Source Term Estimation for Puff Releases Using Machine Learning: A Case Study
Reliable source term prediction for hazardous pollutant puffs in urban microenvironments is challenging, especially for risk management under strict time constraints. Puff movement is highly stochastic due to atmospheric turbulence, intensified by complex urban canopies. This complexity, combined wi...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-06-01
|
| Series: | Atmosphere |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2073-4433/16/6/697 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Reliable source term prediction for hazardous pollutant puffs in urban microenvironments is challenging, especially for risk management under strict time constraints. Puff movement is highly stochastic due to atmospheric turbulence, intensified by complex urban canopies. This complexity, combined with time limitations, makes advanced computational modeling impractical. A more efficient approach is leveraging past and present data using Machine Learning (ML) techniques. This study proposes an ML-based method, enriched with simplified physical modeling, for source term estimation of unforeseen hazardous air releases in monitored urban areas. The Random Forest Regression, commonly used in meteorology and air quality studies, has been selected. A novel variable selection method is introduced, including the following: (a) a model-derived Exposure Burden Index (EBI) reflecting plume–morphology interactions; (b) a plume travel time indicator; (c) the standard deviation of input variables capturing stochastic behavior; and (d) the total dosage-to-mass released ratio at sensor locations as the target variable. The case study examines JU2003 field experiments involving SF<sub>6</sub> puffs released at street level in Oklahoma City’s urban core, a challenging scenario due to the limited number of sensors and historical data. Results demonstrate the approach’s effectiveness, offering a promising, realistic alternative to traditional computationally intensive methods. |
|---|---|
| ISSN: | 2073-4433 |