Physics-Guided AI Improves Fuel Density Prediction for Fire Management.

Accurate fuel density prediction is crucial for effective fire management, particularly for adaptive prescribed burn strategies. Traditional data-driven models often lack the stability and physical plausibility needed for such critical applications. This paper introduces a physics-guided machine learning (PGML) framework designed to improve fuel density prediction. The framework integrates physics constraints and domain knowledge directly into deep learning models, enhancing both accuracy and stability. It explores three deep learning architectures: ConvLSTM, Adaptive Fourier Neural Operator (AFNONet), and Video Vision Transformer (ViViT) to model the spatiotemporal evolution of fuel density. The core of the approach involves incorporating differentiable physics-informed terms into the loss function, specifically a mass-conserving fuel transport term and a rate-of-spread estimation. Experimental results consistently show that this PGML framework outperforms purely data-driven baselines, delivering more accurate and stable predictions. This advancement enables computationally efficient and physically plausible fire forecasting, directly supporting better prescribed burn management.