A Comprehensive Framework Incorporating Hybrid Deep Learning Model, Vi-Net, for Wildfire Spread Prediction and Optimized Safe Path Planning

Abstract

Forest fires are becoming more prevalent than ever, and their intensity and frequency are expected only to increase owing to climate change and environmental degradation. These fires severely threaten the economy, human lives, and infrastructure. Therefore, effective management of wildfires is of utmost importance, and accurately predicting the wildfire spread lies at the core of it. Reliable predictions of fire spread not only provide insights about the at-risk regions but also help in planning several mitigation activities including resource allocation and evacuation planning. This thesis introduces Vi-Net, an innovative hybrid deep learning model, which integrates the localized precision of U-Net with the global contextual awareness of Vision Transformers (ViT) to predict next-day wildfire spread with unprecedented accuracy.

This study utilizes an extensive multimodal dataset that accumulates data from different sources across the United States from 2012 to 2020 incorporating critical factors such as topographical, meteorological, anthropological (population density), and vegetation indices. These elements are vital for modeling the complex dynamics of wildfire spread. A significant challenge in this domain is the class imbalance as the fire points are generally quite less compared to non-fire points. The dataset used in this study had fire regions less than 5% of the total data. To address this issue, advanced loss functions, including Focal Tversky Loss (FTL), are employed, prioritizing accurate segmentation of fire-prone regions while minimizing false negatives. FTL modifies the focus towards hard-to-predict regions and crucial boundaries, thereby enhancing the model's predictive accuracy and reliability in practical scenarios.

Vi-Net addresses the complexities in modeling fire dynamics by synergizing the strengths of U-Net and ViT. Integrating U-Net and ViT in Vi-Net allows for a comprehensive analysis that ensures high precision and recall, effectively balancing the sensitivity and specificity needed in wildfire predictions. This dual approach allows the model to process detailed local information and extensive contextual data, making it exceptionally capable of identifying and predicting fire spread across diverse landscapes. Experimental results highlight the superiority of Vi-Net over traditional models, achieving an F1 Score of 97.25% and an Intersection over Union (IoU) of 94.15% on the test dataset. These metrics highlight its capability to accurately capture localized fire patches and long-range dependencies while avoiding overprediction. These advancements validate the model's potential to offer more nuanced predictions, capturing the interplay between micro and macro-level environmental dynamics.

In addition to predictive modeling, this research extends its practical applicability by integrating the predicted fire masks into an optimized A* algorithm for safe path planning. This step ensures actionable insights for emergency response teams, facilitating efficient evacuation routes and resource allocation while avoiding high-risk fire regions. Qualitative and quantitative analyses confirm the hybrid model's efficacy, with visualizations demonstrating Vi-Net’s ability to preserve spatial detail while capturing broad environmental contexts, and path planning results illustrating the model's robustness and reliability.

This research not only sets a new benchmark for wildfire prediction models but also demonstrates the potential of hybrid deep learning systems in environmental science applications. By providing a robust framework for real-time wildfire management, Vi-Net could significantly influence future strategies in disaster response and resource allocation. Future enhancements could include integrating real-time data feeds to further improve the adaptability and predictive capabilities of the model, potentially revolutionizing wildfire management practices globally.