Background
Air pollution has major adverse effects on health and climate. Air pollutant concentration can exhibit significant
fluctuations over short spatial distances due to unevenly spread emission sources, meteorological conditions, physical characteristics of pollutants, particles and chemical processes. The increasing concern over atmospheric pollution demands strong predictive models for pollutant dispersion. Traditional models including mechanism-based methods, statistics-based machine learning methods and even chemical transport models often face limitations in capturing complex atmospheric dynamics accurately. Air pollution prediction with Physics Informed Machine Learning (PIML) show potential in overcoming these challenges by combining data-driven techniques with a strong foundation in physics.
Objectives
The primary objective of the research is developing a comprehensive PIML model for simulating air pollutant dispersion and transport. A secondary objective involves enhancing accuracy and applicability of the developed PIML model by incorporating real world data sources.
Methods
The first phase of the methodology involves leveraging advection-diffusion equations as the governing equation,
utilizing the Eulerian method to predict pollutant concentrations spatially. By incorporating physical constraints into the learning process, the models will better capture the underlying dynamics of atmospheric pollutant dispersion.
Expected Results
Expected results anticipate a significant advancement in predictive accuracy for spatial distributions for air pollutants compared to traditional machine learning models.
Significance and Impact
The integration of advection-diffusion equation with PIML represents a significant advancement as a predictive model for atmospheric dispersion. This potentially ground-breaking approach holds promise for further refining atmospheric pollutant dispersion predictions