Phys-Diff: A Physics-Inspired Latent Diffusion Model for Tropical Cyclone Forecasting

2026-02-28

AI Summary

This work addresses the common limitation of existing deep learning approaches in tropical cyclone forecasting—namely, their neglect of physical consistency, which often results in implausible relationships among predicted attributes such as track, central pressure, and wind speed. To overcome this, we propose Phys-Diff, a novel model that integrates physics-inspired inductive biases into a latent diffusion framework. Specifically, Phys-Diff decouples latent features into task-specific components and introduces a cross-task attention mechanism to enforce physically consistent constraints across multiple prediction targets. The model further leverages multimodal inputs, including historical tropical cyclone records, ERA5 reanalysis data, and FengWu forecast fields, to enable synergistic prediction. Experiments demonstrate that Phys-Diff achieves state-of-the-art performance on both global and regional datasets, significantly improving both forecast accuracy and physical plausibility.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Multimodal Learning
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search

Abstract

Tropical cyclone (TC) forecasting is critical for disaster warning and emergency response. Deep learning methods address computational challenges but often neglect physical relationships between TC attributes, resulting in predictions lacking physical consistency. To address this, we propose Phys-Diff, a physics-inspired latent diffusion model that disentangles latent features into task-specific components (trajectory, pressure, wind speed) and employs cross-task attention to introduce prior physics-inspired inductive biases, thereby embedding physically consistent dependencies among TC attributes. Phys-Diff integrates multimodal data including historical cyclone attributes, ERA5 reanalysis data, and FengWu forecast fields via a Transformer encoder-decoder architecture, further enhancing forecasting performance. Experiments demonstrate state-of-the-art performance on global and regional datasets.

Problem

• tropical cyclone forecasting
• physical consistency
• deep learning
• physics-inspired modeling
• multimodal data

Innovation

• physics-informed diffusion
• latent disentanglement
• cross-task attention
• multimodal fusion
• tropical cyclone forecasting