Remote Sensing and Geographic Information Systems Detection of Fossil Fuel Air Pollution Impact in Socially Fragile Areas

Abstract

One of the important effects of global warming is the use of fossil fuels. Disadvantaged individuals may be affected by fossil fuel use more than others. In this study, the Kepez district of Antalya province, where the Social Vulnerability Index (SVI) is high, was selected as the study area. Five-year (2019–2023) NO₂, SO₂, and CO concentrations were extracted from the Sentinel-5P TROPOMI satellite with open-source code. These values were combined and compared with Land Use Land Cover (LULC) land classes obtained from the Sentinel-2 satellite. The same process was performed for Land Surface Temperature (LST) obtained from MODIS Terra and Aqua satellites, and interpretation was made according to the LST-LULC map and surface temperature. The integrated SVI was calculated with population, age, education, and gender data from the Turkish Statistical Institute and NO₂, SO₂, and CO concentrations from the Sentinel-5P TROPOMI satellite. It was mapped on a neighborhood basis with zonal statistics. Accordingly, 20.6% of the neighborhoods in Kepez were categorized as very high risk, and 16.2% were categorized as high risk. Integrated SVI with the determination made by evaluating only air pollution gave different neighborhood results. This revealed the importance of using the SVI in disaster risk assessments. This study has the potential to shed light on the social vulnerability-supported disaster risk information system that is likely to be created in the following years.

1. Introduction

Fossil fuels pose risks to the environment and human health during their extraction, transport and use [1]. There is a relationship between air pollution indicators and respiratory and other causes of mortality [2]. Exposure to air pollution (especially PM_2.5) is considered to be a risk factor for mortality globally [3,4]. According to WHO (World Health Organization) data, more than 7 million people die annually due to air pollution [5]. There is an association between air pollution and premature mortality [6].

The quality and quantity of coal used for domestic heating pose an environmental risk [7]. PAH (polycyclic aromatic hydrocarbons) harm human health with domestic coal use [8]. The use of low-quality fossil fuels increases the pollution rate in air pollutants [9]. A relationship has been established between respiratory symptoms and diseases due to domestic coal use and smoking [10]. In China, one of the leading countries in world coal consumption, many people in rural areas still use coal as the primary heat source [11]. In Türkiye, the country where this study was conducted, atmospheric pollutants are mainly from domestic heating, industry and traffic [12].

Particulate matter pollution may increase due to various demographic and socioeconomic factors such as age, ethnicity, income and deprivation [13]. There is a positive correlation between NO₂ and CO obtained with TROPOMI and demographic data [14]. Carbon monoxide, formaldehyde, nitrogen dioxide and the aerosol index can be detected with Sentinel-5P satellite data from the cloud-based Google Earth Engine (GEE) [15]. There is a strong correlation between NO₂ concentration data from the Sentinel-5P TROPOMI satellite and traffic and industrial activities [16]. With Sentinel-5P TROPOMI satellite images, NO₂, SO₂ and CO gas emissions can be determined with GEE and mapped with the kriging interpolation method [17].

In the COVID-19 pandemic, a decrease was observed in the detection of NO₂ data on the closing days with Sentinel-5P TROPOMI [18]. During the COVID-19 pandemic, a decrease in NO₂ and O₃ concentrations was observed [19]. Although the NO₂ concentration determined by Sentinel-5P TROPOMI satellite imagery decreased during the closures during the COVID-19 pandemic, this situation presents high concentrations in highway areas where the continuous urban structure is active and very high concentrations in commercial and industrial areas [18].

There is strong consistency in the comparison of Sentinel-5P TROPOMI and data from the ground-based measurement station [20]. High accuracy was found in the temporal monitoring and comparison of ground data (NO₂ and CO) with Sentinel-5P [21]. A high correlation was found in the comparison of Sentinel-5P TROPOMI and ground stations by Pearson correlation test [22]. Kriging and multiple regression can be used in the mixing of data from Sentinel-2 and ground station data, and high-accuracy results were obtained [23]. The SIM (Continuous Monitoring Center) in Türkiye is inadequate; the distributions can be determined by multiple linear regression, principal component regression and least squares regression with open-source data sets [24].

Within the same category, groups that are disadvantaged compared to other people can be characterized as socially vulnerable [25]. The most important factor for labeling a social group as vulnerable is the potential for risk [25,26]. SVI is used to identify where the vulnerable population is located [27]. SVI measures both the susceptibility of the recipients and the social and economic resources to prevent or mitigate the impacts [28].

SVI is an estimate of the potential for a neighborhood to suffer harmful consequences in the face of natural disasters or disease. It was originally developed to identify vulnerability in the context of disaster management, but it is also used in healthcare studies [29]. Although the SVI was designed to help target emergency personnel in disaster response efforts, the index has moved beyond these contexts [30]. For the SVI, 0 is defined as the least vulnerable and 1 is defined as the most vulnerable in the regions where the data are obtained [30]. Research in the environmental justice literature has analyzed social vulnerability using different kinds of metrics and standards, approaches and lack of consensus [30].

Areas with low socioeconomic status have higher concentrations of air pollution [31,32,33]. Low-income communities in urban areas are likely to be exposed to high concentrations of air pollutants such as ozone and particulate matter [30,34,35,36,37] or lead [30,38,39]. Patients from socially vulnerable communities by age, gender, race, ethnicity and hospitalized patients are more likely to die in hospital and experience major cardiovascular events [40]. Differences in PM_2.5 mortality may vary by education, rural residence, social vulnerability, and race/ethnicity [4].

Identifying factors and mapping social vulnerability have become extremely necessary for environmental management and sustainable development. However, studies linking social vulnerability to air pollution are still insufficient [41]. Although there are many studies in the literature on the detection of individual air pollutants with Sentinel-5P TROPOMI and the Social Vulnerability Index, there are very few studies that combine and compare these two studies. There is no similar study in which different remote sensing parameters such as land classification and surface temperature are also analyzed. The aim of this study is to reveal the spatial relationship between air pollution effects in settlements with socioeconomically disadvantaged population potential and the Social Vulnerability Index by using remote sensing methods and GIS tools. In this context, the Kepez district of Antalya province, which has a population of more than 600,000, is the study area. Kepez was preferred because it is home to disadvantaged individuals, refugees, and foreign nationals, and it includes a ring road, bus terminal, and industrial facilities. The inferences obtained from the study will provide decision-makers with visual information about the location, causes and spatial outputs of regional problems and will increase their ability to fight effectively in the field. Data and analyses will contribute to strategic decisions in the fight against climate change.

2. Study Area

Antalya is an important province in terms of tourism, agriculture and industry that is located in the southwest of Türkiye with a population of more than 2.5 million. The study was conducted in Kepez, which is the central district of Antalya with the highest Social Vulnerability Index. The selection of Kepez was influenced by its population density, agriculture, and industrial facilities. In addition, the fact that socioeconomically disadvantaged individuals live in this district is also an important factor. The fact that Kepez is a pioneer in the province in terms of domestic fossil fuel use as well as industrial facilities is also one of the important parameters of preference [42].

Kepez consists of 68 neighborhoods, and its surface area is 397,869 decares. Approximately 22% of this area is agricultural land. The population is over 600,000 as of 2024 [43]. According to the data obtained from TurkStat, the population average of the last 5 years is 582,696 people. Of this, 171,097 are 0–18 years old and 35,842 are over 65 years old. In a study on poverty alleviation in Antalya Kepez, it was stated that the people living in this district mostly demanded food, clothing, goods, education and fuel aid from local administrations [44]. This shows that there are individuals in the study area who cannot meet their basic needs and that the Social Vulnerability Index is high. Reasons such as a vulnerable population, poverty, the presence of industrial facilities, and domestic fossil fuel use were effective in the selection of Kepez as the study area. The location map of the study area is given in Figure 1.

3. Data Sets, Methods and Analysis

The data sets of the study obtained by remote sensing methods can be listed as NO₂, SO₂ and CO concentrations obtained from the Sentinel-5P TROPOMI satellite, LST obtained from MODIS satellite Terra and Aqua instruments, and LULC obtained from the Sentinel-2 satellite (ESRI Living Atlas of the World). In addition, population, education, age, gender, household, marital status data are statistical data sets obtained from the Turkish Statistical Institute. In this study, zonal statistics with GIS and basic statistical methods were used in SVI calculations. In the study, R statistical software, Google Earth Engine, and Python were used to analyze, map and graph the data using open-source code. This study aims to calculate an integrated SVI index based on air pollution produced with these data sets. In addition, neighborhood-based air pollutants were interpreted with LULC land classification and LST. The workflow chart including the data sets and methods of the study is given in Figure 2.

3.1. Data Sets and Preparation

The data sets in this study were prepared in two stages: remote sensing methods for the calculation of air pollutants and land modeling, the use of statistical methods with data obtained from institutions, and social vulnerability calculation. The data sets used are given in

Table 1. Table of the data set used in the study.

Process	Source	Platform	Data	Resolution (m)	Data Type	Year	Unit
RS (Remote Sensing)	Google Earth Engine [45]	Sentinel-5P TROPOMI	CO	1113.2	Raster Data	2019–2023	mol/m2
			NO2
			SO2
		Terra and Aqua MODIS	LST	1000			°C
	ESRI (Living Atlas of the World) [46]	Sentinel-2	LULC	10		2023	-
Process	Source	Data	Classification		Data Type	Reference [25]
SVI (Social Vulnerability Index)	Turkish Statistical Institute (data taken between 2019–2023)	Population density	Person/km2		Text and number	[47]
		Households	Population per household			[48]
		Marital status (Female/Male separately)	Never married			[25]
			Married
			Divorced
			Dead
		Age	<15			[49,50]
			+65
		Gender	Woman			[49]
			Male
		Education	Illiterate
			Primary school graduate
			Secondary school graduate
			High school graduate
			Graduate of Higher School/Faculty
			Master’s degree graduate

.

3.2. Land Use Land Cover (LULC)

The ArcGIS Living Atlas of the World platform, provided as open source by ESRI, uses 10-m resolution Sentinel-2 satellite imagery. The Sentinel-2 L2A surface reflectance data are generated by a deep learning model using visible blue, green, red, near-infrared and two shortwave infrared bands, which were sampled over 20,000 sites and trained using more than five billion manually labeled Sentinel-2 pixels.

One of the most important capabilities of this application is that it provides dynamic change analysis. Apart from this, it enables dynamic statistical change analysis according to year, map coverage, and class. With the application, filters can be made according to the selected cover class, and regional class statistics can be calculated according to administrative boundaries. It also provides data download and offline use [46]. It was preferred in the study due to all these advantages.

Using these data, it is possible to create a precise and detailed land use thematic map with 9 different classes (water, tree, flooded vegetation, agriculture, building, bare land, snow/ice, cloud, pasture) [51]. LULC classification was made in the study area. Accordingly, the building area, forested area, pasture, agricultural area, and bare land are the land classes identified in Kepez (Figure 3).

3.3. Creating Air Pollutant Maps Using Open-Source Code

The Sentinel-5P satellite, co-funded by the European Space Agency (ESA) and the Netherlands, carries a single payload: the TROPOMI spectrometer. Launched on 13 October 2017 by the ESA, this instrument uses passive remote sensing techniques to measure solar radiation reflected and emitted from the Earth. The spectrometer performs imaging in various ranges of the electromagnetic spectrum: ultraviolet (UV), visible (VIS), near infrared (NIR) and shortwave infrared (SWIR) (

Table 2. Main characteristics of TROPOMI spectral bands [14].

Band		Spectral Coverage (nm)	Span (km)	Spectral Resolution (nm)	Temporal Resolution	Spatial Resolution (km2)
UV	1	270–320	2600	0.49	Diary	7 × 28
	2					7 × 3.5
VIS	3	320–495		0.54
	4
NIR	5	675–775		0.38
	6
SWIR	7	2305–2385		0.25		7 × 7
	8

). In the study, monthly TROPOMI CO, NO₂, and SO₂ concentrations between 2019 and 2023 in the study area were extracted using JavaScript code from the GEE platform. Then, these data were used by taking annual averages.

The Python programming language provides powerful tools for data analysis and visualization. Python libraries Pandas 2.1.4, Matplotlib 3.8 and Seaborn v0.13.0 were used for data analysis and visualization. Data loading, cleaning and analysis operations with Pandas, data visualization and graph creation with Matplotlib, and more aesthetic and information-filled visualization of the data built on Matplotlib with Seaborn were provided.

GEE is an open-source web platform for the cloud-based processing of RS data developed by Google. It allows analyses on its own web interface. While Sentinel 5P-TROPOMI data are provided as Level 2 (geometrically corrected) on the Copernicus platform, GEE converts them into Level 3 (orthorectified) data by grouping them in latitude/longitude instead of time [45].

The reason why this method is preferred in this study is that data security is high and data losses are minimized during the data transformation phase. GEE uses a smaller pixel size (1 km) than the original resolution (3.5 km × 7 km) to prevent data loss when converting the original Level 2 data into Level 3 data during data conversion. The spatial resolution therefore corresponds to 1113.2 km [45].

The monthly and annual average TROPOMI NO₂, CO, and SO₂ concentrations as Level 3 data and MODIS LST values for the study region for the years 2019–2023 were obtained using JavaScript code in GEE. The data were then analyzed using Python on the Google Colaboratory platform Code block, downloading the average pollutants of the study area for each month between 2019 and 2023 in csv format (NO₂ example):

Map.centreObject(table)

var years = ee.List.sequence (2019, 2023);

var months = ee.List.sequence (1, 12);

var data = ee.ImageCollection (“COPERNICUS/S5P/OFFL/L3_NO2”)

According to the data obtained by GEE,

Table 3. Lowest and highest values of pollutants and years of observation.

Air Pollutant	Highest Value		Lowest Value
	5-Year High Value	Year Observed	5-Year Low Value	Year Observed
CO	32.4 × 10−3 mol/m2–34.4 × 10−3 mol/m2/2021	2021	29.4 × 10−3 mol/m2–31.1 × 10−3 mol/m2/2022	2022
SO2	4.2 × 10−4 mol/m2–7.7 × 10−4 mol/m2/2021	2021	3.6 × 10−4 mol/m2–5.9 × 10−4 mol/m2/2019	2019
NO2	7.4 × 10−5 mol/m2–10.2 × 10−5 mol/m2/2023	2023	6.7 × 10−5 mol/m2–8.6 × 10−5 mol/m2/2020	2020

shows the lowest and highest values of the pollutants in the study area and the years in which they were observed.

Maps of air pollutant concentrations superimposed on land classification for the Kepez district for the years 2019–2023 have been created. It is seen that the CO concentration is highest in the residential areas on the southern and southeastern borders of the Kepez district where traffic is intense, and the lowest concentration is in the forested and agricultural areas in the northern part. The general trend in all maps is that CO concentrations are higher in the southern part of Kepez, and the concentration decreases toward the north (Figure 4). The SO₂ concentration ranges from high to low across the following years: 2021, 2020, 2023, 2022, and 2019. In general, when the maps are examined, it is observed that low NO₂ concentrations are concentrated in natural areas such as forested areas and pastures, while high NO₂ concentrations are common in urbanized areas. It was also determined that agricultural areas also contribute to high NO₂ concentrations (Figure 5).

The highest SO₂ concentration of the five years was measured in the Teomanpaşa and Ulus neighborhoods (0.00025215 mol/m²) in 2023, and the lowest SO2 concentration was observed in the Teomanpaşa neighborhood (0.0001380 mol/m²) in 2020 (Figure 6). The highest values in terms of NO₂ concentration were detected in the Yavuz Selim, Kütükçü, Barış, Kuzeyyaka, Yeni Emek and Atatürk neighborhoods (0.0001032 mol/m²) in 2023. The lowest NO₂ value was measured in the Kızıllı neighborhood (0.00007689 mol/m²) in 2020. The highest values in CO concentration were detected in the Altınova Sinan (0.03438 mol/m²), Göksu (0.03435 mol/m²), and Altınova Orta (0.03432 mol/m²) neighborhoods in 2023, while the lowest values were measured in the Başköy (0.02967 mol/m²), Kızıllı (0.02983 mol/m²), Odabaşı (0.02997 mol/m²), Duacı (0.03002 mol/m²) neighborhoods. In terms of LST, the highest values were measured in the Altınova Sinan, Göksu, Altınova Orta, Altınova Düden, Beşkonaklılar and Baraj neighborhoods, and the lowest values were measured in the Başköy, Kızıllı, Odabaşı, Duacı and Kirişçiler neighborhoods.

3.4. Air Pollution in the COVID-19 Pandemic in Kepez

The restrictions related to preventing the spread of COVID-19, which first appeared in Türkiye in March 2020, started in April 2020 and had positive effects on air pollution in Türkiye, as in every country [42]. The five-year low value of NO₂ concentration in Kepez was measured in 2020 (6.7 × 10⁻⁵ mol/m²–8.6 × 10⁻⁵ mol/m²), when people had to stay at home. However, CO and SO₂ were not found to be significant in the COVID-19 pandemic process. The five-year highest value of CO was measured in 2021, and the lowest value was measured in 2022. The five-year highest value of SO₂ was measured in 2021, and the lowest value was measured in 2019. The reason for this situation can be interpreted as the use of domestic fossil fuels for heating purposes in Kepez. However, detailed analyses and studies are necessary for further interpretation.

3.5. Land Surface Temperature (LST)

MODIS is an instrument on the Terra and Aqua satellites. Terra MODIS and Aqua MODIS image an area approximately three hours apart. It collects data in 36 spectral wavelength groups [52]. MODIS has tasks such as monitoring atmospheric components, land cover, sea surface and clouds. MODIS provides daily daytime and night-time LST values at 1 km spatial resolution covering an area of 1200 × 1200 km [53]. LST estimates are obtained with a split-window algorithm using atmospherically corrected longwave infrared channels (bands 31 and 32) [53,54]. LST can provide fundamental information about surface properties and the environment. Thermal infrared (TIR) measurements from satellite sensors are often used to estimate LST on a regional or global scale [55,56]. In this study, LST values of the study area were calculated from MODIS satellite data. Figure 7 shows the graph of monthly average LST values of the study area for five years. Throughout all years, the temperatures show a rising trend starting from January, reaching the highest levels in July or August and then decreasing again. Temperatures peak around 45 °C in July and August and drop to around 10 °C in January and December.

Code block for downloading the 2023 average LST value of the workspace in TIFF format:

var dataset = ee.ImageCollection (‘MODIS/061/MOD11A1’)

        .filter (ee.Filter.date (‘1 January 2023’, ‘31 December 2023’));

var landSurfaceTemperature = dataset.select (‘LST_Day_1 km’);

var clippedLST = landSurfaceTemperature.mean ().clip (polygon)

3.6. Analyses with GIS Tool and Zonal Statistics

In the study, data analyses were performed using QGIS 3.38.3, which is an open-source software widely used in GIS applications for collecting, storing, organizing, analyzing and mapping geographic information.

QGIS offers its users a wide range of tools for processing, visualizing and analyzing data. Thanks to the possibilities offered by this software, complex geographic data can be managed and analyzed effectively. Especially in GIS analyses, QGIS’s powerful and flexible tools allow users to examine data in detail and present results visually.

In this study, the pollution and surface temperature data of Kepez district were analyzed using the zonal statistics tools of QGIS. Zonal statistics tools make it possible to identify differences in all neighborhoods of the Kepez district. These neighborhood-based analyses revealed the spatial distribution of regional pollution levels and surface temperatures in detail.

3.7. Statistical Analyses

R is a web and cloud-based, open source and free programming language for statistical calculations and graphical representations, which is organized according to the GNU certification system [57]. In addition to standard statistical applications, it is used in applications such as artificial intelligence, virtual reality, augmented reality, and machine learning, robotics [57,58,59].

The normalization process, which can be performed by applying different techniques, is frequently used to scale the data size by bringing it to appropriate ranges and to make the operations more efficient and easier [25]. In this method, which is one of the statistical data normalization techniques and widely applied, the mean and standard deviation values of the relevant data are used [25,59]. In this study, the data obtained from the Turkish Statistical Institute were subjected to factor analysis using R 4.4.3 studio software. The prepared data were normalized by the Z-score normalization method for 68 neighborhoods.

In order to evaluate the suitability of the organized data for factor analysis, a Kaiser–Meyer–Olkin (KMO) test and Bartlett’s test of sphericity were applied in R Studio software. The KMO test is another statistical test method used to determine whether the data set is suitable for factor analysis. If the KMO coefficient is small, it is not appropriate to proceed with factor analysis. A KMO coefficient approaching 1 indicates the success and suitability of the test, while any coefficient below 0.5 indicates the failure of the test and unsuitability for factor analysis [25,60]. Since the KMO test coefficient was 0.78 in the study, the sample size was sufficient. The model was found to be suitable for factor analysis. Bartlett’s test of sphericity is used to see whether the correlation matrix is a unit matrix with all diagonal elements 1 and off-diagonal elements 0. Barlett’s test is expected to be less than 0.05 [25]. In the study, the result of Bartlett’s test is significant, since the significance value is 2.2 × 10⁻¹⁶ < 0.05.

After normalization of the data, factor analysis was performed. Factor analysis allows an understanding of the structures in the data set by considering the relationships between variables. It was preferred as it is an ideal method to categorize the components of social vulnerability and to reveal the main factors affecting vulnerability in the Kepez district. It is based on social and environmental variables such as education, age, population, marital status and air pollution.

The eigenvalues of the 16 variables included in the analysis were analyzed. “Kaiser criterion” was used to determine the number of factors. According to this criterion, factors with eigenvalues greater than 1 were considered significant. The first factor explains 63.706% of the total variance, the second factor explains 78.055%, and the third factor explains 86.732% (

Table 4. Table of factor calculations of variables.

Variable	Factor				Initial Eigenvalues			Values After Factorization
	1	2	3	Factor	Total	Variance %	Cumulative %	Factor	Total	% Variance	Cumulative %
Illiterate Population	0.937	0.062	0.118	1	10.193	63.706	63.706	1	10.193	63.706	63.706
Secondary School Graduate	0.929	0.322	0.107	2	2.296	14.349	78.055
Under 15 Years Old Population	0.911	0.347	0.048	3	1.388	8.677	86.732
Primary School Graduates Population	0.903	0.395	0.112	4	0.776	4.849	91.581
Female Population	0.832	0.542	0.056	5	0.514	3.213	94.795
Household Population	0.746	0.647	−0.027	6	0.296	1.851	96.645
High School Graduate	0.701	0.698	0.046	7	0.221	1.38	98.025
Higher School Graduate	0.373	0.895	−0.101	8	0.144	0.899	98.925	2	2.296	14.349	78.055
Master’s Degree	0.05	0.841	−0.347	9	0.079	0.494	99.418
Over 65 Years of Age Population	0.445	0.756	0.139	10	0.047	0.291	99.71
Divorced Population	0.598	0.754	0.066	11	0.022	0.135	99.845
Population Density	0.27	0.736	0.307	12	0.01	0.065	99.91
Spouse Deceased Population	0.631	0.698	0.193	13	0.006	0.04	99.95
CO	0.006	0.054	0.907	14	0.005	0.03	99.98	3	1.388	8.677	86.732
SO2	−0.195	−0.06	−0.791	15	0.002	0.011	99.99
NO2	0.019	−0.064	0.7	16	0.002	0.01	100

). These findings show that the three factors provide sufficient information in the social vulnerability analysis.

After determining the number of factors, common variances were determined. Common variance values (communality), which is the amount of variance shared by each variable with other variables, were analyzed. According to the results obtained in R Studio, it was seen that the common variance values were at appropriate levels and the number of factors determined was sufficient.

The varimax rotation method was used to make the factors meaningful. The rotation process ensured that the correlations between the factors were distributed in an interpretable way. The rotated factor matrix showed in which factor group the variables were grouped.

The total, variance %, and cumulative % were calculated according to the initial eigenvalues of the variables. Then, the variables were grouped into three classes containing 7 in the first group, 6 in the second group and 3 in the third group. After 16 variables were reduced to 3, the total, variance %, and cumulative % were calculated for the three classes (Table 4).

4. Results and Discussion

4.1. Air Pollutant Concentrations and Mapping

Air pollutant (CO, NO₂, SO₂) concentrations were produced monthly with Sentinel-5P TROPOMI. Monthly averages were recalculated annually. The 5-year averages of the pixel-based data were taken, and these maps were overlaid with LULC. Zonal statistics were used to classify the data on a neighborhood basis. The annual averages of each neighborhood for all three air pollutants were plotted with Python. In this way, both the whole study area could be interpreted according to the land class distribution with 5-year data and the pollutant variables could be interpreted locally on a neighborhood basis. These data were also used in the Factor 3 SVI calculation and integrated SVI calculation.

In the general interpretation of the graph in Figure 8, it is determined that CO concentrations show small fluctuations over the years but are largely stable. It is read from the graph that the following years feature the highest concentration to the lowest: 2021, 2023, 2020, 2019, and 2022. The years 2020 and 2023 present very close average values. The highest values in CO concentration were detected in the Altınova Sinan (0.03438 mol/m²), Göksu (0.03435 mol/m²), and Altınova Orta (0.03432 mol/m²) neighborhoods in 2023, while the lowest values were measured in the Başköy (0.02967 mol/m²), Kızıllı (0.02983 mol/m²), Odabaşı (0.02997 mol/m²), and Duacı (0.03002 mol/m²) neighborhoods.

The LULC overlaid map with the 5-year average of CO is shown in Figure 9. This map is spatially similar to the CO annual maps in Figure 4. Pixels with high CO concentrations are observed in the southeast of the study area. There is a decrease in CO concentration toward the north and northwest. This situation coincides with the decrease in building areas and population density and increase in forest and pasture areas determined in the classification made with LULC. In the map, forest and pasture areas are inversely proportional to CO. The density of settlement centers in the southeast of the study area, high population density and high industrial activities in this region can be listed as the reasons for the high CO concentration.

The neighborhood-based graph of NO₂ annual average concentrations is presented in Figure 10. The general impression of the graphs is that the concentrations change over the years but exhibit a certain consistency between the neighborhoods. It is read from the graph that the following years feature the highest to the lowest concentration: 2023, 2022, 2021, 2019, and 2020. The highest values in terms of NO₂ concentration were determined in the Yavuz Selim, Kütükçü, Barış, Kuzeyyaka, Yeni Emek and Atatürk neighborhoods in 2023 (0.0001032 mol/m²). The lowest NO₂ value was measured in 2020 in the Kızıllı neighborhood (0.00007689 mol/m²).

The NO₂ (5-year average)—LULC map is presented in Figure 11. It is possible to interpret the map in three sections (high, medium, low) with an overview. Unlike the CO map, NO₂ concentrations are high in the south of the study area. The concentration of building areas in the south coincides with this situation. Moderate NO₂ concentrations are concentrated around pasture and cultivated areas. It is possible that this situation may be related to agricultural activities as well as being carried by wind from the south. Low concentrations are found in the north of the study area. Forests and bare land classes have the lowest NO₂ concentrations.

SO₂ measurement values do not show a regular distribution on a neighborhood basis (Figure 12). The year 2023 is observed as the year with the highest concentration. The neighborhoods with the highest concentrations in five years were measured in 2023 and are Teomanpaşa (0.00025215 mol/m²) and Ulus (0.00025215 mol/m²). In contrast, Teomanpaşa has the lowest concentration of five years (0.0001380 mol/m²) in 2020. Therefore, a general comment cannot be made on a neighborhood basis in terms of SO₂.

As in the neighborhood graphs, the SO₂-LULC map, where the 5-year average is taken, does not show a regular distribution (Figure 13). Unlike CO and NO₂, low concentrations are observed in the south and high concentrations are observed in the north. Moderate SO₂ concentrations are located in the middle part of the study area in the land class where agricultural and pasture areas are intensively observed. The fact that the lowest SO₂ measurement values are observed in different land classes of the study area prevents a general interpretation. However, the possibility of dispersion by wind can be taken into consideration.

4.2. LST-LULC Maps

The maps show the annual average LST intensity in the study area from 2019 to 2023 (Figure 14). The data obtained during this five-year period were used to examine the spatial and temporal distribution of surface temperatures. Maps created with LST values were overlaid on different land use classifications to create a five-year LST map output. With this method, the effect of different land use types on surface temperatures in the region was analyzed in detail.

When the maps are analyzed, it is seen that the surface temperature distribution has not changed in general for five years. However, small differences were observed between some years. The year with the highest LST value was determined as 2020. This may be due to certain climatic conditions or human activities that occurred that year. The southern region of the district shows generally higher temperatures over the years, while the northern region remains consistently cooler. It can be observed that land use types (pasture, forest, agricultural land, built-up area) have a significant effect on temperature distribution. In particular, built-up areas are associated with higher temperatures.

4.3. Integrated SVI Calculation and Mapping

The entire analytical process of the study was carried out using R Studio. Data normalization and missing data control were performed with dplyr and tidyr libraries. For factor analysis, psych and factoextra libraries were used to extract factors, eigenvalue analysis, calculation of variance percentages and factor rotation. For visualization of the results, the ggplot2 library was used to visualize the distribution of factor scores and vulnerability levels on a neighborhood basis.

The three factors and their factor scores were used to construct the Social Vulnerability Index (SVI) for the Kepez district. Factor scores were calculated according to the weights of the variables on the factors. These scores formed the basis for determining the vulnerability levels for 68 neighborhoods in the Kepez district.

Factor 1 incudes the Illiterate, Secondary School Graduate, Population Under 15, Primary School Graduate, Female Population, Household Population, and High School Graduate variables, Factor 2 includes the High School Graduate, Master’s Degree Graduate, Population Over 65, Divorced Population, Population Density, and Population with Deceased Spouse variables, and Factor 3 includes the air pollutants (NO₂, SO₂, CO) variables. The integrated SVI value was calculated from the sum of these three factors (

Table 5. Factor 1, Factor 2, Factor 3, and integrated SVI values table for neighborhoods.

Neighborhood Name	Factor 1	Factor 2	Factor 3	Integrated SVI	Neighborhood Name	Factor 1	Factor 2	Factor 3	Integrated SVI
Ahatlı	0.93	1.11	0.35	2.39	Habibler	1.56	0.42	0.37	2.34
Aktoprak	0.34	−0.03	0.06	0.36	Hüsnü Karakaş	1.44	0.40	0.35	2.19
Altıayak	−0.81	−0.87	0.18	−1.50	Kanal	0.41	0.62	0.29	1.32
Alt. Düden	−0.71	−0.68	0.35	−1.05	Karşıyaka	0.93	0.95	0.33	2.21
Altınova Orta	−0.24	−0.33	0.33	−0.24	Kazım Karabekir	−0.98	−0.78	0.33	−1.43
Altınova Sinan	−0.38	−0.52	0.33	−0.58	Kepez	−0.85	−0.38	0.38	−0.85
Atatürk	−0.16	0.41	0.32	0.57	Kızıllı	−0.96	−0.89	0.21	−1.64
Avni Tolunay	−0.76	−0.58	0.33	−1.01	Kirişçiler	−1.08	−0.95	0.30	−1.74
Ayanoğlu	0.24	−0.05	0.32	0.51	Kuzeyyaka	2.26	1.07	0.21	3.54
Aydoğmuş	0.41	0.12	0.25	0.78	Kültür	0.48	1.63	−0.06	2.06
Baraj	0.10	−0.36	0.16	−0.09	Kütükçü	−0.23	−0.10	0.05	−0.27
Barış	−0.26	−0.10	0.35	−0.01	M. Akif Ersoy	0.54	0.20	0.35	1.09
Başköy	−1.14	−1.00	0.32	−1.82	Menderes	−0.78	−0.81	0.37	−1.21
Beşkonaklılar	−0.82	−0.79	0.18	−1.44	Odabaşı	−1.11	−0.95	0.32	−1.74
Çamlıbel	−0.24	−0.11	0.34	−0.01	Özgürlük	0.40	1.49	0.37	2.26
Çamlıca	−1.05	−0.73	0.27	−1.51	Santral	−1.16	−0.86	0.09	−1.93
Çankaya	0.13	0.80	−0.03	0.90	Sütçüler	0.16	−0.05	−0.04	0.07
Demirel	−0.75	−0.71	0.25	−1.22	Şafak	1.10	0.36	0.35	1.80
Duacı	−0.92	−0.48	0.18	−1.21	Şelale	0.57	−0.11	0.16	0.63
Duraliler	−0.76	−0.48	0.09	−1.15	Teomanpaşa	2.03	1.07	0.35	3.46
Düdenbaşı	1.23	0.54	0.25	2.03	Ulus	0.55	1.59	0.31	2.46
Emek	0.13	0.21	0.37	0.70	Ünsal	0.50	0.12	0.28	0.90
Erenköy	0.39	0.41	0.40	1.20	Varsak Esentepe	−1.02	−0.92	0.38	−1.56
Esentepe	−0.37	−0.33	0.31	−0.39	Varsak Karşıyaka	0.40	0.16	0.37	0.93
Fabrikalar	−0.07	0.38	0.25	0.57	Varsak Menderes	−1.00	−0.95	0.18	−1.77
Fatih	−0.37	−0.37	0.20	−0.55	Yavuz Selim	−0.97	−0.80	0.30	−1.47
Fevzi Çakmak	0.35	0.13	0.36	0.85	Yeni	0.63	0.50	0.36	1.49
Gazi	0.59	0.22	0.18	0.98	Yeni Doğan	−0.08	0.59	0.34	0.85
Gaziler	−0.70	−0.84	0.27	−1.27	Yeni Emek	0.74	0.46	0.38	1.58
Göçerler	−0.19	−0.40	0.33	−0.26	Yeşiltepe	0.17	0.94	0.35	1.46
Göksu	−0.23	−0.09	0.17	−0.15	Yeşilyurt	0.17	0.50	0.35	1.02
Gülveren	−0.62	−0.22	0.18	−0.66	Yükseliş	−0.03	0.46	0.35	0.78
Gündoğdu	0.99	0.59	0.35	1.93	Zafer	−0.24	0.67	0.35	0.78
Güneş	1.64	0.23	0.36	2.22	Zeytinlik	−0.46	−0.70	0.30	−0.85

).

The three factors and integrated factor values of the calculated SVI values were mapped in the neighborhoods in the study area (Figure 15). As can be seen from the maps, each factor contains different dynamics. Very high and high SVI values are different in each map on a neighborhood basis. This explains the purpose of conducting the study more locally (on a neighborhood basis). The distribution of classes according to neighborhoods is given in

Table 6. Class distribution of the neighborhoods of the variables.

Variable	Number of Neighborhoods Belonging to Classes
	Very High	High	Moderate	Low	Very Low
Factor-1	14	12	13	15	14
Factor-2	14	13	14	13	14
Factor-3	12	14	15	10	17
Integrated SVI	14	12	15	13	14

.

Integrated SVI is an index where both social factors and air pollution variables are calculated together in the study area. In the study area, 14 neighborhoods (Şafak, Ahatlı, Kültür, Özgürlük, Ulus, Yeniemek, Karşıyaka, Kuzeyyaka, Gündoğdu, Teomanpaşa, Habipler, Hüsnükarakaş, Güneş, and Düdenbaşı) have very high SVI values, and 12 neighborhoods (Ünsal, Yeşilyurt, Yenidoğan, Kanal, Erneköy, Yeşiltepe, Yenimahalle, Mehmet Akif Ersoy, Gazi, Varsak Karşıyaka, Çankaya, Fevzi Çakmak) have high SVI values. This determination reveals that priority measures should be taken in these neighborhoods. In addition to the determination of air pollution at the micro level, important determinations have also been made to ensure that socioeconomically disadvantaged individuals are affected in the same way as other individuals in case of disaster.

Our neighborhood-based methods and findings are consistent with previous studies. Air pollution was compared between individual and neighborhood socioeconomic status [32], and it was found that the neighborhood-based socioeconomic group had a strong correlation with air pollution. This provides proof that mass determinations should be made rather than individual determinations in disaster situations.

On the other hand, it was found that social vulnerability and stillbirth were corre-lated with environmental risk factors (air pollution) at a district level [33]. This study also emphasizes the importance of mass surveys rather than individual detection.

Although there are studies applying the Pressure and Release (PAR) model [61] in research on climate disasters, this model has not been applied yet despite the presence of more than one fracture factor in the study area. In this context, GIS and RS studies should be developed.

With the data obtained in the study, social services, health units, and employment should be increased in very high and high-risk neighborhoods, and information against natural disasters should be organized. In addition to social measures, the effects of air pollutants should also be reduced. Efforts should be organized to reduce the use of fossil fuels in industrial activities, industrial facilities and households. The most important social and structural elements for understanding the effects of environmental pollution include education level, income status, social trust and belief in other members of the society [27,62]. Communities that value sustainability and environmental responsibility can motivate individuals to take action to mitigate climate change [63].

Vulnerability can be defined as the extent to which an individual or population is affected by various environmental hazards [27,64]. Efforts to mitigate climate change and global warming in cities and municipalities are critical, and sustainable cities and eco-cities play an important role in the equation [63]. This study, in which the Kepez district, where vulnerable individuals live in large numbers, was analyzed in terms of air pollutants, is an important guide to understand the spatial distribution of air pollution and the factors affecting the concentration of pollutants. Such studies can form the basis for strategic planning to reduce the impact of air pollution on public health. Measures to improve air quality will not only provide environmental benefits but also improve the quality of life of people living in Kepez. In this context, sustainable urban planning and environmentally friendly transport policies will play a critical role in reducing pollution levels and achieving zero carbon in the long term.

5. Conclusions

The study aimed to determine the risks posed by fossil fuel-derived air pollutants in the study area of Kepez district of Antalya province by using UA, GIS technology and open-source tools. According to its social vulnerability, the study was carried out specifically for the region having the highest potential district. The selected district consists of 68 neighborhoods. This amount of neighborhood data for the application facilitates statistical studies. The research seeks to answer the fundamental question of what is the relationship between air pollution (NO₂, SO₂, CO) and socially vulnerable populations. It also includes spatial data such as LST and LULC.

In this study, the versatile use of remote sensing was discussed. Monthly and annual averages of NO₂, SO₂, and CO data sets obtained from the Sentinel-5P TROPOMI satellite between 2019 and 2023 via the GEE platform were used. In addition, LST data were obtained with the MODIS instrument on the Terra and Aqua satellites. These data were processed for 68 neighborhoods in the Kepez district using the zonal statistics tool in QGIS software. An LULC map with 10 m spatial resolution created with Sentinel-2 satellite images from the ESRI platform and neighborhood population data from TurkStat were obtained as well.

Separate NO₂, SO₂, and CO maps for 5 years were created by overlapping with LULC. Thus, interpretations were made according to land classes on a pixel basis. Surface temperatures in land classes were also evaluated in the LST-LULC map. Then, new maps were created by taking 5-year average values of NO₂, SO₂, and CO pollutants. These values were used in Factor 3 of the SVI calculation.

R software was used to calculate the statistics. The integrated SVI value to be used in the study was calculated by adding the three factor values. Neighborhood-based factor maps were obtained. Accordingly, a five-stage classification was created in the maps. In the integrated SVI map, it was determined that there are 14 neighborhoods with very high status and 12 neighborhoods with high status.

In developing countries such as Türkiye, industrial plant activities are high. Low-income communities are more likely to be located near environmental hazards such as waste facilities and factories [30,65,66,67]. Significant differences are observed between the 5-year average map of NO₂, SO₂, and CO pollutants used in the study and the integrated SVI map. This is proof that the focus of the study should not be only air pollution. This situation shows the necessity of evaluating many variables together. In addition, the integrated satellite monitoring of air pollution, which is currently measured by ground stations, will enable detection in larger areas.

In the context of global climate change, there is an opportunity to assess and reduce exposures among the most vulnerable to reduce inequality in health problems and promote health equity [33]. In the measures to be taken by decision-makers against climate change, priority should be residential centers with low SVI. As in this study, local factors such as neighborhood should be evaluated. This will make things much easier in developing preparation and action plans against disasters. Future research needs multi-regional applications of SVI, extending it with different models such as PAR [61] and measurements comparing ground and satellite data. In addition, since air pollution is defined as a long-term disaster, this study has the characteristics of reducing the effects of disasters.