Periodic Reporting for period 1 - MODELAIR (Groundbreaking tools and models to reduce air pollution in urban areas)
Reporting period: 2023-01-01 to 2024-12-31
Urban air pollution poses a significant threat to public health and the environment, with the majority of Europe's urban population exposed to pollutant levels exceeding the World Health Organization's guidelines. Human activities, particularly in densely populated cities, are primary contributors to this issue. Addressing urban air pollution is crucial for achieving the European Union's mission of creating climate-neutral and smart cities, thereby reducing greenhouse gas emissions and providing cleaner air for citizens.
In response to this pressing challenge, the MODELAIR project aims to develop an Artificial Intelligence (AI)-based tool designed to facilitate informed and effective decision-making for controlling air pollution in urban environments. This initiative integrates theoretical, experimental, numerical, and data-driven approaches to simulate, control, and design innovative technologies tailored for sustainable urban development. A key component of MODELAIR is the specialized training of 12 doctoral candidates, equipping them with the expertise to implement these advanced technologies within city councils and relevant industrial sectors.
The project's pathway to impact involves close collaboration with city councils and industries in cities such as Bristol, Brussels, and Madrid. By enhancing current modeling capabilities, MODELAIR will account for the effects of urban infrastructures—like buildings and roadways—on the flow and dispersion of air pollutants. The AI-based tool's effectiveness will be evaluated through three specific applications:
i. Analyzing Urban Topology's Impact on Air Pollution: Investigating how different urban layouts influence pollutant dispersion to inform city planning and pollution mitigation strategies.
ii. Developing a Real-Time Decision-Making Tool for Brussels' Ixelles District: Creating a system to optimize the placement and maintenance of air quality sensors, ensuring high-quality monitoring and rapid response capabilities.
iii. Optimizing Traffic Routes to Minimize Emissions: Studying emission sources to devise traffic management plans that reduce pollution levels.
By producing transferable outputs and employing novel methodologies, MODELAIR aims to significantly improve air quality dispersion models. These advancements will empower city councils and industries to develop and implement effective strategies to combat urban air pollution, thereby contributing to the broader goal of sustainable and healthy urban living environments.
While the primary focus of MODELAIR is on technological and scientific innovation, the integration of social sciences and humanities is essential for understanding the societal implications and ensuring the successful adoption of these new technologies. Engaging with local communities, policymakers, and other stakeholders will facilitate the development of solutions that are not only technically effective but also socially acceptable and aligned with public needs and values.
In response to this pressing challenge, the MODELAIR project aims to develop an Artificial Intelligence (AI)-based tool designed to facilitate informed and effective decision-making for controlling air pollution in urban environments. This initiative integrates theoretical, experimental, numerical, and data-driven approaches to simulate, control, and design innovative technologies tailored for sustainable urban development. A key component of MODELAIR is the specialized training of 12 doctoral candidates, equipping them with the expertise to implement these advanced technologies within city councils and relevant industrial sectors.
The project's pathway to impact involves close collaboration with city councils and industries in cities such as Bristol, Brussels, and Madrid. By enhancing current modeling capabilities, MODELAIR will account for the effects of urban infrastructures—like buildings and roadways—on the flow and dispersion of air pollutants. The AI-based tool's effectiveness will be evaluated through three specific applications:
i. Analyzing Urban Topology's Impact on Air Pollution: Investigating how different urban layouts influence pollutant dispersion to inform city planning and pollution mitigation strategies.
ii. Developing a Real-Time Decision-Making Tool for Brussels' Ixelles District: Creating a system to optimize the placement and maintenance of air quality sensors, ensuring high-quality monitoring and rapid response capabilities.
iii. Optimizing Traffic Routes to Minimize Emissions: Studying emission sources to devise traffic management plans that reduce pollution levels.
By producing transferable outputs and employing novel methodologies, MODELAIR aims to significantly improve air quality dispersion models. These advancements will empower city councils and industries to develop and implement effective strategies to combat urban air pollution, thereby contributing to the broader goal of sustainable and healthy urban living environments.
While the primary focus of MODELAIR is on technological and scientific innovation, the integration of social sciences and humanities is essential for understanding the societal implications and ensuring the successful adoption of these new technologies. Engaging with local communities, policymakers, and other stakeholders will facilitate the development of solutions that are not only technically effective but also socially acceptable and aligned with public needs and values.
The MODELAIR project integrates theoretical, experimental, numerical, and data-driven approaches to develop groundbreaking tools and models for reducing air pollution in urban areas. The project is structured around key technical objectives, which have been pursued through seven work packages (WP), each contributing to the advancement of urban air quality modeling and AI-based decision-making.
Key Technical Achievements
1. High-Fidelity Urban Airflow Simulations and Experimental Validation
MODELAIR has successfully conducted high-resolution simulations and experimental validations to enhance the accuracy of urban pollution models:
- Direct Numerical Simulations (DNS) and Large Eddy Simulations (LES) were used to model urban airflow over buildings and pollutant transport mechanisms.
- CFD simulations captured turbulence effects and pollutant dispersion, integrating real urban geometries in cities like Madrid, Brussels, and Bristol.
- Wind tunnel experiments at the University of Bristol validated numerical models with Particle Image Velocimetry (PIV) and hot-wire anemometry.
- Drone-based LiDAR scanning generated high-resolution 3D urban geometries for accurate digital reconstructions.
2. Data Merging and Sensor Integration for Real-Time Pollution Forecasting
To improve urban air quality predictions, MODELAIR developed advanced data assimilation techniques that combine numerical simulations with real-world data.
- Ensemble Kalman Filter (EnKF)-based Data Assimilation (DA) techniques were applied to reduce uncertainty in urban pollution forecasts.
- Machine learning-based sensor placement strategies optimized real-time monitoring and enhanced predictive accuracy.
- Deep generative models and diffusion models reconstructed sparse sensor data into high-fidelity urban pollution maps.
3. Physics Identification and Pattern Extraction Using AI
MODELAIR is leveraging AI-driven tools to extract meaningful physics from complex urban flow dynamics:
- Physics-Informed Neural Networks (PINNs) integrated physical constraints into data-driven models for pollutant transport analysis.
- Causal AI frameworks were developed to identify hidden relationships between urban topology and pollutant dispersion.
- Higher-Order Singular Value Decomposition (HOSVD) and Dynamic Mode Decomposition (DMD) extracted dominant airflow and pollution patterns.
4. Development of Reduced Order Models (ROMs) for Real-Time Air Quality Assessment
The project developed low-dimensional surrogate models that enable real-time decision-making by city councils and industries:
- Autoencoder-based ROMs compressed high-dimensional urban airflow data while preserving key turbulence structures.
- Hybrid forecasting models integrated Singular Value Decomposition (SVD) with recurrent neural networks (RNNs) for real-time airflow prediction.
- Deep reinforcement learning models were applied for non-intrusive sensing and sensor placement optimization.
Expected Outcomes and Future Steps
- Deployment of AI-based tools for air pollution mitigation in collaboration with city councils and industry partners.
- Integration of high-fidelity CFD models with real-time sensor data to provide actionable insights for urban planning.
- Validation of reduced-order models (ROMs) in real urban environments to enable fast and efficient air quality forecasting.
- Further improvements in data assimilation and uncertainty quantification techniques to enhance prediction reliability.
-During the next reporting period, the MODELAIR tools are planned to be tested in three cities, to evaluate their real-world effectiveness:
- Bristol: Analysis of urban topology’s effect on pollution dispersion using city council air quality data.
- Brussels (Ixelles district): Deployment of a real-time decision-making tool integrating data assimilation and sensor networks.
-Madrid: Optimization of traffic routes to reduce air pollution through high-fidelity CFD and AI-driven predictive models.
Key Technical Achievements
1. High-Fidelity Urban Airflow Simulations and Experimental Validation
MODELAIR has successfully conducted high-resolution simulations and experimental validations to enhance the accuracy of urban pollution models:
- Direct Numerical Simulations (DNS) and Large Eddy Simulations (LES) were used to model urban airflow over buildings and pollutant transport mechanisms.
- CFD simulations captured turbulence effects and pollutant dispersion, integrating real urban geometries in cities like Madrid, Brussels, and Bristol.
- Wind tunnel experiments at the University of Bristol validated numerical models with Particle Image Velocimetry (PIV) and hot-wire anemometry.
- Drone-based LiDAR scanning generated high-resolution 3D urban geometries for accurate digital reconstructions.
2. Data Merging and Sensor Integration for Real-Time Pollution Forecasting
To improve urban air quality predictions, MODELAIR developed advanced data assimilation techniques that combine numerical simulations with real-world data.
- Ensemble Kalman Filter (EnKF)-based Data Assimilation (DA) techniques were applied to reduce uncertainty in urban pollution forecasts.
- Machine learning-based sensor placement strategies optimized real-time monitoring and enhanced predictive accuracy.
- Deep generative models and diffusion models reconstructed sparse sensor data into high-fidelity urban pollution maps.
3. Physics Identification and Pattern Extraction Using AI
MODELAIR is leveraging AI-driven tools to extract meaningful physics from complex urban flow dynamics:
- Physics-Informed Neural Networks (PINNs) integrated physical constraints into data-driven models for pollutant transport analysis.
- Causal AI frameworks were developed to identify hidden relationships between urban topology and pollutant dispersion.
- Higher-Order Singular Value Decomposition (HOSVD) and Dynamic Mode Decomposition (DMD) extracted dominant airflow and pollution patterns.
4. Development of Reduced Order Models (ROMs) for Real-Time Air Quality Assessment
The project developed low-dimensional surrogate models that enable real-time decision-making by city councils and industries:
- Autoencoder-based ROMs compressed high-dimensional urban airflow data while preserving key turbulence structures.
- Hybrid forecasting models integrated Singular Value Decomposition (SVD) with recurrent neural networks (RNNs) for real-time airflow prediction.
- Deep reinforcement learning models were applied for non-intrusive sensing and sensor placement optimization.
Expected Outcomes and Future Steps
- Deployment of AI-based tools for air pollution mitigation in collaboration with city councils and industry partners.
- Integration of high-fidelity CFD models with real-time sensor data to provide actionable insights for urban planning.
- Validation of reduced-order models (ROMs) in real urban environments to enable fast and efficient air quality forecasting.
- Further improvements in data assimilation and uncertainty quantification techniques to enhance prediction reliability.
-During the next reporting period, the MODELAIR tools are planned to be tested in three cities, to evaluate their real-world effectiveness:
- Bristol: Analysis of urban topology’s effect on pollution dispersion using city council air quality data.
- Brussels (Ixelles district): Deployment of a real-time decision-making tool integrating data assimilation and sensor networks.
-Madrid: Optimization of traffic routes to reduce air pollution through high-fidelity CFD and AI-driven predictive models.
The MODELAIR project has advanced urban air pollution modeling by integrating AI-driven forecasting, high-fidelity CFD simulations (DNS, LES, RANS), and real-world data assimilation (EnKF, ROMs, PINNs) to enhance prediction accuracy and decision-making. Its innovative deep learning-based ROMs enable rapid pollution assessments, while wind tunnel experiments and LiDAR-based 3D reconstructions improve model validation.