Parameter Flexible Wildfire Prediction Using Machine Learning Techniques: Forward and Inverse Modelling

• Published in: Remote sensing (Basel, Switzerland) Vol. 14; no. 13; p. 3228
• Main Authors: Cheng, Sibo, Jin, Yufang, Harrison, Sandy P., Quilodrán-Casas, César, Prentice, Iain Colin, Guo, Yi-Ke, Arcucci, Rossella
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.07.2022; MDPI
• Subjects: Air pollution; Algorithms; Artificial Intelligence; California; Computer applications; Computer Science; convolutional autoencoder; Data assimilation; Data collection; Data compression; Deep learning; Dynamical systems; Environmental Engineering; Environmental Sciences; Fluid dynamics; Kalman filters; latent assimilation; Learning algorithms; Machine Learning; Mathematical models; Model reduction; Modelling; Parameter estimation; Parameter identification; prediction; Prediction models; Real time; Reduced order models; reduced-order modelling; Satellite observation; satellites; Statistics; Vegetation; Weather forecasting; wildfire prediction; Wildfires; United States > US; California; reduced-order modelling; data assimilation; machine learning; latent assimilation; convolutional autoencoder; parameter identification
• ISSN: 2072-4292; 2072-4292
• DOI: 10.3390/rs14133228

Parameter identification for wildfire forecasting models often relies on case-by-case tuning or posterior diagnosis/analysis, which can be computationally expensive due to the complexity of the forward prediction model. In this paper, we introduce an efficient parameter flexible fire prediction algorithm based on machine learning and reduced order modelling techniques. Using a training dataset generated by physics-based fire simulations, the method forecasts burned area at different time steps with a low computational cost. We then address the bottleneck of efficient parameter estimation by developing a novel inverse approach relying on data assimilation techniques (latent assimilation) in the reduced order space. The forward and the inverse modellings are tested on two recent large wildfire events in California. Satellite observations are used to validate the forward prediction approach and identify the model parameters. By combining these forward and inverse approaches, the system manages to integrate real-time observations for parameter adjustment, leading to more accurate future predictions.