Related Experiment Video
Updated: Aug 6, 2026
07:12
Façade-Level Monitoring of CO2 Variability under Urban Heat Island Conditions using Low-Cost Sensor Data Loggers
Published on: December 12, 2025
Integrating Audiovisual Data for Short-Term Particulate Matter Concentration Prediction on Urban Sidewalks
Xujing Yu1, Jun Ma1, Waishan Qiu1
1Department of Urban Planning and Design, Faculty of Architecture, The University of Hong Kong, Hong Kong000000, China.
Environmental Science & Technology
|July 22, 2026
Summary
This study uses street-view videos and machine learning to predict sidewalk air pollution (PM2.5 and PM1). This novel approach offers high-resolution urban air quality monitoring, improving pedestrian exposure assessment.
Area of Science:
- Environmental Science
- Data Science
- Urban Planning
Background:
- Urban air pollution poses significant risks to pedestrians.
- Traditional monitoring methods lack the spatial resolution for accurate exposure assessment.
- High costs limit the deployment of dense monitoring networks.
Purpose of the Study:
- To develop a machine learning framework for predicting short-term sidewalk particulate matter (PM2.5 and PM1) concentrations.
- To utilize multimodal audiovisual features from street-view videos for air pollution estimation.
- To enhance urban air quality management through scalable, high-resolution monitoring.
Main Methods:
- A mobile monitoring campaign was conducted in Shenzhen, China.
- Multimodal audiovisual features were extracted from self-collected street-view videos.
- Machine learning models (Linear Regression, XGBoost, LightGBM) were evaluated with meteorological and background pollution data.
- Temporal resolutions of 10 seconds and 1 minute were assessed.
Main Results:
- LightGBM model achieved R2 values of 0.64-0.65 for 10-second predictions and 0.80 for 1-minute predictions (random cross-validation).
- Spatial cross-validation showed moderate generalizability (R2 of 0.41-0.48 at 1-minute resolution).
- A hybrid model incorporating geospatial context improved predictive accuracy.
- Audiovisual features provided significant predictive power beyond background pollution and meteorological data.
Conclusions:
- Integrating multimodal audiovisual sensing with ancillary data enables scalable, high-resolution estimation of street-level PM.
- This approach effectively complements conventional monitoring for urban air quality management.
- The framework offers a cost-effective solution for fine-scale exposure assessment in urban environments.
