Probabilistic Forecasting of Wildfire Size Using Conditional Flow Matching

Key points

• Rather than predicting spatially resolved fire progression, the proposed model forecasts a time series of fire acreage, substantially reducing computational cost and enabling efficient integration with downstream applications such as early warning systems and risk assessment tools.1
• Training data are drawn from 140 coupled atmosphere-wildfire simulations generated using WRF-SFIRE, paired with weather data from the North American Mesoscale (NAM) model.1
• Model performance is evaluated against reference solutions from WRF-SFIRE.1
• Recursive multi-step predictions over 24 hours are performed for 12 held-out test cases, where the flow matching model achieves a final fire area sMAPE of 0.39 while further providing meaningful quantification of forecast uncertainty.1

Sources (1)

• [1]Probabilistic Forecasting of Wildfire Size Using Conditional Flow Matching
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 11, 11:32 PM

  In this work we present a probabilistic model for predicting the temporal evolution of wildfire size using conditional generative algorithms, namely flow matching.

  Rather than predicting spatially resolved fire progression, the proposed model forecasts a time series of fire acreage, substantially reducing computational cost and enabling efficient integration with downstream applications such as early warning systems and risk assessment tools.

Extractive summary: sentences quoted from the sources.