijing. Human activities directly affect the PM2.5 concentration.

thumbnail
Download:
Fig 4. The seasonal average PM2.5 concentration distribution over Beijing during 2014 to 2016.

https://doi.org/10.1371/journal.pone.0240430.g004

Fig 5 is the annual average PM2.5 concentration distribution over Beijing during 2014 to 2016. An obvious decrease of PM2.5 concentration can be intuitively found.

thumbnail
Download:
Fig 5. The annual average PM2.5 concentration distribution over Beijing during 2014 to 2016.

https://doi.org/10.1371/journal.pone.0240430.g005

Although the satellite-derived PM2.5 monitoring method can provide larger spatial coverage than ground monitoring stations, the satellite has lower temporal coverage mainly due to bad observation conditions, such as clouds and fogginess. In our study, there were 477 model-valid days during 2014 to 2016. In addition, due to cloudy or foggy weather in Beijing giving rise to inefficient sampling frequency of available satellite observations, the AOD retrieval algorithm may be not valid. Thus other monitoring approaches with larger spatial coverage and smaller weather limitations should be developed.

Overall, the seasonal estimated PM2.5 concentration showed good consistency with those of ground measurements. We attempted to use relative humidity to correct MODIS AOD, but this brought little effect. Additional work should be done in this aspect.

Conclusion

This paper focused on analysing the influence of BLH and f(RH) on the linear relationship between AOD and PM2.5 concentration and attempted to evaluate the improvement provided by modified regression models. As aerosol scale height, BLH can effectively improve the AOD-PM2.5 correlation by transforming AOD into near-surface aerosol extinction coefficient. When introduced NDVI, ST and SPD, the R2 in all seasons have about 2% increase. The spatial and temporal PM2.5 concentration distribution characteristics over Beijing during 2014 to 2016 were well revealed by seasonal and annual PM2.5 concentration distribution maps. The results confirmed the effectiveness of China’s clean air actions implemented from 2013 to 2017.

Supporting information

S1 Table. Geographical coordinates of 15 air quality monitoring stations used in this survey in Beijing.

https://doi.org/10.1371/journal.pone.0240430.s001

(DOCX)

Acknowledgments

The authors would like to thank the reviewers and editors for their valuable comments and suggestions.

References

  1. 1. Li X, Sun Y, An Y, Wang R, Lin H, Liu M, et al. Air pollution during the winter period and respiratory tract microbial imbalance in a healthy young population in northeastern China. Environ Pollut. 2019;246:972–979. pmid:31126003
  2. 2. Dabass A, Talbott EO, Rager JR, Marsh GM, Venkat A, Holguin F, et al. Systemic inflammatory markers associated with cardiovascular disease and acute and chronic exposure to fine particulate matter air pollution (PM2.5) among US NHANES adults with metabolic syndrome. Environ Res. 2018;161:485–491. pmid:29223110
  3. 3. Lin C, Li Y, Yuan Z, Lau AK, Li C, Fung JC. Using satellite remote sensing data to estimate the high-resolution distribution of ground-level PM2.5. Remote Sens Environ. 2015; 156:117–128.
  4. 4. Huang T, Yu Y, Wei Y, Wang H, Huang W, Chen X. Spatial-seasonal characteristics and critical impact factors of PM2.5 concentration in the Beijing-Tianjin-Hebei urban agglomeration. PLoS ONE. 2018;13(9):e0201364. pmid:30235240
  5. 5. 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(13):7436–7444 pmid:24901806
  6. 6. Zhao R, Gu X, Xue B, Zhang J, Ren W. Short period PM2.5 prediction based on multivariate linear regression model. PLoS ONE. 2018; 13(7): e0201011. pmid:30048475
  7. 7. Griggs M. Measurements of atmospheric aerosol optical thickness over water using ERTS-1 data. J Air Pollut Control Assoc. 1975; 25(6):622–626. pmid:1141544
  8. 8. Butt MJ, Assiri ME, Ali MA. Assessment of AOD variability over Saudi Arabia using MODIS deep blue products. Environ Pollut. 2017; 231:143–153. pmid:28800483
  9. 9. Bilal M, Nichol JE. Evaluation of the NDVI-based pixel selection criteria of the MODIS C6 dark target and deep blue combined aerosol product. IEEE J Sel Top Appl Earth Observ Remote Sens. 2017; 10(8):3448–3453.
  10. 10. Schäfer K, Harbusch A, Emeis S, Koepke P, Wiegner M. Correlation of aerosol mass near the ground with aerosol optical depth during two seasons in Munich. Atmos Environ. 2008; 42(18):4036–4046.
  11. 11. Xin J, Zhang Q, Wang L, Gong C, Wang Y, Liu Z, et al. The empirical relationship between the PM2.5 concentration and aerosol optical depth over the background of North China from 2009 to 2011. Atmos Res. 2014; 138:179–188.
  12. 12. Tai AP, Mickley LJ, Jacob DJ. Correlations between fine particulate matter (PM2.5) and meteorological variables in the United States: Implications for the sensitivity of PM2.5 to climate change. Atmos Environ. 2010; 44(32):3976–3984.
  13. 13. Lee HJ, Liu Y, Coull BA, Schwartz J, Koutrakis P. A novel calibration approach of MODIS AOD data to predict PM2.5 concentrations. Atmos Chem Phys. 2011; 11(15):7991–8002.
  14. 14. Xie Y, Wang Y, Zhang K, Dong W, Lv B, Bai Y. Daily estimation of ground-level PM2.5 concentrations over Beijing using 3 km resolution MODIS AOD. Environ Sci Technol. 2015; 49(20):12280–12288. pmid:26310776
  15. 15. 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.
  16. 16. 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(3):184.
  17. 17. Guo Y, Tang Q, Gong DY, Zhang Z. Estimating ground-level PM2.5 concentrations in Beijing using a satellite-based geographically and temporally weighted regression model. Remote Sens Environ. 2017; 198:140–149.
  18. 18. Wang W, Mao F, Du L, Pan Z, Gong W, Fang S. Deriving hourly PM2.5 concentrations from Himawari-8 AODs over Beijing-Tianjin-Hebei in China. Remote Sens. 2017; 9(8):858.
  19. 19. Zhang X, Hu H. Improving satellite-driven PM2.5 models with VIIRS Nighttime Light data in the Beijing-Tianjin-Hebei region, China. Remote Sens. 2017; 9(9):908.
  20. 20. Eck TF, Holben BN, Reid JS, Xian P, Giles DM, Sinyuk A, et al. Observations of the interaction and transport of fine mode aerosols with cloud and or fog in northeast asia from aerosol robotic network and satellite remote sensing. J Geophys Res-Atmos. 2018; 123:5560–5587. pmid:32661496
  21. 21. Wei J, Li ZQ, Peng YR, Sun L. Modis collection 6.1 aerosol optical depth products over land and ocean: validation and comparison. Atmos Environ. 2019; 201:428–440.
  22. 22. Wang Q, Zeng Q, Tao J, Sun L, Zhang L, Gu T, et al. Estimating PM2.5 concentrations based on MODIS AOD and NAQPMS data over Beijing-Tianjin-Hebei. Sensors. 2019; 19(5):1207.
  23. 23. You W, Zang ZL, Pan XB, Zhang LF, Chen D. Estimating PM2.5 in Xi’an, China using aerosol optical depth: A comparison between the MODIS and MISR retrieval models. Sci Total Environ. 2015; 505:1156–1165. pmid:25466686
  24. 24. Geng G, Zhang Q, Martin RV, Donkelaar A, Huo H, Che H, et al. 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.
  25. 25. Hoff RM, Christopher SA. Remote sensing of particulate pollution from space: have we reached the promised land? J Air Waste Manage Assoc. 2009; 59(6):645–675.
  26. 26. Koelemeijer RBA, Homan CD, Matthijsen J. Comparison of spatial and temporal variations of aerosol optical thickness and particulate matter over Europe. Atmos Environ. 2006; 40(27):5304–5315.
  27. 27. Guo YJ, Feng N, Christopher SA, Kang P, Zhan FB, Hong S. Satellite remote sensing of fine particulate matter (PM2.5) air quality over Beijing using MODIS. Int J Remote Sens. 2014; 35(17):6522–6544.
  28. 28. Tsai TC, Jeng YJ, Chu DA, Chen JP, Chang SC. Analysis of the relationship between MODIS aerosol optical depth and particulate matter from 2006 to 2008. Atmos Environ. 2011; 45(27):4777–4788.
  29. 29. Q , Zheng YX, Tong D, Shao M, Wang SX, Zhang YH, et al. Drivers of improved PM2.5 air quality in China from 2013 to 2017. P NATL ACAD SCI USA. 2019; 116(49):24463–24469.