Quantifying wildfire impacts on atmospheric pollutants using fire exposure metrics and machine-deep learning
Khushal Das1, Sergio Flesca2, Claudia Roberta Calidonna3
1DIMES, University of Calabria, Rende CS, 87036, Italy. khushaldasparmar@gmail.com.
Scientific Reports
|May 8, 2026
Summary
Wildfire smoke significantly impacts regional air quality, affecting gas concentrations like carbon monoxide. This study developed a Fire Exposure Index (FEI) and used machine learning to predict these effects, improving air quality management strategies.
Area of Science:
- Atmospheric Science
- Environmental Science
- Data Science
Background:
- Wildfires are recognized as significant sources of atmospheric pollution.
- The spatial-temporal impact of wildfires on regional air quality is not well understood.
- Accurate assessment of wildfire-induced air pollution is crucial for public health and environmental management.
Purpose of the Study:
- To quantify the impact of wildfire activity on atmospheric gas concentrations (CO, CO2, CH4, BC).
- To develop and validate a novel Fire Exposure Index (FEI) for assessing wildfire plume influence.
- To evaluate the effectiveness of advanced machine learning techniques in predicting wildfire-related air quality changes.
Main Methods:
- Integration of continuous atmospheric observations and wildfire data.
- Development of a Fire Exposure Index (FEI) considering fire proximity, burned area, and wind patterns.
- Application of correlation analysis, baseline machine learning models (Gradient Boosting, Random Forest, Decision Tree), and recurrent neural networks (LSTM, GRU, BiLSTM).
- Implementation of a stacked ensemble architecture combining multiple machine learning models.
Main Results:
- Correlation analyses showed clear distance and wind-dependent relationships between wildfires and gas concentrations, with short-term lag effects.
- Recurrent neural networks demonstrated stronger temporal dependency capture compared to baseline models, especially for CO2 and CO.
- The hybrid ensemble framework achieved high performance (R-squared values 0.959-0.9897) in predicting atmospheric variability.
- The developed FEI effectively quantified wildfire plume influence on the observatory.
Conclusions:
- Integrating exposure metrics with hybrid ensemble learning provides a robust strategy for predicting wildfire-induced atmospheric variability.
- The proposed approach offers a promising and interpretable method for enhancing air quality management in fire-prone regions.
- Accurate prediction of wildfire impacts is essential for mitigating air pollution and protecting regional air quality.
