Seyed Hamid Delbari
Ph.D. Student | September 2022
Hamid joined SFU/SEE/CREATE research lab as a Ph.D. student in September 2022. He has a BSc. in mechanical engineering from Shahid Beheshti University and an MSc. in energy systems engineering specializing in hybrid renewable electricity planning from the University of Tehran, Tehran, Iran. Hamid also has working experience in both the industry and building sectors as an M&V expert focusing on energy auditing.
His research will focus on incorporating AI-enhanced data-driven methods into physics-based atmospheric dispersion models to study the impact of traffic-related air pollutants (TRAP) under alternative transportation scenarios, such as reduced vehicle kilometers traveled (VkT) and mobility as a service (MaaS). He is part of a research team involved in the ongoing Decarbonization of Urban Mobility by Reducing VkT project for the city of Burnaby.
The summary of the main points and challenges of this project is outlined below:
- In this project, the effects of various transportation modes on the concentration of criteria air contaminants (CACs) are studied using a coupled numerical weather prediction (NWP) and chemical transport modeling (CTM) approach implemented in WRF-CMAQ.
- The complex and multiscale spatiotemporal dynamics of atmospheric pollutant species require high-resolution simulations over the area of interest, which are prohibitively resource-intensive in terms of computational cost.
- The predicted surface-level CACs concentration output from the model will be used to investigate the co-impacts of air pollution exposure and health, emphasizing the necessity of rendering accurate and reliable predictions.
- Furthermore, utilizing an agent-based model to design alternative transportation modes, which takes into account individual preferences and their interactions with external factors such as weather, investment, and policy, results in a high-dimensional scenario space, each of which requires a computationally demanding atmospheric CTM simulation.
To decrease both the computational time and cost of evaluating a variety of scenarios while achieving highly accurate and reliable results, it is necessary to study the potential of integrating WRF-CMAQ with AI-based data-driven approaches. A few promising integration opportunities are:
- Rendering high-resolution outputs (downscaling) from faster, low-resolution CTMs using deep neural network (DNN) architectures.
- Leveraging DNNs and CTMs to infer the complex nonlinear response of pollutant concentrations to their precursor emissions.
- Emulating semi-empirical microphysics schemes of physics-based models with physics-informed neural networks.
- Developing fast, robust neural network-assisted solvers for stiff kinetics in atmospheric chemistry.
- Bias correction of numerical model outputs using standard machine learning methods or DNNs.
A hybrid AI and physics-based atmospheric dispersion model leverages the vast amount of historical remote sensing data (satellite retrievals, synoptic and air quality monitoring stations, land use, topology, etc.) to learn the latent correlations in an atmospheric system, reducing the computational cost of air pollution modeling while enhancing the robustness and reliability of predictions.