Emulating Wildfire Plume Injection Using Machine Learning Trained by Large Eddy Simulation (LES).

Area of Science:

• Atmospheric Science
• Environmental Modeling
• Machine Learning Applications

Background:

• Wildfires significantly impact the Earth system and society, but accurately modeling their atmospheric effects remains challenging.
• Wildfire plume rise (injection) is a poorly resolved process, introducing significant uncertainty into air quality and climate model assessments.
• Existing plume rise models, like the widely used Freitas Scheme, exhibit substantial uncertainties.

Purpose of the Study:

• To develop and evaluate a novel machine learning-based plume rise emulator for wildfire modeling.
• To enhance the accuracy and computational efficiency of representing wildfire plume injection in air quality and climate models.
• To create a robust, transparent, and interpretable machine learning system for wildfire impact assessments.

Main Methods:

• Developed a machine learning plume rise emulator (PRESML) trained on high-resolution, turbulence-resolving large eddy simulation (LES) data coupled with microphysics.
• Utilized a bagging ensemble technique to improve the emulator's robustness and mitigate internal variability.
• Implemented measures to ensure model transparency, prevent overtraining, and validate physical soundness of results.

Main Results:

• The machine learning emulator demonstrated superior accuracy and computational efficiency compared to the benchmark Freitas scheme.
• The bagging ensemble further enhanced the robustness of the plume rise predictions.
• The developed emulator provides interpretable and physically sound results.

Conclusions:

• The Plume Rise Emulating System using Machine Learning (PRESML) presents a promising advancement for wildfire impact modeling.
• This system offers a more accurate and efficient solution for incorporating wildfire plume dynamics into regional and global air quality and chemistry-climate models.
• PRESML has the potential to reduce uncertainties in evaluating the broader impacts of wildfires on air quality and climate.