PhysFire-WM: A Physics-Informed World Model for Emulating Fire Spread Dynamics

2025-12-18

AI Summary

Existing wildfire spread prediction methods rely on binary masks, resulting in sparse spatiotemporal signals and weak dynamic modeling capability; while world models show promise for video generation, their physical inconsistency undermines prediction reliability. This work addresses fine-grained wildfire spread forecasting by integrating infrared thermal imagery with fire mask sequences to build a physically consistent world model. Key contributions are: (1) the first incorporation of combustion dynamics priors into a world model architecture; (2) a cross-task collaborative training (CC-Train) strategy that jointly optimizes thermal radiation modeling and fire boundary geometry estimation; and (3) a physics-informed framework featuring simulator-guided structured prior encoding, cross-task parameter sharing with gradient coordination, and multimodal temporal representation learning. Evaluated on a newly constructed fine-grained multimodal wildfire dataset, our method significantly improves trajectory prediction accuracy while enhancing physical plausibility and geometric fidelity.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Fine-grained fire prediction plays a crucial role in emergency response. Infrared images and fire masks provide complementary thermal and boundary information, yet current methods are predominantly limited to binary mask modeling with inherent signal sparsity, failing to capture the complex dynamics of fire. While world models show promise in video generation, their physical inconsistencies pose significant challenges for fire forecasting. This paper introduces PhysFire-WM, a Physics-informed World Model for emulating Fire spread dynamics. Our approach internalizes combustion dynamics by encoding structured priors from a Physical Simulator to rectify physical discrepancies, coupled with a Cross-task Collaborative Training strategy (CC-Train) that alleviates the issue of limited information in mask-based modeling. Through parameter sharing and gradient coordination, CC-Train effectively integrates thermal radiation dynamics and spatial boundary delineation, enhancing both physical realism and geometric accuracy. Extensive experiments on a fine-grained multimodal fire dataset demonstrate the superior accuracy of PhysFire-WM in fire spread prediction. Validation underscores the importance of physical priors and cross-task collaboration, providing new insights for applying physics-informed world models to disaster prediction.

Problem

• Predicting fine-grained fire spread dynamics using multimodal data.
• Addressing physical inconsistencies in world models for fire forecasting.
• Integrating thermal and boundary information to improve prediction accuracy.

Innovation

• Physics-informed world model with combustion dynamics priors
• Cross-task collaborative training integrating thermal and boundary information
• Parameter sharing and gradient coordination for physical realism