Forecasting wind-driven wildfires using an inverse modelling approach

Abstract

A technology able to rapidly forecast wildfire dynamics would lead to a paradigm shift in the response to emergencies, providing the Fire Service with essential information about the ongoing fire. This paper presents and explores a novel methodology to forecast wildfire dynamicsin wind-driven conditions, using real-time data assimilation and inverse modelling. The forecasting algorithm combines Rothermel’s rate of spread theory with a perimeter expansionmodel based on Huygens principle and solves the optimisation problem with a tangent linear approach and forward automatic differentiation. Its potential is investigated using synthetic data and evaluated in different wildfire scenarios. Theresults show the capacity of the method to quickly predict the location of the fire front with a positive lead time (ahead of the event) in the order of 10 min for a spatial scale of 100 m. The greatest strengths of our method are lightness, speed andflexibility. We specifically tailor the forecast to be efficient and computationally cheap so it can be used in mobile systems for field deployment and operativeness. Thus, we put emphasis on producing a positive lead time and the means tomaximise it.