Toward Intelligent Forest Fire Management: Hybrid Tree- Neural Models for Extreme Forest Fire Spread Prediction

The frequency and intensity of forest fires are escalating owing to climate change, requiring sophisticated and prompt management techniques. This study proposes an integrated framework that integrates forest fire detection with fire spread prediction through hybrid tree-neural models. Deep learning architectures, including CNN, EfficientNet, and ConvLSTM, are utilized for precise fire detection, whilst a hybrid XGBoost–LSTM model is implemented to forecast burned area and extreme fire spread. The framework integrates spatial, temporal, and meteorological features, alongside preprocessing and exploratory analysis, to address issues including data imbalance and heavy-tailed distributions. Experimental findings indicate that EfficientNet-B0 attains a detection accuracy of 96.9%, whilst the hybrid model surpasses others with a R² = 0.93 and reduced prediction errors. The suggested methodology facilitates early detection, enhanced decision-making, and optimal resource distribution, thereby enhancing intelligent forest fire management and scalable disaster response frameworks.