Machine Learning and Deep Learning for Wildfire Spread Prediction: A Review

Henintsoa S. Andrianarivony, Moulay A. Akhloufi

The increasing frequency and intensity of wildfires highlight the need to develop more efficient tools for firefighting and management, particularly in the field of wildfire spread prediction. Classical wildfire spread models have relied on mathematical and empirical approaches, which have trouble capturing the complexity of fire dynamics and suffer from poor flexibility and static assumptions. The emergence of machine learning (ML) and, more specifically, deep learning (DL) has introduced new techniques that significantly enhance prediction accuracy. ML models, such as support vector machines and ensemble models, use tabular data points to identify patterns and predict fire behavior. However, these models often struggle with the dynamic nature of wildfires. In contrast, DL approaches, such as convolutional neural networks (CNNs) and convolutional recurrent networks (CRNs), excel at handling the spatiotemporal complexities of wildfire data. CNNs are particularly effective at analyzing spatial data from satellite imagery, while CRNs are suited for both spatial and sequential data, making them highly performant in predicting fire behavior. This paper presents a systematic review of recent ML and DL techniques developed for wildfire spread prediction, detailing the commonly used datasets, the improvements achieved, and the limitations of current methods. It also outlines future research directions to address these challenges, emphasizing the potential for DL to play an important role in wildfire management and mitigation strategies.