Estimating groundlevel PM_{10} using satellite remote sensing and groundbased meteorological measurements over Tehran
 Saeed Sotoudeheian^{1} and
 Mohammad Arhami^{1}Email author
DOI: 10.1186/s4020101401226
© Sotoudeheian and Arhami; licensee BioMed Central Ltd. 2014
Received: 19 February 2014
Accepted: 24 August 2014
Published: 7 September 2014
Abstract
Background and methodology
Measurements by satellite remote sensing were combined with groundbased meteorological measurements to estimate groundlevel PM_{10}. Aerosol optical depth (AOD) by both MODIS and MISR were utilized to develop several statistical models including linear and nonlinear multiregression models. These models were examined for estimating PM_{10} measured at the air quality stations in Tehran, Iran, during 20092010. Significant issues are associated with airborne particulate matter in this city. Moreover, the performances of the constructed models during the Middle Eastern dust intrusions were examined.
Results
In general, nonlinear multiregression models outperformed the linear models. The developed models using MISR AOD generally resulted in better estimate of groundlevel PM_{10} compared to models using MODIS AOD. Consequently, among all the constructed models, results of nonlinear multiregression models utilizing MISR AOD acquired the highest correlation with ground level measurements (R^{2} of up to 0.55). The possibility of developing a single model over all the stations was examined. As expected, the results were depreciated, while nonlinear MISR model repeatedly showed the best performance being able to explain up to 38% of the PM_{10} variability.
Conclusions
Generally, the models didn't competently reflect wide temporal concentration variations, particularly due to the elevated levels during the dust episodes. Overall, using nonlinear multiregression model incorporating both remote sensing and groundbased meteorological measurements showed a rather optimistic prospective in estimating groundlevel PM for the studied area. However, more studies by applying other statistical models and utilizing more parameters are required to increase the model accuracies.
Keywords
PM10 Particulate matter Remote sensing Aerosol optical depth AOD MODIS MISR Multivariable regression modelsBackground
Increasing levels of air pollutants has become a complex issue affecting public health and environment in various cities of the developing countries during the recent years [1]. Serious adverse health effects such as respiratory problems, cardiovascular and lung disease and other damaging effects on human health has been associated to the air pollutants [2][5]. Among different pollutants, particulate matter (PM), including PM_{10} and PM_{2.5} (PM with aerodynamic diameters less than 10 μm and 2.5 μm, respectively), raised thoughtful concerns regarding public health [6][10]. In order to effectively manage the pollutants and evaluate the efficiency of different control strategies, it is crucial to determine the pollutant levels and their variations in different environments [11] which is generally done through air pollution monitoring networks.
Although the groundlevel measurements are generally referred to as accurate methods, these measurements indicate the pollution concentrations of a small area around the monitoring stations. Consequently, the studies of air pollutants and their adverse effects are impeded by limited coverage and irregular distribution of monitoring stations at ground level [12]. In fact, achieving comprehensive pollutants coverage from ground based measurements is difficult due to the limited number of stations equipped with costly instruments [13]. Hence, researchers have been constantly seeking new methods to attain more comprehensive measurements.
In the past decades, innovations in the field of remote sensing techniques by satellites opened a new era for different measurements including air pollutants measurements. Particular attempts have been made in the satellites based remote sensing of PM concentration in the lower troposphere since the late 1970s. Several sensors measure the parameters associated to concentration of aerosol in the atmosphere [14].
The Moderate Resolution Imaging Spectroradiometer (MODIS) sensor on Terra and Aqua satellite and Multiangle Imaging SpectroRadiometer (MISR) on Terra sensors measure the particle abundance and their composition by determining Aerosol Optical Depth (AOD) with temperate spatial resolutions [15]. AOD which reflects optical characteristics of aerosols is also determined by other sensors such as Ozone Monitoring Instrument (OMI) on Aura and is measured by sunphotometer in the Aerosol Robotic Network (AERONET) network. The AOD measurements have been used in several studies to estimate PM_{2.5} or PM_{10} concentration at the ground level [16].
Using the measurements by satellite sensors to estimate groundlevel particulate concentration is a challenging endeavor. Several factors such as particle composition and physical properties affect the optical properties of particles and consequently influence the relationship between satellite data and PM_{10} concentration. Also upper air obstacles including dense cloud cover, and variations in vertical PM profile could disturb this relation, or lead to missed data in the pixels of interest. Hence, it is crucial to integrate other variables and build a model to estimate ground based particulate levels. In these models satellite based measurements are combined with other variables and compared to concentration data to estimate PM concentration and provide valuable information to establish effective air quality strategies and high accuracy predicting models. Two approaches have been implemented in modeling process to estimate PM_{10} concentration. First approach is utilizing the deterministic models requiring intensive data including the inventory of pollution sources, which can be difficult to quantify. Second approach, which is the focus of our study, utilizes statistical models to optimize the relationship between PM levels and independent variable [12].
Several studies have been conducted to develop a relationship between satellite AOD data and PM concentration at the ground surface. Examples of these studies, which were mainly carried to obtain reliable estimates of PM concentration, are presented in the following. Wang et al. [17] examined linear relationship between hourly PM_{2.5} concentration and AOD from MODIS at 7 stations in Jefferson County, Alabama (R = 0.7). The rather high correlation was found between monthly average of PM_{2.5} and AOD (R > 0.9) [17]. Some other similar works obtained relationships between PM_{2.5} and PM_{10} concentration and AOD with R^{2} range of 0.58 to 0.76 [18][20].
By advancing studies in the field of using remote sensing to estimate PM concentration, the structures of models changed from simple linear models to the more complex nonlinear models by using the other affecting parameters such as meteorological data. In this regard, Liu et al. [21] generate empirical model to estimate PM_{2.5} concentration at surface level using the AOD data from MISR sensor. Results show that their model can explain 48% of the variability in PM_{2.5} concentration. They found that several factors such as relative humidity, planetary boundary layer height, season and geographical characteristics of area can affect the relationship between PM_{2.5} and AOD data [21]. Liu et al. [15] continued their studies in 2007 and developed, two general linear regression models by using AOD from both MODIS and MISR sensors and compared their performances. The MISR model was able to explain 62% of variability in PM_{2.5} concentration, while MODIS model explained 51% of concentrations variability [15]. Vladutescu et al. [22] showed that incorporating variation of planetary boundary layer height in such models can improve the accuracy of the models. Also, they showed physical characteristics and hygroscopic of particles (relative humidity of atmosphere) have significant effect on models performance [22]. In the similar work, Koelemeijer et al. [23] advanced PMAOD relationship by incorporating effect of several factors such as boundary layer height and relative humidity on particles size. Average correlation coefficients between measured and modeled levels were 0.5 and 0.6 for PM_{10} and PM_{2.5}, respectively in rural and suburban regions [23]. Pelletier et al. [24] also improved the performance of linear model between AOD and PM concentration by adding auxiliary parameters, mainly meteorological variables. Their improved linear model could explain 76% of concentrations variability [24]. Vidot et al. [25] continued Pelletier et al. [24] general idea, and amended PM and AOD (from SeaWiFS imagery) relationship with effective meteorological information through a statistical approach. They obtained determination coefficient of 0.42 and 0.48 for PM10 and PM2.5 respectively [25]. Also, Gupta et al. [26] developed multiple regression models between meteorological parameters and AOD data from MODIS sensor. Results show that significant improvement in correlation coefficients which was obtained by multiple regression tool and increasing meteorological parameters number [26]. In a more recent study, Tian et al. [12] generated a semiempirical model in the regional scale to estimate hourly PM_{2.5} concentration. This model utilizes a modified AOD value based on boundary layer height and meteorological characteristics, which resulted in explaining 65% of PM_{2.5} variability [12]. Subsequently, Tian et al. [27] emphasized that the spatial and temporal variation of relationship between AOD and PM_{2.5} isn't clear so far due to: meteorological condition, land use, cloud contamination, station location, and particle size. Majority of these recent investigations were able to build models to estimate PM_{2.5} concentrations rather than PM_{10}.
These studies concurred on the capability of utilizing remote sensing as a powerful tool in predicting PM concentration at the ground surface especially for areas without a monitoring network. However, the results of most of these studies prove the need for more studies to increase the accuracy and reliability of the models by incorporating optimal parameters. The models and estimations accuracy was shown to vary for different regions of the world. Moreover, the ability of models to estimate high PM levels due to different phenomena such as dust storms, which could result in wide range of PM levels, has not been evaluated.
Tehran is one of the most polluted cities in the world, facing major issues raised by airborne particulate matter [28]. Many factors including growing populations, extensive transportation network, industrial emissions, and dust storm from deserts in neighboring countries such as Iraq and Saudi Arabia, affect the air quality situation in this metropolis. High PM levels also occasionally occur in this city's atmosphere due to different phenomena such as dust storms, which results in wide range of PM levels. Despite the significance of the airborne particulate problem and possibility of using AOD data to predict surface PM_{10} concentration, no such research has been carried out in this region.
In this study, a semiempirical equation was developed and examined to estimate the PM_{10} concentration in Tehran's stations by utilizing the AOD data from MODIS and MISR sensors. Initially, individual models were developed, validated and evaluated for each station. Subsequently, general models for the entire region were built and examined. Due to the lack of such research in this region, results of this study could help the future investigations, and strategy developments to control airborne PM.
Methods
Data extraction and processing
where C_{ ext } is extinction crosssectional area, which is a function of particle size and refractive index, m, and size distribution of particle, n(r) [21].
MODIS sensors are installed on Terra and Aqua satellite platforms designed to retrieve aerosol properties over land and ocean [14]. These sensors collect data in 36 channels every 12 days depending on the data location [29]. This temporal resolution of AOD data make them appropriate for air quality assessments [14]. MODIS sensors with a 2330 km viewing swath provide almost complete global coverage in one day [30]. The MODIS AOD data are derived at three wavelengths of 0.47, 0.66, and 2.1 μm via over land retrieval algorithm. Total AOD at 0.55 μm is calculated by solving an inversion problem using independent observations of spectral reflectance data derived in three wavelengths (0.47, 0.66 and 2.1 μm) [31],[32].
In this research AOD calculated at 0.55 μm from the Level2 of MODIS (collection 5), which reflects the concentration of pollutant, is utilized due to its better quality and higher resolution (~10 km spatial resolution). The estimated uncertainty of MODIS data for the level 2 over land is 0.05 ± 0.15AOD[32]. The MODIS sensors onboard the Terra and Aqua satellite overpass Tehran at approximately 11 a.m. and 13:30 p.m., local time, respectively. After evaluating the data obtained on Terra and Aqua satellite, AOD data recorded by MODIS on Aqua in 2009 and 2010 were utilized due to the appropriateness and frequency of the circuit with Iran's local stations (http://ladsweb.nascom.nasa.gov/).
MISR sensor is installed on Terra platform to study climate and environmental condition on a global scale. This sensor can provide images in nine view angles at four wavelength bands to retrieve aerosol information over land. MISR AOD data are recorded at a 17.6 km resolution with a temporal resolution of 2 to 9 days depending on the latitudes [15]. The AOD data by MISR which are obtained at 10:3011:30 a.m. of Tehran's local time due to the Terra platform orbiting schedule, were extracted for year 2009 and 2010 for the means of this study (http://eosweb.larc.nasa.gov).
The hourly groundlevel PM_{10} concentrations used for modeling were extracted for the time at which the satellite data are recorded (12 a.m. and 1 p.m. for MODIS & 10 a.m. and 11 a.m. for MISR). Average PM_{10} levels over these time spans were utilized in the models. Once the models were calibrated using the processed data throughout 2009, data for the first half of 2010 were used to validate the models.
Size distribution, particle composition and vertical profile of aerosol are important factors affecting the relationship between satellite data and PM_{10} concentration at the ground level. Therefore, these factors should be considered in the models by implementing appropriate and plausible variables [21]. Meteorological parameters such as wind speed, wind direction, temperature and relative humidity have significant effects on PM_{10} concentration and particle optical properties. Particle properties can affect the relationship between PM_{10} concentration and corresponding remotely sensed data. For example, changes in relative humidity (RH) or temperature can directly or indirectly alter the particles composition due to change in photochemical oxidation and condensation processes, which affect optical properties of particles.
Since the RH values could change particles composition and optical properties, this parameter was also incorporated in the models to improve their ability to estimates the PM_{10} concentration. Since hourly value of RH data weren't available for the studied period, daily value of RH reported from synoptic station were used. Another implemented parameter was Planetary Boundary Layer's Height (PBLH) which is the depth of the surface layer of the atmosphere. The fluid dynamic properties of this layer are directly influenced by contacting the planetary surface, and its height plays an important role in pollutants behavior in the atmosphere. In this study the PBLH were extracted from the Global Data Assimilation System (GDAS) files (http://www.ncdc.noaa.gov/). Spatial resolution of these data is 1^{ ° } × 1^{ ° } with a temporal resolution of 3 hours. PBLH data at 12:00 p.m. were used, which was the closest time to the AOD data recording time.
Quality Assurance and Quality Control (QA/QC) was performed on AOD and PM_{10} dataset. Initially by extracting the average and standard deviation (σ) of data, the data out of 3σ to 3σ from average were flagged. The flagged data were checked to see if they reflect a true event or they are outliers, which need to be eliminated. Since, there are no surface measurements of AOD parameters available, such as LIDAR measurements, during QA/QC procedure AOD and PM_{10} data were also evaluated with each other during the modeling period. In fact, AOD and PM_{10} values were patterned together, and days with extreme changes in each of AOD or PM_{10} values without considerable change in the other parameter were flagged as suspicions value and double checked for potentially being outliers. By applying this method about 1015% of data were excluded. This screening process performed on the dataset to lessen the effects of potential errors in AOD measurement by satellite and PM_{10} recorded at stations. However this would not completely eliminate the potential errors in such measurements, which could impose some extent of uncertainties to the modeling's results.
Regression models
Where T, W, Dir, RH, AOD, PBL are the temperature, wind speed, wind direction, relative humidity, aerosol optical depth and planetary boundary layer height parameters, respectively. α_{0} is intercept of general equation and α_{ i }s are the regression coefficients of the independent variables.
This equation incorporates physical interoperations of relation between meteorological parameters and particulate properties into the model as extensively described in previous studies [15],[21]. In Equation 6 it is assumed that vertical profile of aerosol is smooth and its concentration at different altitudes is correlated to the ground level PM concentration [12]. Due to nonlinear growth of particle size with increasing of relative humidity, exponential function of RH was used as well [21].
All the statistical analyses were performed by Statistical Analysis System (SAS) software (version 9.1). Statistical analysis performed on the data set included to fit linear and nonlinear multivariable regression models, and calculation of regression coefficients, R^{2} values and other statistics for all kind of models. In order to perform correlation analysis and check linear singlevariable regression model, simple correlation was performed and Pearson correlation coefficients (r) were calculated between AOD and PM_{10}. Also, for the regression analysis and fitting multiple variable regression models least square analysis was done by SAS to fit regression models.
Results and discussion
Data overview
Statistical overview of PM _{ 10 } and meteorological measurements during year 2009
Parameter  PM_{10, MODIS}$\left(\raisebox{1ex}{$\mathit{\mu g}$}\!\left/ \!\raisebox{1ex}{${\mathbf{m}}^{\mathbf{3}}$}\right.\right)$  PM_{10, MISR}$\left(\raisebox{1ex}{$\mathit{\mu g}$}\!\left/ \!\raisebox{1ex}{${\mathbf{m}}^{\mathbf{3}}$}\right.\right)$  Temp_{MODIS}(°C)  Wind_{MODIS}(m/s)  Temp_{MISR}(°C)  Wind_{MISR}(m/s)  RH (%)  PBLH (m) 

Average  72.27  76.70  21.04  4.41  21.04  4.09  38.42  1173.05 
STDEV  64.19  62.68  9.97  2.48  9.67  2.48  16.13  879.07 
MIN  4.51  14.62  0.58  0.00  0.25  0.00  9.50  98.60 
MAX  962.31  947.40  41.00  15.44  40.50  12.95  87.00  4000 
Result of linear singlevariable regression model
Parameter  R^{2}  

Aghdasiyeh  Golbarg  Poonak  Shahre Rey  
AOD (MODIS)  0.15  0.16  0.15  0.16 
AOD (MISR)  0.32  0.34  0.43  0.27 
Regression analysis
Regression coefficients for linear and nonlinear multivariable regression models
Model type  Sensor  α _{0}  α _{ AOD }  α _{ T }  α _{ W }  α _{ Dir }  α _{ RH }  α _{ PBLH }  R ^{2}  ${\mathit{R}}_{\mathbf{Adj}}^{\mathbf{2}}$  P  Value  RMSE 

Linear  MODIS  58.96  15.49  23.15  14.36  14.52  19.32  0.26  0.31  0.23  0.0011  25.58 
MISR  75.20  30.72  28.90  13.08  5.54  22.79  19.17  0.47  0.30  0.0400  19.71  
Nonlinear  MODIS  4.02  0.17  0.26  0.19  0.18  0.28  0.06  0.32  0.25  0.0006  0.33 
MISR  4.05  0.50  0.36  0.06  0.07  0.29  0.24  0.49  0.33  0.0290  0.29 
In the study region anthropogenic sources are the main reason of high level of PM concentration most days in a year. In addition, dust storms from western parts of Iran increase particles concentration in the dusty episodes [28]. According to this explanation, wind could have various effects on PM concentrations; in one aspect, wind could prevent stable condition in the region and cause pollutant to flush out and dilute and spread pollutants in wider region in height and area. So, it causes PM concentration to be low in the stations. In the other hand, it could resuspend and transport mineral dust of different size distributions to the study area and increase PM concentration at the surface. Also, due to the dust storm phenomena in our region, wind could transport large amount of dust in to the area and increase PM level in the selected stations. In our case the first situation was dominant and in most cases wind coefficient had negative sign (by increasing in wind speed value, PM concentration decrease), since data from urban stations rather close to the major pollutant sources (vehicular sources) were utilized.
High temperatures could be a sign for intensifying generation of the secondary pollutants due the photochemical activity, increasing the PM_{10} concentration at the surface level. On the other hand, the elevated PM levels could also occur during the cold periods of the year, such as during the occurrence of the inversion phenomenon in winter times. So both positive and negative coefficients were obtained for the independent variable of temperature. Negative signs were obtained for relative humidity, which represents the reverse effect of RH on AOD. Under high relative humidity condition (RH=>80%) hygroscopic particles (e.g. ammonium nitrate and ammonium sulfate) can grow into 210 times of their normal size, increasing the light extinction efficiencies of particle, while PM_{10} are measured at the surface stations under the controlled condition (RH=40%). Hence, the same AOD value at high relative humidity corresponds to lower particle dry mass compared to obtained value at low humidity [21].
Multivariable regression models for individual stations
Statistical Parameters for validation period of models
Station  MISR AOD  

Linear model  Nonlinear model  
R^{2}  Slope  Intercept  RMSE  MAE  Bias  R^{2}  Slope  Intercept  RMSE  MAE  Bias  
Aghdasiyeh  0.31  0.29  48.3  23.9  19.8  2  0.23  0.18  49.5  26.6  21.1  8.4 
Golbarg  0.16  0.14  55.7  25.1  20  7.7  0.14  0.1  56.2  26.4  21.1  10.4 
Poonak  0.22  0.26  44.6  15.9  13.8  0.2  0.23  0.25  44.3  15.7  13.5  1.1 
Shahr Rey  0.25  0.34  53.5  28.9  26.8  20.9  0.25  0.32  51.6  26.8  24.5  18.2 
Station  MISR AOD  
Linear model  Nonlinear model  
R ^{ 2 }  Slope  Intercept  RMSE  MAE  Bias  R ^{ 2 }  Slope  Intercept  RMSE  MAE  Bias  
Aghdasiyeh  0.41  0.45  41.7  19.2  15.5  9.4  0.51  0.41  42.9  16.8  13.1  6.6 
Golbarg  0.3  0.51  35.1  22  17.4  5.2  0.35  0.39  41.7  20  16.5  4.4 
Poonak  0.50  0.89  11.9  17.9  16.2  6.3  0.55  0.65  25.1  14.7  13  6.8 
Shahr Rey  0.17  0.41  60.1  36.3  34.1  25.2  0.30  0.64  46.4  32.9  30.1  26.1 
It could be inferred from MODIS sensor validation results, shown in Table 4, the developed models performed rather similarly in predicting PM_{10} concentration over all four stations except for Golbarg station. Although, models in Shahr Rey have a moderate R^{2} (0.25) but its RMSE, MAE and bias (the difference between average of measured and predicted concentrations) show high values of error in predicted concentrations. On the other hand, models developed by data from MISR sensor showed weaker performance at Golbarg and Shahr Rey stations (low R^{2} and high error). Since these stations are located in the regions near the central desert of Iran and satellite sensors encounter more uncertainties in AOD retrieval from bright surface. These uncertainties could diminish models accuracy in such stations.
Estimating PM _{ 10 } concentration during the dust episodes using AODs from MODIS and MISR sensors
Station  MODIS sensor  MISR sensor  

18 May  22 Jun  18 May  22 Jun  
Measured$\left(\raisebox{1ex}{$\mathit{\mu g}$}\!\left/ \!\raisebox{1ex}{${\mathbf{m}}^{\mathbf{3}}$}\right.\right)$  Predicted$\left(\raisebox{1ex}{$\mathit{\mu g}$}\!\left/ \!\raisebox{1ex}{${\mathbf{m}}^{\mathbf{3}}$}\right.\right)$  Measured$\left(\raisebox{1ex}{$\mathit{\mu g}$}\!\left/ \!\raisebox{1ex}{${\mathbf{m}}^{\mathbf{3}}$}\right.\right)$  Predicted$\left(\raisebox{1ex}{$\mathit{\mu g}$}\!\left/ \!\raisebox{1ex}{${\mathbf{m}}^{\mathbf{3}}$}\right.\right)$  Measured$\left(\raisebox{1ex}{$\mathit{\mu g}$}\!\left/ \!\raisebox{1ex}{${\mathbf{m}}^{\mathbf{3}}$}\right.\right)$  Predicted$\left(\raisebox{1ex}{$\mathit{\mu g}$}\!\left/ \!\raisebox{1ex}{${\mathbf{m}}^{\mathbf{3}}$}\right.\right)$  Measured$\left(\raisebox{1ex}{$\mathit{\mu g}$}\!\left/ \!\raisebox{1ex}{${\mathbf{m}}^{\mathbf{3}}$}\right.\right)$  Predicted$\left(\raisebox{1ex}{$\mathit{\mu g}$}\!\left/ \!\raisebox{1ex}{${\mathbf{m}}^{\mathbf{3}}$}\right.\right)$  
Linear  Nonlinear  Linear  Nonlinear  Linear  Nonlinear  Linear  Nonlinear  
Aghdasiyeh  55.80  56.30  56.50  80.90  69.00  65.80  62.40  39.10  53.20  93.60  91.10  83.00 
Golbarg  61.50      81.30  67.60  68.70  61.20  28.90  44.90  80.80  98.10  84.60 
Poonak  57.20  55.76  57.8  62.10  71.16  70.40  44.80  17.50  40.10  65.70  89.00  74.30 
Shahr Rey  79.40  66.55  67.8  40.80  72.91  67.30  84.00  64.80  71.30  70.00  110.90  102.30 
Multivariable regression models over all the stations
Statistical coefficients obtained from all the data
Statistical parameter  MODIS sensor  MISR sensor  

Linear  NonLinear  Linear  NonLinear  
R ^{2}  0.24  0.22  0.37  0.34 
${R}_{\mathrm{Adj}}^{2}$  0.21  0.20  0.34  0.30 
RMSE  22.62  0.32  21.70  0.32 
P  Value  <0.0001  <0.0001  <0.0001  <0.0001 
Statistical Parameters for validation period for models were developed in all stations
Type of model  MODIS sensor  MISR sensor  

R^{2}  Slope  Intercept  RMSE  MAE  Bias  R^{2}  Slope  Intercept  RMSE  MAE  Bias  
Linear model  0.21  0.19  55.3  26.5  21.3  1.3  0.3  0.6  34.3  23.7  19.8  11 
Nonlinear model  0.18  0.13  53.6  27.5  21  4.3  0.38  0.51  33.9  18.5  15.1  5.9 
Comparison between measured, and estimated levels by model obtained over all the stations for the new stations used for validation
Station  MODIS sensor  MISR sensor  

Measured$\left(\raisebox{1ex}{$\mathit{\mu g}$}\!\left/ \!\raisebox{1ex}{${\mathit{m}}^{\mathbf{3}}$}\right.\right)$  Predicted$\left(\raisebox{1ex}{$\mathit{\mu g}$}\!\left/ \!\raisebox{1ex}{${\mathit{m}}^{\mathit{3}}$}\right.\right)$  Measured$\left(\raisebox{1ex}{$\mathit{\mu g}$}\!\left/ \!\raisebox{1ex}{${\mathit{m}}^{3}$}\right.\right)$  Predicted$\left(\raisebox{1ex}{$\mathit{\mu g}$}\!\left/ \!\raisebox{1ex}{${\mathit{m}}^{\mathit{3}}$}\right.\right)$  
Linear  Nonlinear  Linear  Nonlinear  
Geophysic  57.97  60.21  59.86  60.98  36.62  47.00 
Park Roz  51.61  60.54  60.18  45.79  36.57  46.00 
Ostandary  54.23  60.50  60.00  73.26  36.84  48.00 
Shahrdari 4  38.79  63.80  62.49  38.98  37.10  48.56 
Shahrdari 11  59.77  60.50  60.20  70.90  36.24  46.65 
Shahrdari 16  65.70  60.34  60.11  69.28  36.00  47.00 
Shahrdari 19  84.27  60.38  60.14  87.99  36.04  46.36 
The results obtained in this study were rather in similar ranges of previous studies for other regions. Models develop by Liu et al. [21] using AOD data from MISR sensor were able to estimate PM_{2.5} concentration by correlation coefficient (R^{2}) of 48% in the eastern United States during the period of 2001. They continue their studies in 2007 and developed a linear regression model to predict PM_{2.5} concentration by using AOD data from MODIS and MISR with R^{2} of 51% and 62% respectively [15]. Tian et al. [12] developed a semiempirical model by considering AOD, meteorological and boundary layer height to improve model ability. Finally, their model could explain 65% of PM_{2.5} concentration variability at the ground surface [12]. In the current study we attempted to incorporate other parameters to improve the accuracy of modeling PM_{10} by AOD data. Several models based on data from each station and total data were developed, which could acquire up to 55% explanation of PM_{10} concentrations variability by nonlinear model constructed with MISR AOD data. It should be noted that most of previous studies were performed on PM_{2.5}, which generally correlates better with AOD compared to PM_{10}. Until now there wasn't any relevant study in the studied region, so these results could be a good basis for the future investigation to use remote sensing data to estimate groundlevel PM.
Conclusions
Several statistical models utilizing satellite based measurements were developed and their capabilities in predicting PM_{10} concentration at the ground surface were evaluated. Linear and nonlinear multiregression models were constructed to incorporate meteorological parameters in addition to MODIS and MISR AOD data. These models were examined for stations in Tehran, Iran. Despite the significance of airborne particulate problems and the need for examining new measurement techniques, no such studies had been conducted in this area. The possibility of developing a single model over all the stations was examined and its results were compared to the individual models for each station. Performances of the constructed models to estimate PM_{10} levels during the Middle Eastern dust intrusions were examined.
In general, results of MISR models had better correlation with groundlevel PM_{10} concentration compared to those of MODIS models. Nonlinear multiregression models also generally outperformed the linear models. Among all the constructed models, nonlinear multiregression models utilizing MISR AOD data resulted in the best estimates of groundlevel PM_{10} ( R^{2} of up to 0.55). Generally, the models didn't competently reflect wide temporal concentration variations, particularly due to the elevated levels during the dust episodes. However, like other periods, nonlinear models performed slightly better than linear models during the dust episodes. Applying a single model over all the stations depreciate the results, while nonlinear MISR model repeatedly showed the best performance being able to explain up to 38% of the PM_{10} variability. These models defined by using all data were not able to reflect the spatial variations of concentrations in the studied area. Overall, using nonlinear multiregression model incorporating both remote sensing and meteorological parameters showed a prospective in estimating groundlevel PM for the studied area. However, more studies by applying other statistical models and utilizing more parameters are required to increase the model accuracies.
Abbreviations
 PM:

Particulate Matter
 MODIS:

Moderate Resolution Imaging Spectroradiometer
 MISR:

Multiangle Imaging SpectroRadiometer
 AOD:

Aerosol Optical Depth
 AERONET:

AErosol RObotic NETwork
 OMI:

Ozone Monitoring Instrument
 AQCC:

Air Quality Control Company
 TEOM:

Tapered Element Oscillating Microbalance
 RH:

Relative Humidity
 PBLH:

Planetary Boundary Layer's Height
 GDAS:

Global Data Assimilation System
Declarations
Acknowledgements
The authors would like to give special thanks to the Tehran AQCC and Iran Meteorological Organization for providing the air pollution and meteorological data used in the current study. The authors express their special gratitude to Mr. Wasim Tayyeb for his useful help in extracting AOD data from the available files.
Authors’ Affiliations
References
 Air Quality Guidelines for Europe. Regional Office for Europe, Copenhagen; 2000.
 Kampa M, Castanas E: Human health effects of air pollution. Environ Pollut 2008, 151: 362367. 10.1016/j.envpol.2007.06.012View ArticleGoogle Scholar
 Air Quality Guidelines: Global Update 2005: Particulate Matter, Ozone, Nitrogen Dioxide, and Sulfur Dioxide. Regional Office for Europe, Copenhagen; 2006.
 Naddafi K, Hassanvand MS, Yunesian M, Momeniha F, Nabizadeh R, Faridi S, Gholampour A: Health impact assessment of air pollution in megacity of Tehran, Iran. Iran J Environ Health Sci Eng 2012, 9: 28. 10.1186/17352746928View ArticleGoogle Scholar
 Goudarzi G, Zallaghi E, Neissi A, Ahmadi Ankali K, Saki A, Babaei AK, Alavi N, Mohammadi MJ: Cardiopulmonary mortalities and chronic obstructive pulmonary disease attributed to ozone air pollution. Arch Hyg Sci 2013, 2: 6272.Google Scholar
 Atkinson RW, Anderson HR, Sunyer J, Ayres J, Baccini M, Vonk JM, Boumghar A, Forastiere F, Forsberg B, Touloumi G, Schwartz J, Katsouyanni K: Acute effects of particulate air pollution on respiratory admissions—results from APHEA 2 project. Am J Respir Crit Care Med 2001, 164: 18601866. 10.1164/ajrccm.164.10.2010138View ArticleGoogle Scholar
 Brunekreef B, Forsberg B: Epidemiological evidence of effects of coarse airborne particles on health. Eur Respir J 2005, 26: 309318. 10.1183/09031936.05.00001805View ArticleGoogle Scholar
 Li JJ, Shao LY, Yang SS: Adverse effect mechanisms of inhalable particulate matters. J Environ Health 2006, 3: 185188.Google Scholar
 Pope CA, Dockery DW: Health effects of fine particulate air pollution: lines that connect. J Air Waste Manage Assoc 2006, 56: 709742. 10.1080/10473289.2006.10464485View ArticleGoogle Scholar
 Samoli E, Analitis A, Touloumi G, Schwartz J, Anderson HR, Sunyer J, Bisanti L, Zmirou D, Vonk JM, Pekkanen J, Goodman P, Paldy A, Schindler C, Katsouyanni K: Estimating the exposureresponse relationships between particulate matter and mortality within the APHEA multicity project. Environ Health Persp 2005, 113(1):8895. 10.1289/ehp.7387View ArticleGoogle Scholar
 Kumar N, Chu A, Foster A: An empirical relationship between PM2.5 and aerosol optical depth in Delhi Metropolitan. Atmos Environ 2007, 41: 44924503. 10.1016/j.atmosenv.2007.01.046View ArticleGoogle Scholar
 Tian J, Chen D: A semiempirical model for predicting hourly groundlevel fine particulate matter (PM2.5) concentration in southern Ontario from satellite remote sensing and groundbased meteorological measurements. Remote Sens Environ 2010, 114: 221229. 10.1016/j.rse.2009.09.011View ArticleGoogle Scholar
 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: 47774788. 10.1016/j.atmosenv.2009.10.006View ArticleGoogle Scholar
 AlSaadi J, Szykman J, Pierce RB, Kittaka C, Neil D, Chu DA, Remer L, Gumley L, Prins E, Weinstock L, MacDonald C, Wayland R, Dimmick F, Fishman J: Improving national air quality forecasts with satellite aerosol observation. Amer Meteor Soc J 2005, 86: 12491261. 10.1175/BAMS8691249View ArticleGoogle Scholar
 Liu Y, Franklin M, Kahn R, Koutrakis P: Using aerosol optical thickness to predict groundlevel PM_{ 2.5 } concentrations in the St. Louis area: a comparison between MISR and MODIS. Remote Sens Environ 2007, 107: 3344. 10.1016/j.rse.2006.05.022View ArticleGoogle Scholar
 Péré JC, Pont V, Mallet M, Bessagnet B: Mapping of PM_{ 10 } surface concentrations derived from satellite observations of aerosol optical thickness over SouthEastern France. Atmos Res 2009, 91: 18. 10.1016/j.atmosres.2008.05.001View ArticleGoogle Scholar
 Wang J, Christopher SA: Intercomparison between satellitederived aerosol optical thickness and PM2. 5 mass: implications for air quality studies. Geophys Res Lett 2003, 30: 2095. 10.1029/2003GL018174View ArticleGoogle Scholar
 Slater JF, Dibb JE, Campbell JW, Moore TS: Physical and chemical properties of surface and column aerosols at a rural New England site during MODIS overpass. Remote Sens Environ 2004, 92: 173180. 10.1016/j.rse.2004.05.011View ArticleGoogle Scholar
 Chu DA, Kaufman YJ, Zibordi G, Chern JD, Mao J, Li C, Holben BN: Global monitoring of air pollution over land from the Earth Observing SystemTerra Moderate Resolution Imaging Spectroradiometer (MODIS). J Geophys Res 2003, 108: 4661. 10.1029/2002JD003179View ArticleGoogle Scholar
 Kim JE, Ryu SY, He Z, Kim YJ: Spectral aerosol optical depth variation with different types of aerosol at Gwangju, Korea. J Atmos SolTerr Phy 2006, 68: 16091621. 10.1016/j.jastp.2006.05.008View ArticleGoogle Scholar
 Liu Y, Saranat JA, Kilaru V, Jacob DJ, Koutrakis P: Estimating groundlevel PM_{ 2.5 } eastern United States using satellite remote sensing. Environ Sci Technol 2005, 39: 32693278. 10.1021/es049352mView ArticleGoogle Scholar
 Vladutescu V, Diaz J, Charles L, Moshary GF, Barry M, Ahmed SA: Aerosol layer properties and their effect on optical depth relations to PM_{ 2.5 } concentrations. Opt Remote Sens Lab 2006, 2008: 589592.Google Scholar
 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: 53045315. 10.1016/j.atmosenv.2006.04.044View ArticleGoogle Scholar
 Pelletier B, Santer R, Vidot J: Retrieving of particulate matter from optical measurements: a semiparametric approach. J Geophys Res Atmos (19842012) 2007, 112: 118.Google Scholar
 Vidot J, Santer R, Ramon D: Atmospheric particulate matter (PM) estimation from SeaWiFS imagery. Remote Sens Environ 2007, 111: 110. 10.1016/j.rse.2007.03.009View ArticleGoogle Scholar
 Gupta P, Christopher SA: Particulate matter air quality assessment using integrated surface, satellite, and meteorological products: multiple regression approach. J Geophys Res Atmos (19842012) 2009, 114: 113.Google Scholar
 Tian J, Chen D: Spectral, spatial, and temporal sensitivity of correlating MODIS aerosol optical depth with groundbased fine particulate matter (PM_{ 2.5 }) across southern Ontario. Can J Remote Sens 2010, 36: 119128. 10.5589/m10033View ArticleGoogle Scholar
 Givehchi R, Arhami M, Tajrishy M: Contribution of the Middle Eastern dust source areas to PM_{ 10 } levels in urban receptors: case study of Tehran, Iran. Atmos Environ 2013, 75: 287295. 10.1016/j.atmosenv.2013.04.039View ArticleGoogle Scholar
 Schaap M, Apituley A, Timmermans RMA, Koelemeijer RBA, de Leeuw G: Exploring the relation between aerosol optical depth and PM_{ 2.5 } at Cabauw, the Netherlands. Atmos Chem Phys 2009, 9: 909925. 10.5194/acp99092009View ArticleGoogle Scholar
 Savtchenko A, Ouzounov D, Ahmad S, Acker J, Leptoukh G, Koziana J, Nickless D: Terra and Aqua MODIS products available from NASA GES DAAC. Adv Space Res 2004, 34: 710714. 10.1016/j.asr.2004.03.012View ArticleGoogle Scholar
 Remer LA JKY, TanrÉ D, Mattoo S, Chu DA, Martins JV, Li RR, Ichoku C, Levy RC, Kleidman RG, Eck TF, Vermote E, Holben BN: The MODIS aerosol algorithm, products, and validation. J Atmos Sci 2005, 62: 947973. 10.1175/JAS3385.1View ArticleGoogle Scholar
 Levy RC, Remer L, TanrÉ D, Mattoo S, Kaufman YJ: algorithm for remote sensing of tropospheric aerosol over dark targets from MODIS: collections 005 and 051:revision 2. 2009, 96:. Download from 2009:196. Download from [http://modisatmos.gsfc.nasa.gov/_docs/ATBD_MOD04_C005_rev2.pdf]
 EngelCox JA, Holloman CH, Coutant BW, Hoff RM: Qualitative and quantitative evaluation of MODIS satellite sensor data for regional and urban scale air quality. Atmos Environ 2004, 38: 24952509. 10.1016/j.atmosenv.2004.01.039View ArticleGoogle Scholar
 EngelCox JA, Hoff RM, Rogers R, Dimmick F, Rush AC, Szykman JJ, AlSaadi J, Chu DA, Zell ER: Integrating lidar and satellite optical depth with ambient monitoring for 3dimensional particulate characterization. Atmos Environ 2006, 40: 80568067. 10.1016/j.atmosenv.2006.02.039View ArticleGoogle Scholar
 Gupta P, Christopher SA: Seven year particulate matter air quality assessment from surface and satellite measurements. Atmos Chem Phys 2008, 8: 33113324. 10.5194/acp833112008View ArticleGoogle Scholar
 Emili E, Popp C, Wunderle S, Zebisch M, Petitta M: Mapping particulate matter in alpine regions with satellite and groundbased measurements: an exploratory study for data assimilation. Atmos Environ 2011, 45: 43444353. 10.1016/j.atmosenv.2011.05.051View ArticleGoogle Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.