NDVI-based Analysis of Green Space Decline and Air Quality in Dhaka: Implications for Sustainable Development Goals

The migration of individuals from rural to urban areas, commonly referred to as urbanization, has been a widespread occurrence on a global scale for decades. This process has resulted in a number of negative effects, including the loss of urban green space and an increase in air pollution [1-3], as well as a number of positive effects. Urban inhabitants can reap various advantages from green spaces, such as parks, gardens, and forests, encompassing better air quality, reduced heat island effects, and augmented mental health and well-being [4]. Trees serve as a carbon sequestration mechanism by intercepting particulate matter and microscopic pollutants, thereby enhancing the deleterious impacts of climate change. The expansion of cities often results in the relinquishment of green spaces to accommodate new constructions and infrastructure, consequently causing a decrease in the standard of living for inhabitants of urban areas. The global urban region experienced a significant increase of 168% from 2001 to 2018, with the most notable growth rates observed in Asia and Africa [5]. The inhalation of air pollution has been linked to various health issues, such as respiratory and cardiovascular ailments, lung cancer, and stroke [6, 7]. In 2019, a thorough evaluation of the worldwide burden of diseases revealed that air pollution, originating from indoor and outdoor sources, constituted the fourth most significant risk factor for mortality and disease globally. This factor was responsible for 6.67 million deaths. Air pollution accounted for 12% of fatalities in children under 5 [8]. Outdoor air pollution is responsible for causing approximately 4.2 million untimely fatalities globally each year, with the most significant impact observed in countries with low- and middle-income economies [1, 4].

Bangladesh has been categorized as a lower-middle-income nation since 2015 and is one of the nations with the highest population density globally, with a populace exceeding 160 million individuals [9, 10]. In recent years, the nation has undergone a swift urbanization process, whereby the proportion of individuals residing in urban areas has increased from 17% in 1974 to 36% in 2018 [11]. This has resulted in the expansion of cities and villages, resulting in a decrease in green spaces. The Bangladesh Bureau of Statistics conducted a study that revealed that the country's forest cover decreased from 17% in 1975 to 10% in 2010 [12]. The depletion of green areas has had a noteworthy effect on both the ecosystem and the well-being of the populace, with the issue of air pollution emerging as a significant apprehension in urban areas [13, 14].

The city of Dhaka, located in Bangladesh, is widely recognized as a highly urbanized and heavily polluted city characterized by elevated levels of environmental pollution and a notable scarcity of green spaces [5]. Over the past two decades, Dhaka has undergone a warming trend of approximately 3 degrees Celsius, which can be attributed to various factors such as the depletion of green spaces and swift urbanization. This is noteworthy as the global community aims to limit the increase in global temperature to below 1.5 degrees Celsius [15]. UNEP has mandated that a sustainable urban area should possess a minimum of 25% open space, whereas Dhaka only possesses 8% of such spaces [16]. Urban green infrastructure, comprising vegetation such as trees and shrubs, provides many benefits to urban environments. These benefits include the provision of shade, regulation of wind speeds, the absorption of hot air to mitigate elevated urban temperatures, the preservation of soils, the facilitation of wastewater treatment, and the reduction of air pollution and associated health risks through sewage water for irrigation purposes.

The major contributors to air pollution in Dhaka are brick kilns (40%), vehicle emissions (30%), construction (20%), industry and open burning (10%) [17]. In the vicinity of Dhaka City, brick kilns emit a total of 23,000 tons of PM_2.5 into the atmosphere on a yearly basis [18]. Exposure to the polluted air in Dhaka has resulted in the contraction of various respiratory illnesses such as infectious lung disease, pneumonia, acute respiratory infections, tuberculosis, and other related ailments. About two-thirds of Bangladesh's respiratory health problems and 10% of the country's total disease burden are attributable to air pollution [19].

This study aims to investigate the adverse effects of urbanization on green spaces and air quality in the context of global urbanization and potential measures to alleviate these issues, and ascertain the evolution of vegetation coverage in Dhaka City across different time periods. Besides, the vegetation coverage was quantified using Normalized Difference Vegetation Index (NDVI) analysis. Thus, the results of this analysis were subsequently examined in relation to the phenomena of rapid urbanization and degradation of Air Quality.

In this research, Landsat imagery from 1991 to 2021 with five year time intervals was used to conduct NDVI analysis. The data were prepared for further processing by performing specific preprocessing procedures. The following stage was extracting the area of interest from satellite images using the boundary shapefile of the study area. Afterward, an approach was made to conduct an NDVI analysis to classify the acquired data. The next step was to assess the accuracy of the expected data. Eventually, the area of the classified images was measured, and the findings were interpreted. The flowchart of our methodology is shown in Figure 1 below.

1.png

Figure 1. Methodological workflow of the study

3.1 Study area

2.png

Figure 2. Study area map

Dhaka, located in the middle of Bangladesh, is one of the most polluted cities in the world [24]. It is located at 23°46'37" north and 90°23'58" east. The overall land area of the city of Dhaka is 306 square kilometers [25], and its population is 12.6 million [26]. As Dhaka is the commercial hub of Bangladesh, every year, half a million [27], individuals from all around Bangladesh relocate to Dhaka. The city has undergone massive urbanization and rapid development in response to the rising need for excessive populations. These enormous populations and the rural nature of the city augmented the transportation demand. Consequently, numerous new automobiles are added to the road each year, causing daily traffic congestion and CO₂ emissions. The study area map has been shown in Figure 2.

3.2 Data source and collection

The NDVI analysis was used to quantify the quantity of healthy vegetation in the study area. This study required remote sensing satellite images. Several sources produce remotely sensed images of the earth's surface features. The Landsat data provided by the United States Geological Survey (USGS) was used for this research because it is publicly available and contains the oldest and most extensive collection of satellite images. All necessary data were retrieved from the USGS's official website.

3.2.1 Landsat data description

The Landsat imagery used in this study is listed with appropriate description in Table 2. Four types of sensor data were used, each with a spatial resolution of 30 meters. The Landsat TM sensor has 7 bands and captures visible and near-infrared light. The Landsat ETM+ sensor has 8 bands, including four visible and near-infrared (VNIR), two short wave infrared (SWIR), one thermal infrared (TIR), and one panchromatic (PAN) band. The Landsat OLI sensor has 11 bands, including visible, near-infrared, and shortwave infrared light. The Landsat TIRS sensor has 2 bands and captures thermal infrared radiation [28, 29].

Worldwide Reference System - 2 (WRS -2) [30] is used by Landsat 4, 5, 7, 8, and 9. All of our data was taken under cloud-free weather. All of the satellite data was recorded during the daytime. Except for the thermal and panchromatic bands, all the bands of Landsat data have spatial resolutions of 30 meters. The study area shapefile was produced by georeferencing.

Table 2. Description of USGS Landsat data used for conducting NDVI analysis

3.3 Data processing

3.3.1 Image Pre-process

3.png

Figure 3. Resolving errors in data

After retrieving the data, an approach was made to verify data integrity. Certain scanline errors have been found for some data in this instance (Figure 3). These records contain L7 ETM+ (path 137, rows 44) of 2006 and 2011 (Table 2). Scanline errors in collected data can be resolved with the Fill nodaata (Raster>Analysis>Fill nodata) tool in QGIS, an open-source GIS application.

Every Landsat image has been projected to the Universal Transverse Mercator (UTM) Zone 46 North Geographic Coordinate System and georeferenced to the WGS 1984 datum.

3.3.2 NDVI Analysis of the study area

The raster data were imported into ESRI ArcGIS after preprocessing the satellite images (version 10.8.1). The software provides imagery tools and methods for visualization and analysis, as well as the world's largest imagery database, making it ideal for mapping and analyzing satellite imagery.

Red Band and Near Infrared Band (NIR) were employed for NDVI analysis. NIR and Red bands on various Landsats have different wavelengths mentioned in Table 1.

NDVI was calculated using the Float command in Raster Calculator for precision.

$N D V I=\frac{\text { Float }(N I R-R)}{\text { Float }(N I R+R)}$       (2)

Several classification methods are used to categorize NDVI values, including thresholding, maximum likelihood, random forest, KNN with DTW, and support vector machine. This study used the maximum likelihood classification method (Table 3). This method classifies pixels into different vegetation types: forest, grassland, and shrubland. The classifier uses the known distribution of NDVI values for different vegetation types to classify pixels [31].

Table 3. Suitable NDVI ranges utilized for the land cover classification in this study

3.4 Accuracy assessment

From 1991 to 2021, seven different years of data were considered at five-year intervals for NDVI analysis. For a more precise assessment of the accuracy of the data, it is typically preferable to survey a sample of randomly selected points by visiting the study area and analyzing each point. However, it can occasionally be costly, time-consuming, or even impossible to visit and inspect the randomly chosen points physically. To authenticate our data, we utilized high-resolution images of the real world provided by Google Earth Pro as ground-truthing reference data.

Ground-truthing was conducted in December for each year's Google Earth Pro imagery aligned with the respective Landsat images. However, ground-truthing was not possible for 1991 and 1996 due to the lack of clear reference images.

4.png

Figure 4. Ground truthing points of the study area

Accuracy improves with larger sample points. For accuracy assessment, a simple random sampling method was used to choose 30 random data points, where new points were selected for each year (Figure 4).

Parameters such as overall accuracy, user accuracy, producer accuracy, and kappa coefficient were determined for each year's data. The following formulae were used to calculate these parameters:

$\begin{gathered}\text { Overall Accuracy }= \\ \frac{\text { Total number of correctly classified pixels (diagonal) }}{\text { Total number of reference pixels }} \times  100\end{gathered}$      (3)

$\begin{gathered}\text { User's Accuracy }= \\\frac{\text { Number of correctly classified pixels in each category }}{\text { total number of classified pixels in that category (the row total) }} \times 100\end{gathered}$      (4)

$\begin{gathered}\text { User's Accuracy }= \\\frac{\text { Number of correctly classified pixels in each category }}{\text { total number of classified pixels in that category (the column total) }} \times 100\end{gathered}$     （5）

The Kappa coefficient [32] is a statistical method for assessing the accuracy of classifications. The bigger the kappa value, the stronger the connection between theoretical and actual data (Table 4). There may be a variety of kappa coefficient calculation formulae; nevertheless, the following equation was used to calculate the kappa coefficient:

$\begin{gathered}\text { Kappa Coefficient }= \\ \frac{(T S \times T C S)-\Sigma(\text { Column total } \times \text { Row total })}{T S^2-\Sigma(\text { Column Total }- \text { row total })} \times 100\end{gathered}$     (6)

Here,

TS = Total number of samples taken

TCS = Total number of correct samples

Table 4. Rating criteria for Kappa statistics [33]

The study utilized NDVI analysis as the preferred method for vegetation assessment due to its specific design for quantifying vegetation quantity and health. This choice was made because the study aimed to investigate the dynamics of green cover and its correlation with air quality. This section will present the findings of the study, including the obtained outcomes, their accuracy, and the subsequent interpretation.

5.1 Green space scenario of Dhaka City from 1991-2021

Dhaka has undergone a massive and rapid land cover change, especially in vegetation and built-up areas from 1991 to 2021. To understand this change, the area of each land cover for the studied year has been computed using NDVI analysis data. Table 5 reveals that Dhaka City had around 167 square kilometers of green space in 1991 but only 153 square kilometers in 1996, a roughly 14 square kilometers loss. In the next studied year, 2001, the City witnessed a significant loss of green space and retained merely 107 sq. km of green space. In contrast, it expanded significantly in 2006, adding green space totaling 139 square kilometers. In the subsequent year of study, 2011, the expansion was negligible, adding just 1 square kilometer. However, after 2011 data shows a gradual extenuation of green space where the city had about 111 sq. km and 90 sq. km in respective years in 2016 & 2021. Green space and urbanization dynamics over time can be observed in Table 5.

Additionally, the thematic land cover changes in the study area that occurred within the study period (from 1991 to 2021) are illustrated in Table 6.

The land cover changes from 1991 to 2021 for each year was illustrated through NDVI analysis in Figure 5.

Table 5. NDVI analysis of Dhaka City from 1991 - 2021 using average value range

Table 6. Resultant thematic change statistics “from-to” matrices (1991-2021)

5.png

Figure 5. Composite image of NDVI analysis (1991 - 2021)

Table 7. Overall calculation of accuracy and Kappa coefficient

[Note: UA = User’s Accuracy; PA = Producer’s Accuracy; OA = Overall Accuracy]

5.2 Data accuracy analysis and interpretation

For data accuracy analysis, this study utilized the Error Matrix method, a statistical method for data accuracy assessment that uses users, producers, overall accuracy, and kappa coefficient [43].

User accuracy assesses the proportion of correctly classified instances within a predicted class, indicating errors of commission. Producer accuracy assesses the proportion of correctly predicted instances within the actual class, capturing errors of omission. The overall accuracy in the error matrix method is a metric that evaluates the correctness of instance classification across all classes. It quantifies the percentage of accurately classified instances from the total instances considered within the error matrix. The Kappa coefficient in the error matrix method measures the agreement between observed and predicted classifications, accounting for chance agreement [33].

The accuracy evaluation for 2001, 2006, 2011, 2016, and 2021 was performed using a high-resolution Google Earth Pro image. We were unable to determine accuracy for 1991 and 1996 because of a lack of quality imagery from Google Earth Pro for our study area prior to 2001.

In 2001, the average user and producer accuracy for classified images was 86.15% and 88.75%, respectively. In 2006, the average user and producer accuracy was around 89.75% and 92.5%, respectively. The average user accuracy in 2011 was 85.55%, and the producer accuracy was 90.62%. In contrast, the average user and producer accuracy in 2016 was 91.87% and 93.22%, respectively. Furthermore, in 2021, the average user accuracy was 88.43%, while the average producer accuracy was 86.6%.

The overall accuracy for 2001, 2006, 2011, 2016, and 2021 was 86.67%, 90.0%, 86.67%, 93.33%, and 90.0%, respectively. Kappa values were 0.82, 0.86, 0.82, 0.90, and 0.85 for 2001, 2006, 2011, 2016, and 2021, respectively. All data relating to accuracy assessment are presented in Table 7.

This research has comprehensively analyzed the temporal change of vegetation and its impact on air quality in Dhaka City. The findings clearly indicate that the city has experienced a significant loss of green space over the past three decades due to population growth, urbanization, and increased housing demand. As a result, Dhaka's air quality has deteriorated considerably, with industrial emissions, vehicular traffic, and construction activities contributing to increased pollution levels. The decline in vegetation has further exacerbated this issue, as fewer trees and plants are available to absorb air pollutants and release oxygen. The ecological balance has been disrupted, leading to the city's displacement and endangerment of various species.

The long-term effects of vegetation loss, rapid urbanization, and air quality degradation are extensive and multifaceted. The disruption of ecosystems and reduction in biodiversity have created ecological imbalances that require urgent attention. The intensification of industrial activities and vehicle emissions has increased health risks for urban residents, resulting in escalating healthcare costs and burdening healthcare systems. Additionally, the heat island effect and energy consumption for cooling purposes strain the power grid and contribute to rising electricity production and distribution costs. The challenges of water management, including reduced rainwater absorption and increased flooding risks, further compound the issues faced by Dhaka City. Moreover, social and economic disparities are exacerbated, impacting marginalized communities disproportionately.

To address these challenges and work towards a more sustainable future, it is imperative to implement comprehensive strategies. These strategies should encompass sustainable urban planning, preservation of green spaces, air pollution control measures, and improved water management practices. Transitioning to renewable energy sources can positively impact air quality, and the government should provide incentives and support for the adoption of solar, wind, and other renewable technologies. Waste management initiatives, including segregation, recycling, and composting, coupled with investments in waste treatment facilities, can contribute to a cleaner environment. Integrating green spaces, parks, and urban forests into city planning can help mitigate air pollution and enhance overall air quality. The preservation and creation of green areas should be prioritized, including innovative approaches such as rooftop gardens, vertical green walls, and urban parks.

Similar challenges faced by other cities, such as Lahore, Hotan, New Delhi, and Kathmandu, emphasize the need for a collaborative approach in addressing air pollution and green space decline. By striking a balance between urban development and the preservation of open spaces, cities can pave the way for a healthier and more sustainable urban environment.

The findings of this study present a clear call to action for stakeholders, policymakers, and the community at large to prioritize sustainable development goals and work towards a greener, cleaner, and more resilient Dhaka City. By implementing the recommended approaches outlined in this paper, we can contribute to a better quality of life for present and future generations, fostering a city that thrives both economically and environmentally.