Fire Intensity and spRead forecAst (FIRA): A Machine Learning Based Fire Spread Prediction Model for Air Quality...

Area of Science:

• Atmospheric Science
• Environmental Science
• Computer Science

Background:

• Wildfires release hazardous pollutants, impacting air quality and public health.
• Current air quality forecast (AQF) models often lack real-time fire spread dynamics, limiting prediction accuracy.

Purpose of the Study:

• To develop a novel machine learning (ML) model, Fire Intensity and spRead forecAst (FIRA), for predicting wildfire spread and intensity.
• To enhance AQF models by integrating dynamic fire characteristics like spatial distribution and fire radiative power (FRP).

Main Methods:

• Developed the FIRA ML model using 2020 CONUS and historical California fire data.
• Applied FIRA's FRP predictions as input to the Unified Forecast System coupled with smoke (UFS-Smoke) model.
• Evaluated FIRA's performance against near-real-time fire products using a September 2020 California fire case.

Main Results:

• FIRA demonstrated strong performance in capturing fire spread (R-squared ≈ 0.7) and spatial similarity (≈95%).
• FIRA predictions showed good agreement with UFS-Smoke simulations, indicating accurate representation of future fire activities.
• Scaled FIRA predictions significantly improved predictions of aerosol optical depth, 3D smoke content, and surface PM2.5 concentrations compared to the baseline UFS-Smoke model.

Conclusions:

• The FIRA model offers a significant advancement in forecasting wildfire behavior for AQF applications.
• Integrating FIRA into AQF systems can lead to more accurate predictions of wildfire smoke impacts on air quality and public health.
• While FIRA may underestimate fire intensity, applying scaling factors effectively mitigates this uncertainty.