MACHINE LEARNING-BASED FOREST FIRE PREDICTION MODEL UTILIZING UAV VIDEOS

Forest fires represent a pervasive and significant natural calamity, yearly devastating millions of hectares of forest and presenting a grave danger to human life and property.Precise quantitative forecasting of forest fire propagation is crucial for formulating prompt risk management plans and executing efficient firefighting methods.This work presents a Forest Fire Spread Behavior Prediction (FFSBP) model, which has two essential components: the Forest Fire Spread Process Prediction (FFSPP) model and the Forest Fire Spread Results Prediction (FSRP) model.The FFSPP model predicts the direction and velocity of forest fire propagation by integrating the Cellular Automata model with the Wang Zhengfei model.The FFSRP model, instead, emphasizes predicting the magnitude of the burnt area by using machine learning techniques.A genuine case study using the "3.29 Forest Fire" disaster in China is conducted to assess the usefulness of the presented models.The FFSPP model is verified for this example, while the FFSRP model is evaluated using a real fire dataset sourced from Montesinho National Forest Park in Portugal.The validation procedure indicates that during the natural growth phase of the "3.29 Forest Fire," the FFSPP model forecasts a burnt area of 286.81 hm², with a relative error of 28.94%.This relative inaccuracy is much less than those recorded in the Farsite and Prometheus fire behavior simulation models.The FFSRP model has notable prediction efficacy, especially with small and medium-sized fire situations.The results highlight the FFSBP model's potential as an effective instrument for improving forest fire forecast accuracy and enabling more effective risk reduction and firefighting techniques