A Forest Fire Prediction Model Based on Cellular Automotadata and Machine Learning

Abstract

Forest fires constitute a widespread and impactful natural disaster, annually ravaging millions of hectares of forests and posing a severe threat to human life and property. Accurate quantitative prediction of forest fire spread is essential for devising swift risk management strategies and implementing effective firefighting approaches. In response to this imperative, this study introduces a Forest Fire Spread Behavior Prediction (FFSBP) model, encompassing two integral components: the Forest Fire Spread Process Prediction (FFSPP)model and the Forest Fire Spread Results Prediction (FFSRP) model. The FFSPP model involves the prediction of the direction and speed of forest fire spread, achieved through a fusion of the Cellular Automatamodel and the Wang Zhengfeimodel. On the other hand, the FFSRP model focuses on forecasting the extent of the burned area, leveraging machine learning methods. To validate the efficacy of the proposed models, a real case study is undertaken using the ‘‘3.29 Forest Fire’’incident in China. The FFSPPmodel is validated against this case, while the FFSRPmodel is assessed using a real fire dataset obtained from Montesinho National Forest Park in Portugal. Results from the validationprocessrevealthatduringthenaturaldevelopmentperiodofthe ‘‘3.29ForestFire,’’the FFSPPmodelpredictsa burned area of286.81 hm2 , with anassociated relativeerror of28.94%. This relative error is notably smaller than those observed in the Farsite and Prometheus fire behavior simulation models. Additionally, the FFSRP model demonstrates commendable predictiveperformance, particularlyinthecontextofsmallandmedium-sizedfirescenarios. These findings underscore the potential of the FFSBPmodel as a valuable tool in enhancing forest fire prediction accuracy and facilitating more robust risk mitigation and firefighting strategies.