Covergent Thinking - Global Air Pollution

[salihakalender]

Global Air Pollution

Problem Statement:

Air pollution remains one of the most pressing environmental and health challenges worldwide, contributing to millions of premature deaths each year. While the Air Quality Index (AQI) gives a general measure of pollution levels, it doesn’t always explain what’s driving the problem in each city. To take effective action, it’s important to move beyond overall AQI values and analyze which pollutants—such as PM2.5, NO₂, CO, or O₃—are the main contributors to poor air quality in different locations. This project uses 2024 global air pollution data to explore how individual pollutants impact AQI, classify cities by their dominant pollution sources, and uncover regional pollution patterns. By doing so, we aim to support more targeted and informed air quality management strategies.

Background Review

The background research explores:

• Health and economic impacts of air pollution (WHO, World Bank, GBD 2024)
• Key pollutants (PM2.5, NO2, CO, O3) and their sources
• Urbanization, low-income exposure, and regional variation
• Prior data science efforts in predicting AQI
• Tools like OpenAQ, CAMS Reanalysis, Kaggle datasets

Summary of Problem Domain (With Systems Thinking)

Air pollution is a complex system involving:
Inputs: Vehicle emissions, industrial output, energy generation
Processes: Chemical transformations, meteorological effects
Outputs: Health outcomes, environmental degradation, economic loss
Stakeholders: Governments, healthcare systems, citizens, data providers

Systems thinking helped us understand:

• Interdependence of pollutant types
• Regional and seasonal loops (e.g., winter smog vs. summer ozone)
• Feedback loops between regulation and health

Research Questions

"Can we quantify and categorize the dominant pollutant driving AQI in each global city, and visualize regional trends in pollutant profiles using clustering and ratio-based analysis?"

Novel Contribution

In this project, we will begin with a comparative analysis of AQI levels across cities and regions, followed by a detailed examination of pollution profiles to identify dominant pollutants in each location. As a novel approach, rather than simply reporting overall AQI, we go a step further to reveal why cities have poor air quality by calculating the contribution ratios of individual pollutants (PM2.5, NO₂, CO, and O₃). These insights allow us to classify cities based on their dominant pollutant and can support more targeted pollution control strategies—for example, implementing PM2.5-focused policies in particulate-heavy cities or ozone reduction measures where O₃ is the primary concern.

[linahKhayri]

Why This Matters

Air pollution is one of the most underestimated killers in the world today. Unlike more visible threats, polluted air is invisible and easy to ignore until it becomes personal.
My mother has asthma. I’ve seen how a single visit to a polluted area can leave her struggling to breathe for weeks, but still, this invisibility of polluted air is what makes it so dangerous and easy to be overlooked.
I think that by telling a story or a perspective that humanizes the data, we can connect readers to the real stakes behind the statistics -this should be done in the repo readme-

My questions were designed from these two thoughts:

Are children losing years of life before they even get the chance to grow up just because of the air they breathe?

Is the air in our cities more dangerous than the crime on our streets?

1. How many years of life are lost (YLL) annually due to air pollution exposure in children under 10 across high-pollution cities globally?

FRESCO Evaluation

• Feasible: Leverages datasets like WHO, Global Burden of Disease (GBD), and UN’s RCP scenarios that include age-specific Years of Life Lost (YLL) and pollution exposure data.
• Relevant: Highlights the disproportionate impact on vulnerable populations (young children) .
• Ethical: Uses public, aggregate-level data with no privacy concerns.
• Specific: Focused on YLLs for children under 10 in high-pollution urban areas.
• Complex: Requires integration of pollution data, population distributions, and health outcomes.
• Original: Most studies focus on national aggregates; this city-level, child-focused analysis provides a novel and human-centered lens.

2. Is chronic air pollution exposure more statistically linked to premature death than violent crime in major urban centers?

FRESCO Evaluation

• Feasible: Draws from public datasets on urban air pollution mortality and violent crime death rates.
• Relevant: Offers a striking comparison between two major but differently perceived urban threats.
• Ethical: Works with aggregated, city-level data respecting privacy and consent.
• Specific: Focuses on premature mortality linked to both pollution and crime.
• Complex: Involves data normalization (per capita), age-adjustment, and statistical inference.
• Original: Provides a fresh comparative framing that can shift public dialogue and inform policy across domains.

⚠️ Note: These questions can be refined or reframed for clarity, focus, or stronger analytical power.

[salihakalender]

Hello Linah,

The research questions you explored are really great. You’ve clearly put a lot of thought into them — well done!

We could also consider adding the health effects aspect to our global air pollution research. I’ll look into how we might combine these topics and whether the available data would support that.

[linahKhayri]

Hi Saliha.

Thank you for your kind words! I appreciate your support.

[Adamx090]

Great work Linah! This is a thoughtful take . I like the fact that you personalized the research as someone with an invested interest in the topic.

[FalaqMajeed]

@salihakalender I really like your idea and how it looks at the bigger picture. Most people focus on which countries have high AQI, but not on why or which gas is causing it. Your question is clear, useful, and to the point.

[Adamx090]

Carbon Dioxide Role

Introduction

Carbon dioxide (CO₂) is not classified as a traditional air pollutant under most air quality standards, as it is a naturally occurring, non-toxic gas at ambient atmospheric levels and plays an essential role in Earth's carbon cycle. However, despite its relative safety to breathe, CO₂ is a primary greenhouse gas and a major driver of global warming. Importantly, there is a notable relationship between CO₂ emissions and the presence of harmful air quality pollutants such as particulate matter (PM2.5), nitrogen oxides (NOₓ), and volatile organic compounds (VOCs), which has two key dimensions:

• Source-level correlation: The industrial and combustion processes that release CO₂, such as fossil fuel burning for transportation, electricity generation, and manufacturing, simultaneously emit harmful air pollutants.

• Climate-driven correlation: Rising global temperatures caused by elevated CO₂ levels can indirectly worsen air quality by intensifying the formation of ozone, increasing the frequency and severity of wildfires, and altering atmospheric conditions in ways that trap or distribute pollution.

These intertwined dynamics suggest that monitoring and modeling CO₂ may offer insights into the behavior of air quality indicators, raising an important question.

Research Question

To what extent can carbon dioxide (CO₂) concentrations be used to predict PM2.5 levels across different cities and seasons?"

FRESCO Evaluation

• F – Feasible: Public datasets for both CO₂ and PM2.5 are widely available from global monitoring networks and satellite sources.
• R – Relevant: PM2.5 is a major public health concern, and CO₂ is central to climate change, making their relationship crucial to understand.
• E – Ethical: The research uses only publicly available, non-personal environmental data, posing no ethical risks.
• S – Specific: The question clearly defines both the independent variable (CO₂) and the dependent variable (PM2.5), with scope across cities and seasons.
• C – Complex: The relationship involves multiple influencing factors, allowing for advanced modeling techniques like regression or time series analysis.
• O – Original: While CO₂ and PM2.5 are well-studied individually, using CO₂ to predict PM2.5 is a novel application with practical implications.

Expanded Version of the Question

How well do carbon dioxide (CO₂) levels explain variations in both PM2.5 concentrations and overall Air Quality Index (AQI) ratings, and how do these relationships differ across cities and climate zones?

Practical Significance of the Research

This research is especially useful because CO₂ is much easier to measure than pollutants like PM2.5 or composites like AQI, which require multiple, often expensive sensors. CO₂ sensors are low-cost, stable, and widely deployed, especially since CO₂ is already monitored globally for its greenhouse gas role in climate change. As a result, the data is not only easier to collect but is often already available, meaning that if CO₂ proves to be a reliable proxy for air quality, this approach could significantly reduce the cost and complexity of air pollution monitoring, especially in low- and middle-income countries or rural areas.

[Adamx090]

@salihakalender Great work! This is starting to look like an interesting topic already!

[ObayCipher]

🌍 Global Air Pollution: A Data-Driven Crisis

Introduction
Air pollution is responsible for an estimated 7 million premature deaths annually and continues to pose one of the greatest environmental risks to human health, especially in urban centers. According to the World Health Organization (WHO), over 90% of the global population breathes air that exceeds safe pollution limits, with disproportionate effects on children, low-income populations, and those in the Global South WHO, 2024. Yet many countries still lack the tools to diagnose which pollutants matter most and how to act effectively.

Air Quality Index (AQI) is commonly used to summarize pollution, but it fails to indicate which pollutants (PM2.5, NO₂, CO, O₃, Hg) are the dominant contributors in each city. Policymakers need more precise diagnostics—not just a single AQI number—to develop targeted mitigation strategies.

Recent research from platforms like OpenAQ, NASA’s CAMS reanalysis, and the Global Burden of Disease (GBD 2024) provides publicly accessible, high-resolution air pollution data. Combined with tools from data science and machine learning, we can now move beyond generalizations to classify cities by pollution type, identify health impacts on vulnerable groups, and uncover climate-pollution interdependencies (e.g., CO₂ and ozone formation).

New pollutants like mercury (Hg)—a toxic heavy metal—are gaining attention for their neurodevelopmental harm and links to fossil fuel combustion, especially in coal-heavy regions UNEP Mercury Assessment, 2023.

🔍 Research Questions for Discussion

🧪 Pollutant Profiling and Classification

• Which pollutants (PM2.5, NO₂, CO, O₃, Hg) are the dominant drivers of AQI in the most polluted cities globally in 2024?

• Can global cities be grouped into pollution types (e.g., “PM2.5-dominant”, “O₃-driven”) using clustering algorithms?

• What are the seasonal and regional trends in dominant pollutant types across continents and climate zones?

🤖 Predictive Modeling and CO₂ Relationships

• To what extent can carbon dioxide (CO₂) levels predict PM2.5 or AQI values across cities and seasons?

• How well do machine learning models forecast AQI when trained on pollutant concentrations, weather, and CO₂ data?

🧍 Public Health and Human-Centered Analysis

• How many years of life are lost (YLL) due to air pollution exposure in children under 10 in high-pollution cities?

• Is the mortality impact of air pollution statistically greater than violent crime in major urban centers?

🌐 Spatial and Environmental Justice Insights

• How do industrial hotspots, low-income zones, and mercury emissions overlap with poor AQI levels on the global map?

• Are there urban clusters where mercury (Hg) co-occurs with elevated PM2.5 or ozone, and what are the regulatory blind spots?

[salihakalender]

Hello all,

@Adamx090 You made a great point in today’s meeting: in almost every city, PM2.5 appears to be the most dominant pollutant. Given this, we can add the following enhancements to our study:

✅ 1. Pollutant Composition Index (PCI)

What is it?
A custom metric that captures how diverse or concentrated the pollution profile of a city is. It measures whether a city’s air pollution is dominated by a single pollutant or is spread across multiple pollutant types.

Example:

If PM2.5 contributes 90% of the AQI in City A, the PCI is low (high dominance, low diversity).

If all four pollutants (PM2.5, NO₂, CO, O₃) contribute equally, the PCI is high (balanced profile).

Why it matters:
Cities with high PCI values may require more comprehensive control strategies, while low-PCI cities can benefit from focused interventions targeting one pollutant.

✅ 2. Policy Simulation Tool

What is it?
A simple model that estimates how changes in a specific pollutant level affect the overall AQI. This helps simulate the potential impact of different environmental policies.

Example:
“If PM2.5 levels in Jakarta are reduced by 30%, the AQI drops from 160 to 110.”

Why it matters:
Most AQI dashboards (like WAQI) only show the current status. Your tool helps decision-makers simulate “what-if” scenarios and prioritize actions based on expected impact.

✅ 3. Regional Signature Profiling

What is it?
The process of identifying typical pollutant patterns for each world region (e.g., South Asia, Europe, Africa, Latin America).

Example:

South Asia → PM2.5-dominated, peaks in winter

Europe → Generally clean air, but O₃ spikes in summer

Sub-Saharan Africa → Increasing NO₂ near cities due to rapid urbanization

Why it matters:
Recognizing these patterns allows for region-specific air quality strategies instead of applying a one-size-fits-all model.

✅ 4. Classify PM2.5-Dominant Cities Internally

Every city may suffer from PM2.5 pollution, but the causes, sources, and timing can differ greatly.

Suggested Subgroups:

1. Source-Based Classification:

Traffic-heavy PM2.5 (e.g., Delhi, Jakarta)

Biomass-burning PM2.5 (e.g., Accra, cities near forests)

Industrial PM2.5 (e.g., Wuhan, suburban Istanbul)

2. Seasonal PM2.5 Patterns:

Cities with winter peaks (e.g., heating-related emissions)

Cities with stable, year-round PM2.5 levels

3. Co-Pollutant Signatures (Pollution Fingerprints):

PM2.5 + high NO₂ → Likely traffic-dominated

PM2.5 + high CO → Suggests biomass burning or incomplete combustion

➕ Example Visualization Idea:
Map of PM2.5-dominant cities, color-coded by secondary pollutant levels (e.g., NO₂, O₃) to highlight underlying causes.

[salihakalender]

🧠 Health Impacts Layer

While our main focus is decomposing AQI into its pollutant components, we expand our analysis by integrating known health impacts of each pollutant type based on public health research and global databases.

✅ 1. Pollutant-to-Health Mapping (Evidence-Based Insight)

We linked each pollutant to its primary health effects using WHO, GBD 2024, and EPA literature:

Pollutant	Main Health Risk	Most Affected Groups
PM2.5	Cardiovascular disease, lung cancer	Elderly, general population
NO₂	Childhood asthma, lung inflammation	Children, school zones
O₃	Respiratory irritation, reduced lung function	Outdoor workers, people with asthma
CO	Neurological damage, oxygen deprivation	Pregnant women, infants

This layer helps us interpret not just how polluted a city is, but what kind of health burden it might face based on pollutant profiles.

✅ 2. Integrating Global Health Burden Data

We enrich our city-level pollution analysis by merging it with country-level health burden data from the Global Burden of Disease (GBD) project. This allows us to compare pollution exposure with actual health outcomes.

For example:
"Many of the most PM2.5-heavy cities are located in countries with high air pollution-related mortality rates, such as India and Bangladesh (GBD 2024)."

✅ 3. Health-Aware Pollution Clusters

In addition to clustering cities by pollutant composition, we identify health-relevant city groups, such as:

“PM2.5-dominant high-risk zones”

“NO₂-driven childhood asthma clusters”

“O₃-heavy but low AQI regions”

This adds a strategic policy dimension to our results, going beyond visualization toward actionable insights.

[salihakalender]

Here is the updated research question. Please check and share your feedback.

Research Question:

“How can we identify and categorize the dominant pollutants driving AQI in global cities, and what regional patterns emerge through clustering and ratio-based analysis? How do these patterns relate to known health impacts from public health data?”

[linahKhayri]

Here is the updated research question. Please check and share your feedback.

Research Question:

“How can we identify and categorize the dominant pollutants driving AQI in global cities, and what regional patterns emerge through clustering and ratio-based analysis? How do these patterns relate to known health impacts from public health data?”

Here is the updated research question. Please check and share your feedback.

Research Question:

“How can we identify and categorize the dominant pollutants driving AQI in global cities, and what regional patterns emerge through clustering and ratio-based analysis? How do these patterns relate to known health impacts from public health data?”

Here is the updated research question. Please check and share your feedback.

Research Question:

“How can we identify and categorize the dominant pollutants driving AQI in global cities, and what regional patterns emerge through clustering and ratio-based analysis? How do these patterns relate to known health impacts from public health data?”

How about framing it this way, focusing on clustering based on dominant pollutants first and with some specific health concerns:

Can we cluster global cities based on the dominant pollutants contributing to AQI, and how do these clusters align with regional respiratory and cardiovascular illnesses

[FalaqMajeed]

I guess Linah your research question is more focused and you dropped some points on calculating the PCI and only focuses on respiratory and cardiovascular illnesses, or that's what you're trying to say

[salihakalender]

Thank you Linah and Falaq for your valuable contributions.

@linahKhayri This version looks great — it’s clearer and more focused.

Shall we go this version? It aligns better with academic language norms.

“Can we cluster global cities based on the dominant pollutants driving AQI, and examine how these clusters correspond to regional patterns in respiratory and cardiovascular health outcomes?”

[Adamx090]

Air Quality Index (AQI)

Hello everyone! I’ve done some research and it seems we slightly misunderstood how the Air Quality Index (AQI) works. AQI is actually assigned to each pollutant separately based on its concentration in the air. The final AQI reported for a city is not calculated from a mix or average of all pollutants, it’s simply the AQI value of the single highest pollutant at that time.

The reason for this is that AQI values fall into health-based brackets (e.g., Good, Moderate, Unhealthy, etc.) - see below table - and each bracket corresponds to a specific level of health concern. Once any one pollutant reaches a level that poses a health risk, that value sets the AQI for the entire city.

If the final AQI were calculated by combining multiple pollutants (some with lower AQI values), it would lower the risk level of the highest pollutant, which could give a false sense of safety.

For example, if PM2.5 is in the Unhealthy range but the other pollutants are only Moderate, averaging them could result in an overall AQI that falsely reports the air as only Moderate. That’s why the system always reports the worst-case pollutant , to better reflect the actual health risk at that moment.

So the AQI of the city is in fact the AQI of the dominant pollutant.

Air Quality Index (AQI) Brackets with Pollutant Concentration Ranges

AQI	Level	Color	PM2.5 (µg/m³)	O₃ (ppm)	NO₂ (ppb)	CO (ppm)	Health Implications
0–50	Good	🟢 Green	0.0 – 12.0	0.000 – 0.054	0 – 53	0.0 – 4.4	Air quality is satisfactory; little or no risk.
51–100	Moderate	🟡 Yellow	12.1 – 35.4	0.055 – 0.070	54 – 100	4.5 – 9.4	Acceptable; some pollutants may affect very sensitive individuals.
101–150	Unhealthy for Sensitive Groups	🟠 Orange	35.5 – 55.4	0.071 – 0.085	101 – 360	9.5 – 12.4	Sensitive groups may experience effects; general public not likely affected.
151–200	Unhealthy	🔴 Red	55.5 – 150.4	0.086 – 0.105	361 – 649	12.5 – 15.4	Everyone may begin to experience health effects; more serious for sensitive groups.
201–300	Very Unhealthy	🟣 Purple	150.5 – 250.4	0.106 – 0.200	650 – 1249	15.5 – 30.4	Health alert: serious health effects for entire population.
301–500	Hazardous	🟤 Maroon	250.5 – 500.4	0.201 – 0.604	1250 – 2049	30.5 – 50.4	Emergency conditions; everyone is more likely to be affected.

Research Question

Can we cluster global cities based on the dominant pollutants driving AQI, and examine how these clusters correspond to regional patterns in respiratory and cardiovascular health outcomes?

This research question looks good, but I have 2 points on this part:
"Can we cluster global cities based on the dominant pollutants driving AQI"

• We should avoid using the word "cluster". If you remember, clustering is only for unsupervised learning when you don't have a label and you want to uncover latent features. But here we have grouping based on the dominant pollutant, so it's more like categorizing.
• The second point is that the answer is simply: Yes, we can. We can easily group cities based on the dominant pollutant. All we have to do is filter the data for the highest pollutant, it can be easily done in excel. This is a data-wrangling step, not a discovery. So it should not be in the research question.

If the idea is to see how the dominant pollutant corresponds to particular heath complications, then I suggest to rephrase the question this way:

How do dominant air pollutants across global cities relate to regional patterns in respiratory and cardiovascular health outcomes?

What do you guys think? I hope to hear your thoughts on this take.

[salihakalender]

@Adamx090 Thanks for the clarification! I understand your concern regarding data wrangling. However, in our context, identifying and categorizing the dominant pollutant in each city is not merely a preparatory step — it serves as an analytical foundation for several core components of our research.

According to the U.S. Environmental Protection Agency (EPA) and the World Air Quality Index Project, the AQI for a city is determined by the pollutant with the highest sub-index at a given time, not by averaging across pollutants. This means that identifying the dominant pollutant is inherently tied to the way AQI is calculated and interpreted (EPA, 2024; WAQI, 2024).

Building on this, categorizing cities by dominant pollutants allows us to construct the Pollutant Composition Index (PCI) — a novel metric that captures the diversity or concentration of pollutant contributions. This approach also enables regional pollution profiling (e.g., PM2.5-dominant in South Asia vs. O₃-dominant in parts of Europe), which has been shown in the literature to be critical for region-specific policy interventions (GBD 2024; WHO, 2023).

Furthermore, different pollutants have distinct and well-documented health effects. For instance:

• PM2.5 is associated with cardiovascular disease and lung cancer (WHO, 2021)

• NO₂ is linked to childhood asthma (HEI, 2020)

• O₃ contributes to respiratory irritation (EPA, 2022)

By aligning dominant pollutant profiles with these known health risks, our research moves beyond simple data grouping — we’re uncovering statistical relationships between environmental exposure and health outcomes, a method supported by studies in urban epidemiology and environmental health sciences.

So while the act of identifying the highest pollutant may seem simple, the interpretive value it generates — when combined with health outcome data — forms the analytical core of our project. We can frame this pollutant analysis as a critical methodological step rather than just background processing.

[salihakalender]

@FalaqMajeed Great question, Falaq! Actually, we’re planning to do all three — starting with identifying regional pollution patterns, then finding the dominant pollutant in selected cities, and finally exploring how these are linked to respiratory and cardiovascular health outcomes.

[salihakalender]

@linahKhayri Absolutely, you’ve captured it well! I feel the same way — both angles are important and will strengthen our analysis by providing different but connected insights.

[Adamx090]

Thank you all for your input . I think we’re all on the same page in terms of approach. But I believe my position might have been slightly misunderstood. I have no issue with identifying and categorizing cities based on their dominant pollutant and then studying health effects based on those categories. In fact, that’s a valuable step in the analysis.
What I’m pointing out is that the research question itself should not ask whether we can do that, because the answer is already known. We can categorize cities by dominant pollutant quite easily; it’s a straightforward data processing step. The research question should focus on the insight or mystery we’re trying to uncover.
So if our approach involves categorizing cities by dominant pollutant and then comparing health outcomes across those categories, the research question should reflect that goal only, not the categorization step itself. For instance:

How do respiratory and cardiovascular health outcomes vary across global cities categorized by their dominant air pollutant?

This question emphasizes the role of categorization in the research, but it doesn't ask if we can do it. That's all the point I am trying to make.

[salihakalender]

I see your point. Your suggested phrasing captures that purpose clearly and keeps the focus where it matters most — on the health implications.

[FalaqMajeed]

Thanks, Saliha and Salih, for the clarification! I think we’re now getting really close to finalizing our research question.so, do we all agree about it?

[Husayn01]

Hi Everyone,

I've carefully read through everyone's contributions, and I'm really impressed with the depth of analysis and thoughtful approaches so far, Well done guys.

Building on these excellent points, and acknowledging the importance of targeted, actionable insights, I'd like to propose an additional, yet directly related, dimension to our project: the socioeconomic factors influencing air pollution and its health impacts.

I propose we integrate an analysis of how socioeconomic indicators (e.g., income levels, population density, industrialization rates, access to green spaces) correlate with dominant pollutant profiles and their associated health burdens. This will allow us to:

• Identify Vulnerable Populations: Pinpoint which communities, often low-income or marginalized, are most exposed to specific dominant pollutants and suffer the most severe health consequences.

• Uncover Systemic Inequalities: Reveal how historical and ongoing socioeconomic disparities contribute to unequal distribution of air pollution and health risks.

• Inform Equitable Policy: Provide data-driven insights for developing environmental policies that not only reduce pollution but also address issues of environmental justice and aim for equitable health outcomes.

Proposed Research Question(s) / Analytical Focus:

Building upon the refined research question: "How do dominant air pollutants across global cities relate to regional patterns in respiratory and cardiovascular health outcomes?"

we could add:

"How do socioeconomic factors and urban development patterns influence the prevalence of dominant pollutants and the severity of associated respiratory and cardiovascular health outcomes across global cities?"

[Ridwanayinde]

I am really impresed with everything I read here. Kudos to you guys.