Why cities choke each winter: AI traces smog to transport emissions

Across rapidly growing cities in South Asia and beyond, smog has overwhelmed air quality controls and straining public health systems. Despite years of regulatory efforts, pollution alerts, and emergency responses, many metropolitan regions continue to experience prolonged episodes of hazardous air, raising questions about whether existing mitigation strategies are targeting the real drivers of the problem. New research using artificial intelligence (AI) suggests that the answer lies not in isolated pollution sources or weather events, but in the interaction between emissions and atmospheric conditions that amplify one another in predictable ways.

The study, AI-Driven Analysis of Meteorological and Emission Characteristics Influencing Urban Smog: A Foundational Insight into Air Quality, published in the journal Gases, applies machine learning to disentangle how human activity and weather jointly shape urban air quality. Using the city of Lahore, Pakistan, as a detailed case study, the research offers broader insights into why smog remains so persistent in densely populated cities with high transport demand and seasonal atmospheric stagnation.

Machine learning exposes how emissions and weather reinforce smog formation

Traditional air pollution studies have often treated emissions and meteorology as separate issues, focusing either on what pollutants are released or how weather affects their dispersion. The new research challenges that separation by modeling both elements together as parts of a single system. The authors argue that this integrated approach is essential for understanding modern urban smog, which rarely results from a single source or event.

The study applies a two-phase machine learning framework to examine how air quality deteriorates during peak pollution periods. In the first phase, the researchers combine daily air quality index data with meteorological variables such as temperature and wind speed, alongside aggregated emissions from vehicles and industry. Two widely used artificial intelligence models, Random Forest and Extreme Gradient Boosting, are trained to predict changes in air quality under real-world conditions.

Both models achieve very high predictive accuracy, explaining the vast majority of air quality variation during the study period. The results show that air pollution levels rise sharply when high emissions coincide with calm winds and stable atmospheric conditions, particularly during cooler months. Under these conditions, pollutants accumulate near the ground instead of dispersing, leading to rapid deterioration in air quality.

Vehicle emissions consistently emerge as the strongest predictor of poor air quality, outweighing industrial emissions and individual meteorological factors. At the same time, weather variables play a crucial role by determining whether emissions disperse or remain trapped over the city. Low wind speeds, temperature inversions, and high humidity act as amplifiers, transforming routine emissions into severe smog episodes.

The findings demonstrate that smog is not simply a matter of how much pollution a city produces, but also of when and how that pollution is released into the atmosphere. By capturing nonlinear relationships between variables, the machine learning models provide a more realistic explanation of smog dynamics than conventional statistical methods, which often underestimate the combined effects of emissions and weather.

Transport emissions dominate urban pollution in the Lahore case study

In the second phase, the study examines the chemical makeup and sectoral origins of emissions to identify which activities contribute most to smog formation. Using detailed emission inventories, the researchers focus on four major pollutants: nitrogen oxides, carbon monoxide, volatile organic compounds, and particulate matter. These pollutants are assessed across three main sectors: transportation, industry, and agriculture.

The results reveal a striking concentration of emissions in the transport sector. In the Lahore case study, transportation accounts for roughly 90 percent of total annual emissions, far exceeding contributions from industrial and agricultural activities. This imbalance highlights the central role of urban mobility patterns in shaping air quality outcomes.

Within the transport sector, emissions are driven largely by incomplete fuel combustion and aging vehicle fleets. Carbon monoxide and particulate matter dominate transport-related pollution, reflecting the prevalence of motorbikes, small personal vehicles, and auto-rickshaws that lack modern emission controls. Motorbikes emerge as the single largest source of emissions, followed by cars and three-wheeled vehicles, while buses and trucks contribute a comparatively smaller share.

Industrial emissions present a different profile. Although their overall contribution to total pollution is limited, they release higher proportions of nitrogen oxides, linked to fuel-intensive processes such as boilers, generators, and manufacturing operations. Agricultural emissions, while smaller in volume, are associated with elevated levels of particulate matter and volatile organic compounds during residue burning and fertilizer use, particularly in seasonal cycles.

The ML model used in this phase demonstrates strong predictive performance across all sectors and pollutants, confirming that emission behavior is both sector-specific and chemically distinct. This level of detail allows policymakers to move beyond generalized pollution controls and design interventions that target the most harmful activities with greater precision.

Health risks, policy lessons, and global relevance beyond one city

Many urban centers across South Asia, the Middle East, and parts of East Asia share key characteristics with Lahore, including rapid population growth, heavy reliance on private transport, aging vehicle fleets, and winter weather patterns that favor pollutant accumulation.

Prolonged exposure to fine particulate matter and nitrogen oxides increases the risk of respiratory infections, cardiovascular disease, reduced lung function, and premature mortality. Seasonal smog episodes place disproportionate burdens on children, older adults, and people working outdoors, while also increasing pressure on healthcare systems and reducing economic productivity.

The study reinforces the need for policies that address urban mobility at its core. Measures such as stricter vehicle emission standards, improved inspection and maintenance programs, promotion of cleaner fuels, and investment in efficient public transport systems are likely to yield the greatest benefits. Traffic management strategies that reduce congestion and limit high-emission activity during unfavorable weather conditions could further reduce pollution spikes.

The research also sheds light on the role of AI in air quality governance. Accurate short-term forecasting models can support early warning systems, allowing authorities to issue timely advisories and implement temporary controls during high-risk periods. More importantly, AI-based attribution tools enable policymakers to test hypothetical scenarios, such as emission reductions in specific sectors, before committing to large-scale interventions.

The authors acknowledge important limitations too. The analysis focuses on a narrow seasonal window, limiting its ability to capture year-round pollution dynamics. Citywide averages may also obscure localized hotspots influenced by traffic density or industrial clustering. Expanding the framework to include longer time series, higher spatial resolution, and additional cities would strengthen its general applicability.