A Unified Model for Human Mobility Generation in Natural Disasters

TL;DR

This work tackles the challenge of generalizing human mobility generation to unseen natural disasters and cities. It proposes UniDisMob, a diffusion-transformer framework augmented with physics-informed prompts and physics-guided alignment, coupled with a meta-learning strategy to separate universal and city-specific knowledge. The spatiotemporal decay model captures universal disaster-induced mobility patterns and guides trajectory generation, while meta-learning enables rapid adaptation to new cities. Extensive experiments across seven cities and multiple disaster types show that UniDisMob significantly outperforms state-of-the-art baselines, including strong zero-shot gains, and ablations confirm the critical roles of alignment and physics-informed conditioning in achieving realistic disaster mobility dynamics.

Abstract

Human mobility generation in disaster scenarios plays a vital role in resource allocation, emergency response, and rescue coordination. During disasters such as wildfires and hurricanes, human mobility patterns often deviate from their normal states, which makes the task more challenging. However, existing works usually rely on limited data from a single city or specific disaster, significantly restricting the model's generalization capability in new scenarios. In fact, disasters are highly sudden and unpredictable, and any city may encounter new types of disasters without prior experience. Therefore, we aim to develop a one-for-all model for mobility generation that can generalize to new disaster scenarios. However, building a universal framework faces two key challenges: 1) the diversity of disaster types and 2) the heterogeneity among different cities. In this work, we propose a unified model for human mobility generation in natural disasters (named UniDisMob). To enable cross-disaster generalization, we design physics-informed prompt and physics-guided alignment that leverage the underlying common patterns in mobility changes after different disasters to guide the generation process. To achieve cross-city generalization, we introduce a meta-learning framework that extracts universal patterns across multiple cities through shared parameters and captures city-specific features via private parameters. Extensive experiments across multiple cities and disaster scenarios demonstrate that our method significantly outperforms state-of-the-art baselines, achieving an average performance improvement exceeding 13%.

Figures (6)

• Figure 1: The overview architecture of UniDisMob, which consists of five modules: (1) Trajectory Encoder, (2) Noise Scheduler, (3) Physics-informed Prompt, (4) Noise Predictor, (5) Physics-guided Alignment.
• Figure 2: The meta-learning framework, which extracts common patterns across multiple cities through shared modules, while capturing city-specific characteristics via private modules.
• Figure 3: Fitting performance of the spatiotemporal decay model across different cities.
• Figure 4: Zero-shot performance on Los Angeles datasets.
• Figure 5: Zero-shot performance on Worcester datasets.