1 Introduction

Air quality is an important indicator of human health and wellbeing. Any increases in the concentration of harmful air pollutants can significantly impact human health, increasing risks of cardiovascular diseases, asthma, and premature deaths. Air pollution is one of the most significant environmental issues that are concentrated in urban cities and poses a significant risk to contemporary society, especially with rapid urbanization, industrialization and migration [1] and [2]. Anthropogenic stressors negatively impact the environment such as polluting the air, the water resources and the soil for different plant species [3]. Most air pollution results from the production and use of energy and emissions from factories, power plants, and automobiles [4]. Air pollution whether indoor or outdoor is responsible for approximately 2 million premature deaths and kills around 7 million people a year globally [5] and [6]. Air pollution increases the risks of respiratory and cardiovascular diseases, hospitalizations, which affect the overall economy of the country [7] and [8]. It is evident that people living in environments with poor air quality are more vulnerable to asthma, ventricular hypertrophy, lung cancer, retinopathy, fatal growth, Alzheimer’s, and Parkinson’s disease. In China, polluted air was responsible for about 1.2 million deaths in 2010 and around 1.6 million deaths in 2014 [9]. Research has shown that air pollution is a contributing factor to high infant mortality rates in developing nations in Africa [10].

In the context of air pollution, particulate matter (also known as atmospheric aerosols) are fine suspended particles in the form of solid or liquid state and they derive from natural sources such as forest fires, volcanic eruption, desert dust or from human-related activities such as biomass or fossil fuels burning [11]. The concentration of PM is known to be high in regions with increased industrial activity, urbanization, traffic, and with poor emission control measures [12]. These particles have a strong interaction with solar radiation, the temperature, vertical profile, earth surface characteristics, and other thermodynamic variables [12]. Aerosols have different effects on human health, climate dynamics, and visibility [13]. The presence of atmospheric aerosols (PM) in high concentrations in the atmosphere compromises the quality of air and contributes to the rising rates of pollution [14].

The two major types of PM are PM2.5 and PM10, where PM2.5 corresponds to a diameter less than 2.5 µm (fine particles) and PM10 that has a diameter of 10 µm and is classified as coarse particles [15]. Given that the size of PM is very small, those particles can easily enter the human lungs through inhalation and cause huge health risks [16]. Absorption and scattering of solar radiation by these particles can cause serious problems for human health [17] and [18]. Indoor air pollution that results from physical and biological aerosols produced from cooking activities, incenses and air conditioners is a leading cause of pneumonia, stroke, IHD, COPD, and cancer that is linked to 4.5 million annual deaths globally [19] and [20]. A study found an association between air pollution and high levels of PM and human health in UAE, specifically serious diseases such as cardiovascular diseases, cancers, and pulmonary inflammation [21].

Aerosol optical depth (AOD) or Aerosol optical thickness (AOT) is a parameter that provides an estimation of aerosol particles in the air from ground based measurements [22]. The status of the air can be indicated based on the AOD value. Low AOD values indicate that the sky is clear and pure, and higher values mean lower in air visibility and purity due to the presence of aerosols [23]. AOD describes the attenuation of solar radiation in the atmosphere as it gets scattered and/or absorbed by the PM [24]. Monitoring of the AOD by satellite remote sensing is widely used along with ground-based in situ measurements corresponding to the satellite data, provided that satellite imagery provides larger spatial coverage [25]. There is a global ground network of air pollution monitoring stations such as Aerosol Robotic Network (AERONET), which assists in tracking the level of PM concentration and its impact on the globe [26]. AERONET is a ground network that measures many parameters of the atmospheric aerosol robotically, including PM10, and performs data processing and data calibration before providing the data [27]. There are local monitoring stations established in different countries in order to complement and densify global networks [28].

Ground monitoring stations measure air quality accurately, at any time, and are good measuring tools for gauging air pollution levels and trends. However, most ground stations are sparsely distributed and are located near cities limiting the effectiveness and representativeness [29]. The information is limited to the region without incorporation of other influencing factors such as meteorological data, land use categories, sources of emission data, topographic, and socioeconomic factors [30]. Given the limitation of ground monitoring stations, satellite remote sensing-based methods have become very popular. Satellite remote sensors are useful in estimating atmospheric content given the limitations of available ground stations and the expensive costs associated with monitoring [31]. Satellite sensors provide information on a wider geographical area than ground stations, are accurate, and offer the required data that can be used in the estimation of PM concentrations in the atmosphere [32]. Subsequently, the use of satellite remote sensing has become a major source of information about the atmosphere.

Remote sensing is the science of obtaining information from a distance and is typically carried out via aircrafts or satellites. The satellites currently in orbit that provide free continuous access to AOD data are Moderate Resolution Imaging Spectroradiometer (MODIS), National Oceanic and Atmospheric Administration (NOAA), Geostationary Operational Environmental Satellite (GOES), Visible Infrared Imaging Radiometer Suite (VIIRS), and Landsat [33]. The free continuous access to such data is an essential component of a monitoring system. These types of sensors are used in obtaining atmospheric information including water and air current flow, as well as the level of pollution through estimating the levels of aerosol in the atmosphere. Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS) images are the main instruments for Landsat 8 [34]. Landsat 8 acquired multispectral bands, panchromatic band, and TIRS bands. The four optical bands of Landsat 8 used in this study have a spatial resolution of 30 m, which is sufficient for PM10 monitoring given that these particles do not propagate for long distances given their sizes [35].

Despite the numerous applications for satellite remote sensing, their use in estimating air pollutants concentrations is limited in the literature, with a few studies in which satellite images were used in prediction models for particulate matter [36,37,38]. Different models were developed in the literature associating spectral information from satellite imagery to the concentration of PM [39,40,41,42]. For the UAE, a country highly susceptible to sandstorms and high aerosol concentrations due to its geography and geomorphology, satellite remote sensing can be useful in creating PM models, which would help to better understand the spatiotemporal behaviour of aerosols as they vary across the country. Only one study has attempted to create a PM10 model for the UAE, for the city of Al Ain specifically, using MODIS Aerosol Optical Depth (AOD) data to create a linear regression model correlating the AOD to the PM10 concentration over the city [43]. Developing a robust model for PM10 concentrations from Landsat imagery is needed given the frequent dust storms in the country and the region and can help not only in mapping, but also monitoring of the PM10 concentrations whenever satellite images are available.

The UAE hosts forty-one ground monitoring stations according to the Air Quality Index Manual, but this number is insufficient since it does not cover all areas, especially the remote and rural areas of the country, given that data is not accessible for most of them [44]. This study aims to monitor PM10 concentration levels covering the period 2016–2020 using Landsat 8 imagery, which will help in establishing a high-resolution remote sensing-based PM monitoring system for UAE although the temporal resolution is 16 days. This cost-effective method would complement PM10 ground monitoring techniques and provides higher spatial resolution.

The UAE is particularly susceptible to sandstorms, which increases the concentration of PM10 in the air. The UAE has also urbanized rapidly over a short period of time, especially the main cities of Abu Dhabi and Dubai, which their sizes have expanded many times their size. Therefore, it is necessary to develop a cost-effective, area-covering, air quality monitoring system that shows the pattern of the varying concentration levels of air pollutants including PM10. The main goal of this study is to develop a satellite remote sensing based PM10 monitoring system for the UAE that uses multi-spectral surface reflectance values from Landsat 8 Operational Land Image (OLI) imager, noting that such a system doesn’y currently exst and also has not been addressed in the literature for the UAE.

2 Methodology

2.1 Study area

The study area of this research is the UAE, which is a nation located in the Arabian Gulf in western Asia at the southeastern end of the Arabian Peninsula as presented in Fig. 1. Geographically, the UAE extends between 22° 50′–26° 4′ N latitudes and 51° 5′–56° 25′ E longitudes, covering a total area of 83,600 km2, more than two-thirds of which is desert [1]. The arid/semi-arid environment of the UAE and the presence of sand and sand dunes, particularly in the southern region of the UAE and the local pollution emissions that arises from the petroleum and construction industry makes it vulnerable to experience frequent sandstorms and natural dust episodes in the atmosphere, which result in increased levels of CO2 and suspended PM [2]. One of the adverse impacts of urbanization is the increase in air pollution.

Fig. 1
figure 1

Study area of the United Arab Emirates (UAE)

Full size image

2.2 Data analysis and processing

In this section, data acquisition, including the procedures of obtaining the data from the ground monitoring stations and downloading the Landsat 8 satellite images are presented. Table 1 provides information about the characteristics of bands of Landsat 8. Landsat has a temporal resolution of 16 days, which means that the monitoring capacity is acceptable, and hence it is sufficient to capture data within the UAE [45]. Moreover, Data processing, including radiometric processing of the optical bands of Landsat 8 into reflectance values and estimating AOD, are presented. The procedure used in this study is presented in the flowchart shown in Fig. 2. Data acquisition consists of the Landsat 8 satellite images obtained for the UAE from the USGS Core Science Systems as well as obtaining PM10 data from ground monitoring stations. The following subsection will provide a detailed overview of the flowchart presented.

Table 1 Landsat 8 bands characteristics
Full size table
Fig. 2
figure 2

Workflow Process for PM10 Estimation

Full size image

2.3 Data acquisition

Landsat 8 images were retrieved from the United States Geological Survey (USGS) Core Science Systems through the Landsat Earth Explorer [46]. The Landsat-8 payload consists of two science instruments: the Operational Land Imager (OLI) and the Thermal InfraRed Sensor (TIRS). Landsat 8 has 11 bands where nine bands are under the OLI sensor, and two bands are under the TIRS sensor [47]. Nine bands are included in the OLI instrument where bands 1–7 and band 9 are spectral bands that have 30 m spatial resolution. However, band 8 is the panchromatic band with a 15 m resolution. Moreover, the TIRS gathers the data in two long wavelength thermal infrared bands which are bands 10 and 11 that have a resolution of 100 m and are resampled to 30 m during delivery of the data product. The acquisition dates of Landsat 8 images used in the study are the years 2016, 2018, and 2020, with cloud cover of 10% or less. Also, visible bands 1, 2, 3, and 4 of Landsat 8 OLI images for the UAE area were provided within the path and row of (159–162) and (48–50), respectively.

2.4 PM10 data

PM10 ground measurement data in this study was obtained from the Air Pollution in World: Real-time Air Quality Index Visual Map website. The website offers daily ambient air quality monitoring data for forty-one stations in the UAE. PM10 data for 30 stations was used to develop the satellite remote sensing based PM10 model, and eleven stations were used for model validation. The literature review proved that generally dividing the testing points using the 75:25 ratio provides very good accuracy [48]. Hence, 30 stations were chosen to perform the study, which is around 75% of the existing available stations, and the other four stations were used to test the validity of the established model. Since eight satellite images are needed to capture the whole UAE domain, eight images were downloaded for each year as depicted in Fig. 3. Table 2 below shows the times at which PM10 data was used from the ground stations.

Fig. 3
figure 3

PM10 ground monitoring stations and Landsat 8 Images covering the UAE

Full size image
Table 2 Reflectance values for the four bands of each image of the selected 30 stations for the year 2016
Full size table

2.5 Data processing

Landsat 8 Level 1 data for the four bands used in this study were first pre-corrected geometrically. Moreover, the obtained images were selected using a cloud cover mask of 10%. The data was quantified and calibrated into scaled Digital Numbers (DN) which were converted into TOA reflectance by applying radiometric correction to transform the DNs into reflectance values.

Conversion of the DNs into TOA Reflectance. The major rescaling factors in converting OLI bands data to TOA planetary reflectance (ρλ’) are independent variables consisting of band-specific multiplicative rescaling factor (Mρ), band-specific additive rescaling factor (Aρ), and quantized and calibrated standard product pixel values of DN (Qcal). These factors were extracted from the imagery metadata files. The following equations were used to compute TOA planetary reflectance (ρλ’) following Landsat 8 standard processing methods [49].

$$\rho \lambda ^{\prime}\, = \,{\text{M}}_{\rho } {\text{Q}}_{{{\text{cal}}}} \, + \,{\text{A}}_{\rho }$$
(1)

Equation 1 calculates TOA reflectance without correction for solar angle and hence, Eq. 2 is used for the correction of the sun angle.

$${\rm P}\lambda \, = \,\rho \lambda{\prime} /{\text{ cos}}\left( {\theta_{SZ} } \right)\, = \,\rho \lambda{\prime} /{\text{ sin}}\left( {\theta_{{{\text{SE}}}} } \right)$$
(2)

where Ρλ refers to TOA planetary reflectance, θSE is the local sun elevation angle, and θSZ is the local solar zenith angle. The next step involves Aerosol Optical Thickness (AOT)/AOD and PM10 correlation. Regression algorithm analysis is applied to correlate atmospheric AOT to PM10. Aerosol Optical Thickness algorithm for a single band is represented in Eq. 3.

$${\text{T}}\alpha = \left( {\frac{4\mu \mu o}{{\omega oPa(\theta s,\theta v,\phi }}} \right)\rho_{atm} \left( {\theta_{s} ,\theta_{v} ,\phi } \right).$$
(3)

where Ʈα is the AOT, ρatm is the atmospheric reflectance, θs is the solar zenith angle, θv is the viewing zenith angle, ɸ is the relative azimuth angle, µ is the cosines of view of directions, and µo is the cosines of the illumination directions. Similarly, AOT can be simplified, and Eq. 3 can be simplified into Eq. 4 with the addition of different constants.

$${\text{AOT}}\left( \lambda \right)\, = \,{\text{a}}_{{\text{o}}} {\text{R }}\lambda_{{1}} \, + \,{\text{a}}_{{1}} {\text{R}}\lambda_{{2}} \, + \,{\text{a}}_{{2}} {\text{R}}\lambda_{{3}} \, + \,{\text{a}}_{{3}} {\text{R}}\lambda_{{4}}$$
(4)

where R λ1, R λ2, R λ3, and R λ4, are the reflectance that correspond to bands 1, 2, 3, and 4 of Landsat 8 respectively. While the aj are the model coefficients that are empirically determined. The relationship between AOT and PM10 is established using a single and homogenous atmospheric layer that contains spherical aerosol particles. Therefore, the particulate matter concentration can be compared significantly with the AOT directly to determine air pollution. Different software systems were used in this study to complete the processing and the model development. The ESRI ® ArcGIS was utilized to calculate the TOA reflectance values using Eq. 2. The band-specific multiplicative rescaling factor (Mρ), band-specific additive rescaling factor (Aρ), quantized and calibrated standard product pixel values of DN (Qcal) and the local sun elevation angle (θSE) were extracted from the MTL Notepad files that come with the images in order to proceed with the reflectance TOA calculations.

Out of the existing forty-one ground monitoring stations in the UAE, only sixteen stations have accessible PM10 data, as mentioned earlier. Table 2 shows the sixteen stations that are used in this study. Data from twelve of these stations were used to develop the model coefficients (ao, a1, a2 and a3) which are shown in Eq. 4. Normally, PM10 can be determined as a function of AOT by incorporating Eq. 4 into Eq. 5 [50] and [51].

$${\text{PM}}_{{{1}0}} \, = \,{\text{a}}_{{\text{o}}} {\text{B}}_{{1}} \, + \,{\text{a}}_{{1}} {\text{B}}_{{2}} \, + \,{\text{a}}_{{2}} {\text{B}}_{{3}} \, + \,{\text{a}}_{{3}} {\text{B}}_{{4}}$$
(5)

where B1, B2, B3, and B4 are the reflectance values calculated using Eq. 2 in ArcGIS raster calculator. The values ao, a1, a2, and a3 are the model coefficients that were calculated using MATLAB in order to solve the PM10 linear equations. Considering the twelve stations by adopting the data of the three selected years resulted in establishing a system of thirty-six linear equations that were solved in MATLAB using the minimum mean square error method to resolve the indeterminate situation.

3 Results and discussion

In this section, the resulting values and coefficients to establish the PM10 monitoring model are presented. Moreover, the model was stablished and validated to determine the PM10 concentration for any location within the UAE. Table 2 shows the reflectance values for the four bands of each image of the selected twelve stations for the year 2016 and similar calculations were performed for years 2018, 2020, and 2022.

Using data from 30 ground stations, 120 equations were established. The minimum mean square error method in MATLAB was used to estimate the model coefficients (ao, a1, a2, and a3), which are shown in Table 3. The finalized spectral model is shown in Eq. 6.

$${\text{PM}}_{{{1}0}} \, = \,{248}0.{\text{5B}}_{{1}} \, + \,{133}0.{\text{3B}}_{{2}} \, + \,{2618}.{\text{8B}}_{{3}} \, + \,{1653}.{\text{4B}}_{{4}}$$
(6)
Table 3 Model coefficients obtained based on the equations
Full size table

As mentioned earlier, data from 30 stations were used to establish the model. The remaining 11 stations were used for validation. The model R-squared (R2) value was estimated to be around 65%. This value indicates that the model can robustly estimate the PM10 concentrations spectrally, given Landsat 8 availability at the location of interest.

In comparison to previous studies, our model R2 value is lower than that obtained by [52] for Delhi, India and by [53] over Kirkuk, Iraq. These studies reported R2 values of 80% and 83%, respectively. The reason for obtaining a lower R2 value for our spectral model is the use of time series data for 2016, 2018, and 2020, respectively. This is an advantage to our model, which takes into consideration the atmospheric variations over the period of study, unlike the previous studies that were conducted over one episode of time. This time consideration is advantageous to our study and definitely has a huge impact on the variability of PM10 concentrations due to many factors. This demonstrates the uniqueness of our model, which reflects the spatial and temporal variation of atmospheric factors such as wind speed, direction, humidity, and atmospheric pressure. In addition to the R2 assessment, Root Mean Square Error is another accuracy measure that shows the robustness of the model. RMSE was calculated for the model and found to be 17.67 µg/m3. Further, data from eleven ground stations were used for model validation. The result revealed a satisfactory R2 value of 82.15% with a RMSE of 16.88 µg/m3.

The model has demonstrated robustness in estimating PM10 concentrations from the four bands of Landsat 8 imagery, Landsat images were downloaded for December 2017, September 2019, and December 2021 to represent different seasons of the year. Landsat images were processed and mosaiced for the purpose of visualizing PM10 concentration, which are estimated with the spectral model in color-coded maps covering the UAE. Figure 4 shows the color-coded images for December 2017, September 2019, and December 2021 images.

Fig. 4
figure 4

A PM10 concentration in a color-coded image covering the whole UAE in December 2017. B PM10 concentration in a color-coded image covering the whole UAE in September 2019. C PM10 concentration in a color-coded image covering the whole UAE in September 2021

Full size image

It can be inferred from Fig. 4that the red areas represent high PM10 concentration, yellow areas represent a medium concentration, and the green areas represent a low PM10 concentration area. Although the month of September is normally not windy, PM10 levels were still higher than in December 2017. This increase in the particulate matter concentration is due to the urbanization expansion and the related increased level of air pollution. The study also confirmed that the time of the year has a huge impact in particulate matter concentration. Furthermore, high levels of PM are observed during the dusty summer season as a result of the Shamal winds, which is known for producing notable sandstorms. In fact, the PM concentration is linked to the emissions of the primary and the secondary PM initiators, including natural atmospheric causes such as the Shamal winds [54]. The color-coded map of December 2017 (Winter season) shows less PM10 concentration than that estimated in September 2019 (End of the summer season), taking into consideration the increase in the urbanization and pollution rates over the years from 2017 to 2019.

4 Conclusion

This study aimed at developing a remote sensing-based model for estimating and monitoring of PM10 levels in the UAE. Given that PM10 is a major contributor to air pollution and has adverse health effects, this study has attempted to develop a spectral model based on Landsat 8 data. The model R-squared value of 65% and the Root Mean Square Error (RMSE) of 12.55 µg/m3 show the model efficiency. The results of the study have demonstrated the successful use of Landsat 8 imagery in estimating PM10 concentration over the UAE. The resulting spectral model, which has a monitoring potential, can be used to estimate PM10 values at any location in the UAE where Landsat 8 data is available. The results of the study are in agreement with that of [33] and [34]. Moreover, ArcGIS and MATLAB software systems were essential in the process to ease the set of calculations in order to obtain the model and complete this study. Furthermore, the results proved that satellite imagery could be used in monitoring PM10 over large areas.

Although the results of the study showed that the model is robust in estimating PM10 concentration in UAE, the model would have been more robust if the data of all the existing ground monitoring stations were accessible. Using more ground monitoring stations in future studies will improve the results, and therefore the model will be improved. Furthermore, studying the contribution of the seasonal variations and the geographical locations of the PM10 concentrations, and the variation in related atmospheric factors would help in understanding the causes of high PM10 concentration variability over time.