Agent-Based Framework for Modelling Hyperlocal Air Quality in Urban Spaces

SATHISH SWAMINATHAN, Raghunathan Rengaswamy, V. Faye McNeill, Columbia University

     Abstract Number: 334
     Working Group: Advancing Aerosol Science through Data Analysis Tools

Abstract
Understanding and managing air quality at hyperlocal scales is crucial for public health, environmental justice, and smart urban planning. Conventional modeling approaches fall into two groups: first principles-based transport models, which are highly accurate but computationally intensive, and empirical models, which are efficient but often lack grounding in the fundamental physics of pollutant behavior. We introduce a novel multi-agent framework that bridges these paradigms by incorporating data-driven learning rooted in fundamental physics. The architecture leverages agent-based modeling to represent heterogeneous urban elements and behaviors, while maintaining a physically consistent core through mass conservation principles. The framework leads to an intelligent system capable of descriptive, predictive, and prescriptive environmental analytics. It is capable of assimilating real-world data to derive insights, make predictions, and inform interventions. By achieving a balance between computational efficiency and physical fidelity, this hybrid framework enables scalable and responsive air quality assessment at street-level granularity. This work details the development of the framework, including its physical and computational structure, and showcase its descriptive, predictive and prescriptive abilities. Hyperlocal mobile monitoring data for PM2.5 from Chennai, India is used to demonstrate the framework. The results show the framework’s potential for enabling context-aware, high-granularity air quality decision support systems in rapidly evolving urban environments.