Mapping Air Quality Using Remote Sensing Technology: A Case Study of Nairobi County ()
1. Introduction
Worldwide, air pollution continues to pose a major threat to public health [1] [2] [3] . The World Health Organization (WHO) estimates that air pollution claims the lives of almost 7 million people annually (WHO, n.d [4] ). Numerous countries have conducted studies on the impact of air pollution on human health, with a focus on respiratory issues [5] and, in severe cases, results in mortality [6] [7] . Air pollution exacerbates respiratory ailments such as lung cancer, bronchitis, pneumonia, and asthma among others [8] . Countries around the world must, therefore, implement best practices to reduce and control the negative effects caused by air pollution.
According to the World Health Organization, air pollution is “contamination of the indoor and outdoor environment by chemical, physical or biological agent that modifies the natural characteristics of the atmosphere” [9] . Air pollutants have been broadly classified as primary and secondary pollutants [10] [11] . Primary air pollutants originate from their sources and are released directly into the atmosphere, while secondary air pollutants are created when primary pollutants react with other substances, like ozone. Examples of primary pollutants include nitrogen dioxide (NO2), carbon monoxide (CO), Sulphur dioxide (SO2), and total suspended particles (TSP). On the other hand, ozone (O3) and particulate matter (PM) are examples of secondary pollutants [12] .
Air pollution sources are of four main categories. They include; mobile, stationary, area, and natural sources [13] [14] . Mobile sources refer to vehicular emissions as well as from airplanes while stationary air pollution sources include emissions from power plants, industries, factories, and oil refineries among other fixed stations. Area sources include agricultural areas and cities while natural causes include wind-blown dust, wildfires, and volcanoes as examples [15] .
In a publication by the United Nations Environment Programme (UNEP) titled Air Pollution Hurts the Poorest Most, (n.d [16] ), it was highlighted that negligence by citizens in most developing countries in upholding regulations around air pollution has led to many recorded deaths associated with atmospheric pollution. In most developing countries, ambient air pollution is mostly attributed to industrialization and urbanization which poses a serious health threat to their ever-growing population [1] [17] [18] .
Similarly, Nairobi’s population is more affected by ambient air pollution than those in other parts of the country [19] . With 4.4 million residents as of 2019, Nairobi is the most populous city in the nation and one of the fastest-growing cities in the world [20] . It also experiences intense traffic situations with a high prevalence of old, unmaintained vehicles running on low-quality fuel with high Sulphur content, which contribute largely to the unclean atmospheric conditions [21] . The increasing population in Nairobi faces health threats caused by air pollutants and as such, it calls for management strategies to be put in place. Nevertheless, before the said strategies are set, it is important to acquire location-based information on air pollution occurrences. One such way is through mapping the air quality situation.
2. Situational Analysis of Air Quality in Nairobi County
Nairobi County experiences rapid industrialization and urbanization [22] . Nairobi County’s air quality is deteriorating at a startling rate. Studies done specifically in the slums of Nairobi have uncovered unhealthy levels of pollution for sensitive groups such as children, who do not have fully developed lungs to contain the detrimental effects of PM particles [23] [24] . These areas experience high population density [25] . Similarly, Nairobi’s traffic situation has been considered a prime mobile source of air pollution [26] [27] . The increase in motorized transport in the city has played a part in contributing to vehicular air pollution [28] . According to current registration rates, the number of vehicles in Nairobi alone is expected to exceed 1.35 million by 2030. With this number of vehicles moving in and out of Nairobi’s Central Business District (CBD), traffic snarl-ups are apparent. Nairobi’s air pollution problem is exacerbated by smoke from factories, industries, and other plants in addition to vehicle emissions [29] [30] . As it stands, the Nairobi air pollution problem has not been sufficiently addressed despite it being a major environmental disaster [31] .
In Nairobi County, the majority of air quality stations currently utilize low-cost sensors (LCS), a cost-effective option chosen for their affordability [19] . However, the utilization of such sensors introduces certain challenges regarding data accuracy and reliability due to occasional defects that may occur as reported on the Nairobi IQAir website (https://www.iqair.com/kenya/nairobi). LCS used in air quality monitoring is affected by environmental factors, necessitating costly periodic calibration to ensure data accuracy [32] . Reference monitors, also referred to as high-accuracy monitors, are costly devices that can cost up to US $100,000 and are utilized for regulatory purposes in many countries [33] [34] .
The adoption of various mechanisms especially by bodies such as Kenya’s National Environmental Management Authority (NEMA) is essential in coming up with policies and regulations on air pollution and enforcing them. Kenya, as of 2021, lacks a standalone national policy on air pollution, but it addresses this issue through broader environmental policies. The Environmental Management and Coordination (EMCA) Regulations were implemented by the Environment Minister in 2014 to address the problem of air pollution and to establish acceptable air pollution limits for areas used for residential, industrial, and occupational purposes.
These standards have been Kenya’s main concern in containing the issue of air pollution in the country. However, initiatives on the issuance of air quality monitors and the creation of publicly available air quality data have been slow [19] . According to a NEMA statement, as of 2010, only two designated laboratories were listed as being capable of monitoring air quality, which is insufficient for the needs of the nation. Additionally, only a small number of designated laboratories are capable of monitoring emissions on behalf of industries [35] . It is for these reasons that the Government of Kenya has been unable to enforce the regulations.
It is safe to conclude that the quality of the measurement and the number of monitoring stations in Nairobi are not sufficient to accurately depict the air pollution situation over the entire county. It is practically impossible to fit monitoring stations everywhere as it would be a laborious and expensive undertaking. Nairobi alone would be economically unviable if more air quality monitoring stations were to be set up. A generalized representation of air quality information can be a means to solve this problem, especially by the use of remotely sensed data that offer a synoptic view.
3. Spatial Indicators for Air Quality Mapping
Various spatial indicators have made it possible to visualize the air quality situation in an area. For instance, aerosol optical thickness which is an indicator of atmospheric pollutant concentration, has been used in air quality studies as a measurement of pollution [36] . The retrieval of aerosol optical depth, extraction of API values, visibility studies quantified in terms of atmospheric turbidity, and identification of suitable Pseudo Invariant Targets (PITs) in satellite images are common spatial indicators used for air quality mapping that rely on remotely sensed images [37] .
An understanding of some of the main causes of air pollution is required to map air quality accurately. The Air Quality Action Plan 2019-2023 report, Table 1 gives a summary of the components of air pollution.
4. Remote Sensing for Air Quality Mapping
The use of remotely sensed data to evaluate air quality has been the subject of numerous studies. The widespread recommendation of remote sensing is primarily attributed to its synoptic characteristic, providing a comprehensive view of a vast area, its high temporal resolution, allowing for frequent monitoring over time, and the continuous surface representation of imageries [39] . In studies where ground monitoring stations are few or nonexistent, remote sensing data has been used to fill in the existing gaps, especially for rural air quality mapping endeavors [40] .
It became evident that air pollution could be mapped using remote sensing when the Measurement of Air Pollution from Satellite (MAPS) instrument, which was launched in 1981 on the space shuttle Columbia, provided the first measurements of high concentrations of CO over the continents of Asia, Africa, and South America [41] . The Total Ozone Mapping Spectrometer (TOMS) and Stratospheric Aerosol and Gas Experiment (SAGE) instruments on the Nimbus 7 satellite measured tropospheric ozone, demonstrating the global nature of air pollution [42] . These were among the earliest attempts to map air quality at a global scale using remotely sensed imageries.
Satellite sensors orbiting in outer space can map aerosols. Examples provided by both Lelieveld et al. [43] and Kaufman et al. [44] , portrayed the worldwide
Table 1. Air pollutants and their impact on health and the environment.
(Modified from source: Nairobi City County, 2019 pg. 11-12 [38] ).
distribution of both naturally occurring and man-made aerosols, collected from a Moderate-Resolution Imaging Spectrometer (MODIS) on the TERRA satellite.
Air pollution as a phenomenon would be more understood by use of satellite imageries as was in the case of El-Askary [45] , where AOD was retrieved from satellite data confirming dust and air pollution events in Cairo, Egypt. In the study conducted by El-Nadry et al. [46] , in mapping the air quality of the North African region, they concluded that satellite data would suffice where ground recording stations were not set up. Also from their study, they exploited the use of TROPOMI among other data sources such as MODIS, which was documented to be the first study to ever exploit data by Sentinel-5P in the northern part of Africa.
More researchers are developing different approaches to depict and comprehend the state of air pollution in African nations, such as employing artificial neural networks and machine learning techniques [47] [48] .
5. Mapping Air Quality for Nairobi County
5.1. Area of Study Description
Kenya’s capital and largest city is Nairobi. It is situated at 1˚9'S, 1˚28'S, 36˚4'E, and 37˚10'E (Figure 1). The city is located roughly 200 kilometers south of the Equator, covering 696 square kilometers in total, and rises to an altitude of about 1700 meters above sea level. Nairobi County consists of 17 administrative sub-counties, which are as follows: Kamukunji, Kasarani, Kibra, Langata, Makadara, Mathare, Roysambu, Ruaka, Starehe, and Westlands. South and North Dagoretti, Embakasi Central, Embakasi East, Embakasi North, Embakasi South, and Embakasi West. As of 2019, the estimated population of Nairobi County was 4.3 million, which is growing at a 4% annual rate [49] . Rapid urbanization has been linked to Nairobi County’s exponential population growth as more people are drawn to the area in quest of job opportunities [50] . Kenya’s capital is one of the developing cities where the majority of people reside in areas with polluted air [51] . Air pollution in Nairobi is a major problem, with mobile sources being the main source of pollutants [52] . Apart from the transient sources, Nairobi’s air pollution has also been caused by pollution from stationary sources [29] . Despite this, Nairobi does not have an inventory of air quality, and studies that are limited to certain pollutants are insufficient to determine the county’s overall level of atmospheric quality [38] .
5.2. Data Sources and Collection
The primary data used was Landsat imageries as outlined in Table 2. Details of the Landsat types and band characteristics are well documented in published journals [53] .
Nairobi county boundary shapefile was obtained from the Independent Electoral and Boundaries Commission (IEBC) database in shapefile format (.shp).
Figure 1. Study area map of Nairobi County.
Table 2. Landsat imagery characteristics used for mapping air quality in Nairobi County.
Source: Google Earth Engine catalog.
5.3. Procedure for Mapping Air Quality for Nairobi County
The study utilized Landsat imagery from three different epochs (2000, 2010, and 2018) acquired from the Google Earth Engine (GEE) cloud server (Figure 2). Simple JavaScript codes were employed to access and retrieve the images, which were subsequently filtered based on cloud metadata, date, and the geographical extent of the area of interest (Nairobi). These filtered images were then exported to a common Google Drive and later saved on the personal computer storage for
Figure 2. A systematic flowchart describing the processes taken in mapping air quality in Nairobi County from Landsat imagery acquisition, processing, extraction of vegetation indices, and computation of air pollution index (API) and visualization of air pollution maps.
further analysis.
In the subsequent steps, the acquired Landsat images were integrated into the QGIS interface [54] . Vegetation indices, namely the Normalized Difference Vegetation Index (NDVI), Transformed Vegetation Index (TVI), and Difference Vegetation Index (DVI), were computed using the QGIS raster calculator. Additionally, the Air Pollution Index (API) classification was computed using the raster calculator tool in QGIS.
For qualitative accuracy assessment, five (5) sample points were drawn from each class representing residential, industrial, green spaces, and traffic hotspots, based on a passive-random sampling in Google Earth Pro software. These sample points were overlaid with the classified API rasters, and API values were extracted from each raster using the raster extraction tool in QGIS. To adhere to the API classification chart used in the study, the calculated API rasters were then reclassified.
Finally, a qualitative accuracy assessment was conducted for each of the sample stations, and deductions were drawn from the obtained API values. This comprehensive process allowed for an in-depth analysis of air quality dynamics in Nairobi and enabled the drawing of valuable conclusions based on the classified imagery and sampled data.
5.3.1. Sampling Data
The data that formed part of the qualitative accuracy assessment approach was sampled across Nairobi County using a passive-quasi-random sampling technique.
5.3.2. Extraction of Vegetation Indices
This was done by the use of QGIS raster calculator. The following vegetation indices were extracted.
1) NVDI (Normalized Difference Vegetation Index) [55] :
(1)
2) TVI (Transformed Vegetation Index) [56] : served to transform the values from NDVI into a normal distribution:
(2)
3) DVI (Difference Vegetation Index) [57] :
(3)
API was obtained from the radiometric reflectance values of NIR and SWIR11 and the vegetation indices (i.e., DVI, TVI).
The API function is as follows [58] :
(4)
5.3.3. API Classification and Visualization
After the API was computed for every Landsat imagery used, they were reclassified as shown in Table 3.
The sample points were then overlaid on every classified imagery. This was done to give reference and aid in the interpretation of the spatial visualization. The overlaid sample points were then used to extract API values using the bi-linear interpolation method and the results were tabulated. The raster-to-point tool in QGIS was exploited and later the results were saved as CSV files.
6. Results
6.1. Air Pollution Information from Landsat 7 Imagery
Varying API values were recorded in Nairobi county in the year 2002 (Landsat 7 imagery) across different sample categories. In the residential category, Karen had the lowest API value (50.36472), while Parklands had the highest (121.6216) Karen exhibited the most favorable air quality category (“good”2), while Parklands had the least desirable air quality (“unhealthy for sensitive groups”).
Kamulu and Ruai were observed to fall under “moderate” air quality with API values of 100.4715 and 92.50559 respectively. Among traffic hotspots, Roysambu stage showed moderate pollution (90.0007), while Koja Bus Station, Railways Bus Station, and Lusaka Road had higher API values. Bamburi Limited was visually observed to have relatively better air quality (“moderate”3) with the lowest API value in the industrial category (81.39734), while Industrial Area, EABL, and Acme Limited showed higher API values. Green spaces presented mixed results in comparison to the visual observations, with Ngong Race Course indicating more favorable air quality (“good”) with the lowest API value (41.48645), while Bomas of Kenya and Marurui recorded higher API values.
Ref: Table A1 (annexed).
Figure 3 shows the air pollution categorized in API values as highlighted in Table 3. The result is from the extraction of API values from the Landsat 7 scene for Nairobi County in the year 2002.
6.1. Air Pollution Information from Landsat 5 Imagery
The investigation revealed that residential areas exhibited comparatively lower
Table 3. Air Pollution Index classification scheme.
Modified excerpt. Source: [59] .
Figure 3. A map showing the air pollution (API) categorization in 2002 for Landsat 7 sample imagery.
levels of air pollution, whereas traffic hotspots and industrial zones displayed elevated pollution levels. Contrary to expectations, green spaces demonstrated higher pollution levels when compared to residential areas. The majority of API values across the sampled areas fell within the classification of “unhealthy for sensitive groups.”4
The results depict varying air pollution levels across different areas in Nairobi. Residential areas displayed an average API value below 100, while Parklands exhibited the highest pollution at 139.764. Traffic hotspots and industrial zones consistently had higher pollution levels, with API values above 115 (traffic hotspot average: 125.1828) and EABL recorded the highest industrial API value at 132.4025. Intriguingly, green spaces also showed elevated pollution, with API values exceeding other areas (e.g., Nairobi National Park: 93.56007, Karura forest: 114.90893, Marurui: 168.5034) ref: Table A1 (annexed).
Figure 4 shows the distribution of air pollution over Nairobi County in 2010. Landsat 5 imagery was used as the primary data.
6.2. Air Pollution Information from Landsat 8 Imagery
Figure 5 shows results obtained using four different Landsat 8 imageries to
Figure 4. A map showing the air pollution index (API) categorization in 2010, Landsat 5 sample imagery.
represent the extracted API values. A different color scheme (sequential) was used for the classification rather than that described in Table 3.
Amongst the API values obtained from the image samples, none of them made logical sense as most of the sample points registered negative values. The API code provided proved indifferent to the use of Landsat 8 imagery.
Normalization of API values for Landsat 8 imagery, year 2018
A sample API data for Landsat 8 for the year 2018 as shown in Figure 5, was normalized using the expression illustrated herein and later reclassified according to Table 3.
Stretching the data to 8-bit
smin = 0; smax = 255
(x − min(x)) * (smax − smin)/(max(x) − min(x)) + smin
NewRaster = (OldRaster − −1) * 255/(1 - −1) + 0
source: (online)
API representation and values for normalized sample Landsat 8 imagery (2018).
Figure 6 shows the result obtained from normalizing the API values to fit the 8-bit value range.
The values obtained after the normalization of the data for the 20 sample
Figure 5. Using Landsat 8 imageries for air pollution index (API) representation.
stations ranged from 106.3265 to 185.7503 as shown in Table A1 (annexed) implying that the 20 samples fall under unhealthy for sensitive groups and unhealthy categories.
7. Discussions of the Results
7.1. Discussions of API Classified Imagery from Landsat 7 for the Year 2002
Residential areas like Karen exhibited lower API values, potentially due to the surrounding green space in the area. In contrast, Parklands’ higher API value can be linked to its proximity to the CBD, with interconnecting residential roads having increased traffic congestion during peak hours. The moderate API values in Kamulu and Ruai, both residential regions, suggest a moderate level of pollution, potentially influenced by moderate traffic density and limited industrial activities.
The variations in API values among traffic hotspots can be attributed to several potential reasons. Roysambu stage’s moderately polluted air could be due to a combination of moderate traffic congestion and limited industrial activities in the area. Koja Bus Station and Railways Bus Station showing similar API values in the “unhealthy for sensitive groups” category may result from their proximity to major roadways, leading to higher vehicular emissions and air pollution. Lusaka Road’s highest API value in the same category could be attributed to heavy traffic flow and with its location being nearer to the industrial area, contributing
Figure 6. Air pollution index (API) categorization map for Landsat 8 in the year 2018-normalized API values.
to increased pollutant concentrations.
In the industrial sample type, Bamburi Limited recorded the lowest API value, showing moderate pollution, while the Industrial Area had the highest API value, placing it in the category of “unhealthy for sensitive groups.” EABL and Acme Limited also fell under the latter category, indicating significant air pollution levels associated with industrial activities.
Unlike the air pollution representation in Landsat 5 API classified imagery (Figure 4), Landsat 7 API classified image 2002, is partially in line with the assumptions made for green spaces. The assumption being that greener spaces are likely to register lower API values as compared to other sample types. Ngong Race Course’s lower API value may be due to its open space nature with limited vehicular traffic and hardly any industrial activities. The moderate API values for the Open Space sample from Karen and Karura Forest could be influenced by vegetation and far distance from pollution sources. Conversely, higher API values at Bomas of Kenya and Marurui indicate potential nearby pollution sources affecting their air quality.
7.2. Discussions of API Classified Imagery from Landsat 5 for the Year 2010
Landsat API classified imagery for the year 2010 provided valuable information on air pollution levels across the different areas in Nairobi. Notably, residential regions exhibited comparatively lower pollution levels in contrast to traffic hotspots and industrial zones. However, it was observed that Parklands, despite being primarily a residential area, displayed the highest API value, signifying “unhealthy for sensitive groups” air quality. This phenomenon could be attributed to its proximity to the CBD of Nairobi, which also demonstrated similar pollution standards. Additionally, the traffic congestion experienced at road intersections within the Parkland area could serve as a contributing factor to the observed high API value.
Regarding traffic hotspots, the extracted API values consistently exceeded 115, with Koja Stage and Railways Bus Station displaying comparable high values. On the other hand, Roysambu Stage had a slightly lower API value within this category. The location of the Lusaka Road junction near Nyayo Stadium near industrial areas and heavy traffic flow might account for its higher API value. All traffic hotspot values fell within the “unhealthy for sensitive groups” category, highlighting the potential health risks associated with these areas.
The API values corresponding to the sampled areas with industrial activities accurately reflected the state of air pollution in these regions. EABL recorded the highest API value, closely followed by Industrial Area, Acme Ltd, and Chloride Ltd. Conversely, Bamburi demonstrated the lowest API value within this category. Bamburi, being a sub-outlet for the cement plant in Athi River, primarily dealing with cement supplies rather than manufacturing, appeared to have a positive influence on its lower API value. However, despite its location in a relatively polluted area, all industrial area API values fell under the “unhealthy for sensitive groups” category.
Conversely, the API values for green spaces presented mixed results, differing from the initial assumption that such areas would have lower pollution levels. The general impression from the extracted values indicated an unhealthy pollution level for the sampled green spaces. Even though Nairobi National Park registered the lowest API value, surpassing that of Karura Forest and Marurui, it was still higher than Simba Villas. This discrepancy raises questions about the API formula, potentially due to the influence of high vegetation indices values, since the methodology adopts vegetation as a pseudo invariant target.
7.3. Discussions of API Classified Imagery from Landsat 8 for the Year 2018
The values obtained after the normalization of the data for the 20 sample stations implied that the 20 samples fall under unhealthy for sensitive groups and unhealthy categories. These remarks are, however, not likely due to the different characteristics in the sample types. The values slightly differ from each other. Green spaces and residential sample types were expected to bring huge disparities as compared to traffic hotspots and industrial sample sites, which was not the case.
7.4. Practical Implications and Limitations of the Study
The study presents valuable insights into Nairobi County’s air quality using Landsat imagery to extract API values, enabling effective monitoring and targeted improvement strategies. The findings emphasize the significance of considering air quality in urban planning, particularly in high-pollution areas, and its implications on public health. The spatial data facilitates health impact assessments, identifying vulnerable populations in need of targeted interventions. Policymakers can utilize the API data to develop regulations and implement pollution-reducing measures.
However, the study faces limitations due to the temporal and spatial resolution constraints of Landsat imagery, potentially overlooking short-term fluctuations and fine-scale variations. The API formula’s suitability, especially when applied to Landsat 8 imagery, warrants further investigation. The absence of ground truthing through on-site air quality measurements may affect result reliability. Moreover, the study does not consider potential land-use changes, which could influence air pollution patterns over time. Nonetheless, it lays the foundation for future research and policy interventions to address air pollution in the region, emphasizing the importance of integrating remote sensing and ground-based data for comprehensive air quality assessments.
8. Conclusion
The application of remote sensing technology to map air quality in Nairobi County has provided valuable insights, particularly in regions where ground-based monitoring stations are scarce, a common challenge in many developing countries. The cost-effectiveness and comprehensive spatial coverage of this technology make it an invaluable tool for understanding pollution patterns. Our analysis revealed distinct pollution patterns, with residential areas exhibiting lower pollution levels compared to traffic hotspots and industrial zones. The consistent recording of high Air Pollution Index (API) values in traffic hotspots, exceeding 115 and categorized as “unhealthy for sensitive groups,” underscores the need for targeted interventions in these areas. Green spaces’ API values varied, contrary to initial assumptions, indicating challenges in the interpretation of pollution levels in these areas. To enhance the applicability of air quality assessments, we propose the recalibration of the API formula at smaller scales, such as the sub-county or ward level. This would enable a more focused approach, allowing us to discern the impact of air pollution on various land cover types more accurately. Furthermore, we recommend further research involving various sensors, including those on board Sentinel 2, to expand the scope and accuracy of air quality monitoring. This study advocates for the continued use of API as a standardized measure of air quality, providing a common metric for comparison and analysis.
Acknowledgments
The authors are grateful to Mr. Onyango Bernard, GIS Developer at the Food and Agriculture Organization, and Mr. Chiro Jelani from the National Environmental and Management Authority for their assistance.
Author Contribution
Quinto Juma Meltus: Conceptualization, methodology, software, data curation, writing, original draft preparation, visualization, investigation. Faith Njoki Karanja: Conceptualization and design of the study, supervision of the research, reviewing, writing and editing the manuscript, granting final approval of the version to be submitted.
Data Availability
We freely obtained the data from the Google Earth Engine catalog. Reach out to the corresponding author @ meltusquinto@gmail.com for data used in the study.
Funding Statement
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Exclusively the authors funded the research.
Annex
Table A1. Extracted Air Pollution Index (API) values for the various epochs over twenty sampled locations Nairobi County.
NOTES
1SWIR1: Short Wave Infrared Red 1 band (Band 6 for Landsat 8 and Band 5 for Landsat 7).
2“good” indicates that air quality is considered satisfactory, and air pollution poses little or no risk to the general population.
3“moderate” in the AQI signifies that air quality is acceptable; however, there may be some health concerns for a few sensitive individuals who are unusually sensitive to air pollution.
4“Unhealthy for sensitive groups” is an air pollution classification indicating heightened health risks for individuals with pre-existing respiratory or cardiovascular conditions, the elderly, young children, and those with compromised immune systems [59] .