IoT- and AI-informed urban air quality models for vehicle pollution monitoring

TL;DR

This paper addresses the need for high-resolution, temporally-dense urban air quality monitoring by integrating low-cost IoT sensors, edge AI-based traffic emission estimation, and HPC CFD simulations within an end-to-end IoT–edge–cloud–HPC pipeline. The approach uses traffic cameras and LCS data fed through edge computing to produce real-time emission estimates, which drive a CFD-based urban dispersion model; data assimilation via Universal Kriging and a Kalman Filter enhance spatial-temporal accuracy. A real-world Barcelona pilot demonstrates privacy-conscious edge processing, reduced data transfer, and improved conformance with reference NO2 measurements relative to static emission inventories. The results highlight the approach's scalability, adaptability to other cities, and potential to support urban planning and policy decisions through higher-resolution exposure assessments and prospective digital twins.

Abstract

With the rise of intelligent Internet of Things (IoT) systems in urban environments, new opportunities are emerging to enhance real-time environmental monitoring. While most studies focus either on IoT-based air quality sensing or physics-based modeling in isolation, this work bridges that gap by integrating low-cost sensors and AI-powered video-based traffic analysis with high-resolution urban air quality models. We present a real-world pilot deployment at a road intersection in Barcelona's Eixample district, where the system captures dynamic traffic conditions and environmental variables, processes them at the edge, and feeds real-time data into a high-performance computing (HPC) simulation pipeline. Results are validated against official air quality measurements of nitrogen dioxide (NO2). Compared to traditional models that rely on static emission inventories, the IoT-assisted approach enhances the temporal granularity of urban air quality predictions of traffic-related pollutants. Using the full capabilities of an IoT-edge-cloud-HPC architecture, this work demonstrates a scalable, adaptive, and privacy-conscious solution for urban pollution monitoring and establishes a foundation for next-generation IoT-driven environmental intelligence.

Figures (5)

• Figure 1: Data flow across the IoT-enabled air quality monitoring framework, showing how raw sensor and video data are processed at the edge, integrated into HPC-based urban simulations, and visualized through a cloud platform for decision-making and public communication.
• Figure 2: Overview of the pilot site and data sources. (a) Map of the intersection showing the deployed IoT sensing infrastructure, including traffic cameras, low-cost sensors, passive dosimeters, and the reference air quality monitoring station. (b) Comparison of NO$_2$ and PM$10$ time series measured by the low-cost sensors and the reference station. (c–d) Views from the two installed traffic cameras covering the monitored area.
• Figure 3: Vehicle detection and emission analysis derived from video-based tracking. (a) Probability density function of detected speeds for all road users. (b) Speed distribution by road user type. (c) Instantaneous speed, acceleration, and emission rates for an illustrative vehicle trajectory. (d) Comparison of one-hour aggregated emissions derived from camera-based estimates and from the HERMESv3 emission model based on average speed and expected vehicle counts.
• Figure 4: Example of a computational fluid dynamics simulation at the Barcelona pilot case. The panels show (a) details of the refined computational mesh, (b) a snapshot of NO2 emissions for each road-link in the domain, and (c) the corresponding resulting NO2 concentration field near ground level.
• Figure 5: Model evaluation and data-fusion results. (a) NO$_2$ concentrations from dosimeters compared with raw CFD and Universal Kriging–corrected fields. (b) Temporal NO$_2$ evolution at the AQMS location for June 1, 2025, showing raw and Kalman Filter–corrected model predictions against observations.