Khalifa University Pioneers AI Forecasting for NO₂ and O₃ Levels in Abu Dhabi

The Groundbreaking Publication from Khalifa University Researchers

In a significant advancement for environmental science and public health, researchers from Khalifa University in Abu Dhabi have published a pioneering study in Nature Scientific Reports titled "Multi-model forecasting of NO₂ and O₃ in Abu Dhabi: benefits of correlation-based feature augmentation." Released on March 17, 2026, the paper demonstrates how artificial intelligence (AI), specifically advanced machine learning (ML) and deep learning models, can dramatically improve predictions for two critical air pollutants: nitrogen dioxide (NO₂) and ground-level ozone (O₃). Led by corresponding author Nazar Ali from the Department of Electrical Engineering, along with Waad Abuouelezz, Zeyar Aung, Ahmed Altunaiji, and Shaik Basheeruddin Shah, the team collaborated with Mansour Alkatheeri from the Environment Agency - Abu Dhabi (EAD). This work underscores Khalifa University's commitment to leveraging cutting-edge technology to address real-world challenges in the United Arab Emirates (UAE).

The study evaluates a suite of models across forecasting horizons of 1 hour, 2 hours, 1 day, and 1 week, using metrics like symmetric Mean Absolute Percentage Error (sMAPE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). Transformers emerged as the top performer, achieving sMAPE values of 0.2354–0.2981 for NO₂ and 0.2030–0.2719 for O₃, highlighting their potential for precise, actionable forecasts.

Air Pollution Challenges in Abu Dhabi: NO₂ and O₃ Explained

Abu Dhabi, the UAE's bustling capital, faces air quality issues stemming from rapid urbanization, industrial activities, vehicular emissions, and natural dust events. NO₂, a reddish-brown toxic gas primarily from combustion sources like vehicles and power plants, irritates the lungs and exacerbates respiratory conditions such as asthma. O₃, a secondary pollutant formed when NO₂ reacts with volatile organic compounds (VOCs) in sunlight, contributes to smog and can trigger inflammation in the airways, particularly affecting vulnerable groups like children and the elderly.

Recent data from EAD's network of over 20 monitoring stations shows Abu Dhabi's annual PM2.5 average around 38 μg/m³, often moderate but spiking during dust storms or high-traffic periods. NO₂ levels have declined post-2020 due to pandemic-related emission reductions, yet O₃ remains a concern in warmer months. In the UAE, air pollution is linked to approximately 4,000 preventable deaths annually, emphasizing the urgency for reliable forecasting to enable timely interventions.

Khalifa University's ENGEOS Lab: A Hub for Environmental Innovation

Housed within Khalifa University's College of Engineering and Physical Sciences, the Environmental and Geophysical Sciences (ENGEOS) Lab specializes in high-resolution weather, air quality, and dust storm forecasting. Equipped with state-of-the-art monitoring platforms, the lab analyzes data from satellites, ground stations, and models like WRF-Chem to support renewable energy and sustainability efforts. Past projects include studying NO₂ periodicity across the UAE and assessing lockdown impacts on particulate matter (PM), revealing complex dynamics where reduced emissions sometimes led to higher dust contributions.

This latest publication builds on KU's track record, integrating EAD data to push boundaries in AI-driven environmental prediction. As UAE universities like Khalifa prioritize research aligned with national visions such as UAE Net Zero by 2050, such initiatives position the institution as a leader in applied sciences.

Decoding the Multi-Model Machine Learning Framework

The researchers employed a comprehensive ensemble of models to benchmark performance:

• Traditional ML: Decision Trees (DT), Random Forests (RF), Support Vector Regression (SVR)
• Deep Learning: Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks
• Time-Series Specialists: Prophet
• Advanced Transformers: Attention-based models excelling in sequence dependencies

Training data from EAD stations captured hourly pollutant levels alongside meteorological variables like temperature, wind speed, and humidity. A novel correlation-based feature augmentation incorporated interactions between NO₂ and O₃, recognizing their chemical interdependence—NO₂ precursors fuel O₃ formation under sunlight. This step-by-step process—data preprocessing, feature engineering, model training, and hyperparameter tuning via cross-validation—ensured robust, station-specific predictions tailored to Abu Dhabi's diverse microclimates from urban cores to industrial zones.

Photo by Markus Winkler on Unsplash

Transformer Models Shine in Short- and Long-Term Forecasts

Transformers outperformed others across all horizons, leveraging self-attention mechanisms to capture long-range dependencies in time-series data. For instance, CNNs ranked second for 1- and 2-hour forecasts, ideal for immediate alerts, while Prophet excelled in weekly predictions by handling seasonality and trends. These results validate deep learning's edge over classical methods in volatile environments like Abu Dhabi, where dust and sea breezes influence pollutant dispersion.

Real-world applicability is evident: a 1-hour sMAPE under 0.24 for NO₂ could trigger traffic adjustments or school closures, preventing exposure spikes.

Leveraging Pollutant Correlations for Enhanced Accuracy

A key innovation was augmenting features with NO₂-O₃ correlations, yielding measurable gains—especially for short-term O₃ (up to 24 hours) and 1-hour NO₂ forecasts. This multi-pollutant modeling acknowledges real atmospheric chemistry, where NO₂ photolysis produces O₃, improving holistic predictions over siloed approaches. Such strategies could extend to PM2.5 or SO₂, amplifying forecasting utility.

Partnership with Environment Agency Abu Dhabi: Data-Driven Collaboration

Data sourced exclusively from EAD's extensive network powered the models, though access is restricted. This KU-EAD synergy exemplifies UAE's public-private research ecosystem, where university expertise meets agency needs for operational tools. EAD's real-time dashboard already disseminates AQI, and AI forecasts could integrate for proactive advisories.

Public Health and Policy Implications

Accurate NO₂/O₃ forecasts enable targeted interventions, reducing health burdens. In Abu Dhabi, where pollution contributes to respiratory diseases amid a growing population, timely warnings protect asthmatics and outdoor workers. Policymakers can optimize emission controls, aligning with UAE's Green Agenda 2030. Globally, this validates AI for arid urban forecasting, adaptable to Gulf cities facing similar dust-traffic mixes.

Photo by Diviya Khanna on Unsplash

Aligning with UAE's Net Zero Ambitions and ADSW 2026

The research dovetails with UAE's sustainability drive, highlighted at Abu Dhabi Sustainability Week (ADSW) 2026, which emphasized AI in clean energy and climate resilience. KU's contributions support national goals for 50% clean energy by 2050, positioning higher education as a catalyst for innovation. For details on the study, visit the full paper.

Future Horizons: Expanding AI at Khalifa University

Authors plan multi-pollutant expansions and real-time deployment, potentially via KU's Environmental Monitoring Platform. As ENGEOS evolves, integrating satellite data could enhance spatial coverage. This positions KU researchers for grants and partnerships, fostering UAE talent in AI-environmental science.

In summary, this publication not only elevates Khalifa University's profile but equips Abu Dhabi with tools for cleaner air, exemplifying higher education's role in sustainable development.