Can AI sift through smog to show Pakistan a cleaner way forward?

The rapid intrusion of artificial intelligence (AI) is quietly materialising in Pakistan, from classrooms and banking to hospitals, traffic systems, and agriculture. Yet, despite these technological advancements, every year during the winter season, thick layers of smog cover the major cities of Pakistan behind a grey wall, for which now AI is being called on to tackle a far more visible and urgent challenge: the air we breathe.

AI is now being utilised to monitor air quality and predict smog events. Although this looks like a great enhancement in introducing a new development that could help authorities to prevent such issues, questions still arise whether AI can realistically overcome the systematic gaps and help Pakistan to breathe clean air?

Pakistan's air pollution crisis

Pakistan’s severe air pollution has been a persistent issue, particularly during the winter, when even the slightest change in the temperature exacerbates this problem. In major cities like Lahore, the Air Quality Index (AQI) repeatedly reaches 'hazardous' levels, posing serious risks to public health.

Lahore, according to the World AQI Index, is yet again the world’s most polluted city. Each winter, the smog not only shrouds the skyline but also threatens public health. In response, cities facing bad air quality are strictly prompted to implement emergency measures such as school closures, partial lockdowns, and market restrictions.

Recently, the Government of Punjab also announced winter vacations for all educational institutes across the province in a bid to take early measures, as temperatures are expected to drop sharply in December.

Elaborating on the air pollution’s impact, Linked Things CEO Sophia Hasnain observes: “Each year from October onwards, the air quality becomes bad. Every person breathing under that air says it's fine for them, even though we have collected data from 37 cities, which tells us that the air quality has become worse. People have gotten so used to it. To them, during the smog season, visibility is fine, it’s just cold weather.”

Major sources of pollution

Pakistan’s most persistent contributors to air pollution include:

• Vehicular emissions.
• Industrial activity.
• Brick kilns.
• Crop residue burning.
• Construction dust.
• Open waste burning.

To detect these sources, the government has started deploying AI-enabled air quality monitoring and forecasting systems across major cities in Pakistan to address worsening air pollution.

Earlier, in October, during the launch of Punjab’s new AI-powered air quality monitoring framework, Senior Minister Marriyum Aurangzeb, terming the AI-based AQI monitoring systems a comprehensive technological breakthrough, said: “These systems are designed to predict accurate air pollution levels, smog intensity, and even the cross-border movement of polluted air drifting in from India.”

She also highlighted that AI technology in these monitors will help the authorities to forecast air quality up to four months in advance.

Whereas, at the institutional level, the Punjab Environmental Protection Agency (EPA) approved the province’s AI-driven AQI framework, terming it a major scientific milestone.

EPA Director General Dr Imran Hamid Sheikh said the model relies on the primary air-quality data is incorporated into international forecasting methodologies.

This is the first such attempt by the country to integrate ground-level sensors with predictive AI models for actionable insights that could help officials plan interventions, issue early warnings, and prioritise which areas need urgent attention.

Talking about the technical side of these AI monitoring systems, Hasnain explains: “Hubs are collecting the data from the different areas, which give you granular data. At the mohalla (neighbourhood) level, if you’ve installed one monitor, then you can easily see the air quality performance of that area.”

She further elaborated that the system works by collecting real-time data from multiple sensors across different areas covered by the monitors; these sensors measure real-time pollutants like Particulate Matter (PM) 2.5, PM10, NO₂, and CO.

From these systems, the data is transmitted to a central system where machine-learning algorithms process it to detect patterns, identify hotspots, and even forecast peaks in pollution, offering actionable insights to authorities and the public.

According to the experts, a major hurdle in combating air pollution is the lack of widespread, reliable monitoring infrastructure.

“Right now in Pakistan, there are probably not more than 300 monitors, but they should be in thousands. Each city should have thousands of them,” added Yasir Darya, Director Climate Action Centre (CAC).

Pointing out the usage of AI, Darya highlights that AI integration could do better in giving alerts where the air quality is expected to be bad so that important measures can take place timely, but if AI is to be utilised in data collection then it must be noted that Pakistan does not produce a lot of data yet, “When you ask an AI about air quality in Pakistan, you get rough figures. The AI itself needs more learning to become aware of the data that is actually available,” he added.

Hasnain also highlighted the limitations, saying, “The systems they have installed are extremely limited in the entire city. Even after the installation of this system can derive some sort of result, but the real issue is that this data collection must be done over a very wide area, and many people should be working on it.”

How AI can make a difference

Understanding on a deeper level that if AI has the potential to reshape the way Pakistan responds to air pollution, the experts gave their insights briefly. At their core, AI can convert a jumble of environmental signals into clear, actionable plans. Machine-learning models analysing historical trends, weather patterns, and satellite indicators can forecast smog in advance.ga

A 2025 study published in the Spectrum of Engineering Sciences predicted Karachi's pollution level with models like Long Short-Term Memory (LSTM) and Random Forest using satellite data and showed how the early warnings can become scientifically reliable tools for intervention.

This is in line with what has already been put into motion in Punjab, where, under the Climate Resilient Punjab Vision and Action Plan, AI-enabled AQI monitors have been used along with satellite-based mapping to identify stubble-burning hotspots, industrial emissions, and spikes in traffic-related pollution.

The scarcity of data limits the degree of accuracy and reliability of insights generated by AI. Pakistan Air Quality Initiative (PAQI) Founder Abid Omar echoes this concern, warning that "AI-driven insights can only be as good as the data and information given to AI to develop those insights."

Without long-term, high-quality, publicly accessible air quality records, AI models cannot generate forecasts that policymakers can confidently act upon.

While these are limiting factors, AI can still make significant contributions in areas where traditional monitoring is lacking. Global research reveals that advanced neural networks create AQI maps even over regions with very few sensors. Such a study, conducted in 2025 on arXiv, used Graph Neural Networks (GNNs) to map air quality across Lahore from sparse sensors.

Already, AI-assisted systems on the ground in Punjab's "smog war room" alert teams if active fires or factory emissions are detected through satellite or drone data, accelerating responses previously delayed by manual reporting.

But as all experts agree, AI on its own cannot clean the air.

It can highlight hotspots, point to causes, and predict what will happen next-but enforcement, political will, and structural reforms must follow. If they don't, the predictions remain just that-predictions.

Limitations and risks

While AI offers significant potential, its effectiveness depends on reliable, comprehensive data. Hasnain explains: “If the authorities are collecting the data properly, then yes, AQI monitor data can be reliable. It depends on whether these monitors are certified and installed at the correct locations where accurate data can be gathered. There are guidelines from the Environmental Protection Agency (EPA) for where monitors should be installed. You can’t just install one in any place according to your need.”

Highlighting the risks and limitations from an expert’s perspective, Omar adds: “The primary limitations aren't just technical; they are structural, data-related, and human."

Data gaps and quality: AI is only as good as the data fed into it. Pakistan lacks long-term, reliable and publicly accessible air quality data.

Insufficient monitors: There is an insufficient number of official, high-quality air quality monitoring stations. While Punjab has been expanding its network (from around 30 to 75+, aiming for 100), many are not properly maintained, leading to missing or unreliable data.

Data silos: Government data is often fragmented, non-digitised, and locked in departmental silos, making it difficult to integrate for a holistic AI model.

Structural and economic realities: AI can identify a pollution source, but it can't solve the underlying economic reason for it. For example, AI can pinpoint stubble burning, but it doesn't make the alternative (like a 'Happy Seeder' machine) affordable for a small-scale farmer.”

AI has the potential to transform Pakistan’s air quality management, but only if foundational issues are addressed, as Omar also emphasises: “The use of technology for air quality management is a great initiative undertaken by the government, which shows its foresight in dealing with this complex problem. However, to train and really utilise and maximise the effective use of AI requires a solid foundation of long-term, reliable, high-quality data, specifically from a nationwide network of in-situ monitors.”

He added that with a limited regulatory monitoring network in Punjab since last year, the temporal scale is not sufficient for reliable air quality forecasting. AI-driven insights could only help develop the insights.

Considering each expert’s cross perspective, the officials from EPA and the Environment Protection and Climate Change Department (EPCCD) were contacted to gain an in-depth understanding of how these AI-powered AQI monitoring systems are being utilised to drive solutions.

However, repeated attempts to secure their input did not yield any response, which leaves a gap in the official take on the aspects of operational and enforcement of these systems.

In Pakistan, AI is known to be a transformative tool in many aspects, but when it comes to fighting one of the country’s biggest issues, air pollution, some experts warn that it cannot work in isolation unless proper policy enforcement, public awareness, and structural monitoring are conducted effectively.

Pareesa Afreen is a staffer at Geo.tv

Header and thumbnail illustration by Geo.tv