Application of a Machine Learning Method for Prediction of Urban Neighborhood-Scale Air Pollution

Area of Science:

• Environmental Science
• Computational Science
• Urban Planning

Background:

• Urban air pollution poses significant health risks, necessitating accurate prediction models for smart city development.
• Computational Fluid Dynamics (CFD) models offer detailed dispersion behavior but are computationally intensive.
• Existing models struggle with the complexity of urban environments and real-time prediction needs.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) model, specifically an Artificial Neural Network (ANN), for predicting vehicle-derived PM10 dispersion in urban canyons.
• To compare the performance and practicality of the ANN model against a traditional CFD model for urban air quality assessment.
• To identify a computationally efficient and accurate model suitable for smart city applications.

Main Methods:

• An Artificial Neural Network (ANN) model was trained using measured meteorological data and PM10 concentrations.
• A building-resolved Computational Fluid Dynamics (CFD) model was established for comparative analysis under identical environmental conditions.
• Model performance was evaluated by comparing predictions against extensive PM10 measurements, focusing on accuracy and computational efficiency.

Main Results:

• The ANN model demonstrated promising accuracy (r = 0.82, fractional bias = 0.002) for long-term PM10 predictions.
• Both ANN and CFD models showed good performance (r > 0.8) in predicting diurnal PM10 variations on clear days.
• The ANN model significantly outperformed CFD in terms of practicality, requiring less than 0.1% of the computational time and fewer operational constraints.

Conclusions:

• The Artificial Neural Network (ANN) model is a highly effective and efficient tool for predicting urban PM10 dispersion.
• ANN models present a superior alternative to CFD for real-time air quality monitoring and management in smart cities.
• The study highlights the potential of machine learning in addressing complex environmental challenges within urban settings.