(394v) Toward Developing an Agent Based Digital Twin for Air Quality Management

Urban ambient air pollution has a significant impact on human health and overall well-being. Tackling this issue requires identifying the sources, understanding the effects, and developing practical interventions. Most existing modelling paradigms tend to focus on just one of these aspects - either describing, predicting, or prescribing solutions. In this work, we present a new framework that applies an agent-based approach to build a digital twin of a region, allowing us to assess and manage air quality with fine spatial and temporal detail. The region is broken down into smaller units, or agents, which interact with each other and the environment through pollutant exchange. At the heart of the model is a mass balance that simplifies the representation of pollution transport, removing the need for complex dispersion models. However, it learns from real world data including monitoring data from sensor networks, land use, meteorology among others. The framework includes several adjustable parameters that are tied to real-world characteristics, making it practical and adaptable. We demonstrate the framework through PM2.5 data collected through a mobile monitoring campaign in Chennai, India. It is successful in capturing the variation in PM2.5 concentration at high granularity. The descriptive, predictive and prescriptive capabilities of the model are also demonstrated. By combining the strengths of both physics-based and data-driven methods, this approach offers a unified way to describe current conditions, predict future scenarios, and suggest targeted interventions for improving air quality.