P2       Recovering fire arrival time from satellite data by machine learning.

Farguell, Angel, James Haley, Lauren Hearn, and Jan Mandel, University of Colorado Denver, and Adam Kochanski, University of Utah

Estimating fire arrival time is one of the key points when using data assimilation in a coupled atmosphere-fire model. A machine learning method to estimate fire arrival time using satellite data for WRF-SFIRE is proposed. First, the Active Fire Satellite data from different instruments are processed defining, at each location in the domain, the last time the satellites observed saved ground and the first time they detected fire. Then, these two sets of data are separated using the Support Vector Machine technique giving the best soft margin hyperplane dividing both groups. Finally, the fire arrival time at each location is defined to be the minimum time forming part of the previous separating hyperplane giving promising results experimenting with 5 large wildland fires.