Application of artificial intelligence methods to model the effect of grass curing level on spread rate of fires

• Published in: Environmental modelling & software : with environment data news 2024-02, Vol.173, p.105930, Article 105930
• Main Authors: Khanmohammadi, Sadegh, Cruz, Miguel G., Mohammadi Golafshani, Emadaldin, Bai, Yu, Arashpour, Mehrdad
• Format: Article
• Language: English
• Subjects: Bushfire; Feature importance; Gene expression programming; Grasslands; Machine learning; Remote regions
• ISSN: 1364-8152
• DOI: 10.1016/j.envsoft.2023.105930

Artificial intelligence (AI) enables new approaches to fire behaviour models of operational relevance, including prescribed burns. This is particularly important in modelling of processes that are poorly understood, such as live fuel's effect on fire propagation. The objective of this study was to apply AI algorithms to quantify the effect of the proportion of dead fuels in a senescing grassland, the curing level, on reducing the rate of fire spread relative to the fully cured condition. We applied three different machine learning (ML) models, regression trees, support vector regression (SVR) and Gene expression programming (GEP), two ensemble ML methods, Random Forest and GEP Forest, and non-linear regression analysis to an experimental fire dataset. Results show SVR and GEP as the best ML methods to model the curing level impact on fire spread. No differences in model fit were observed between the best ML methods and non-linear regression analysis. •The effect of the proportion of dead fuels in a senescing grassland on the rate of fire spread is evaluated by using Artificial intelligence (AI).•Three different machine learning (ML) models, regression trees, support vector regression (SVR) and Gene expression programming (GEP), and two ensemble ML methods, Random Forest and GEP Forest are employed in this study.•The performance of AI models including ensemble models is compared against the non-linear regression model as a traditional approach.