LLMs as World Models: Data-Driven and Human-Centered Pre-Event Simulation for Disaster Impact Assessment

2025-06-02

AI Summary

To address the need for proactive pre-disaster preparedness against sudden earthquakes, this paper proposes leveraging large language models (LLMs) as interpretable and generalizable world models to build a human-perception-oriented pre-event impact simulation framework. Methodologically, it pioneers the integration of multimodal heterogeneous data—including geospatial, socioeconomic, building attribute, and street-view imagery—combined with retrieval-augmented generation (RAG), in-context learning, and multimodal joint encoding to enable fine-grained prediction of pre-event Modified Mercalli Intensity (MMI) and community-level perceptual impact. Key contributions are: (1) establishing the first LLM-driven paradigm for pre-disaster perception modeling; and (2) introducing a novel simulation pathway that jointly incorporates physical laws and public experiential cognition. Evaluated on the Napa and Ridgecrest earthquake cases, ZIP-level predictions achieve Pearson correlation of 0.88 and RMSE of 0.77; incorporating visual inputs significantly improves accuracy, outperforming conventional numerical models.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Computer Vision: Large Vision Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Efficient simulation is essential for enhancing proactive preparedness for sudden-onset disasters such as earthquakes. Recent advancements in large language models (LLMs) as world models show promise in simulating complex scenarios. This study examines multiple LLMs to proactively estimate perceived earthquake impacts. Leveraging multimodal datasets including geospatial, socioeconomic, building, and street-level imagery data, our framework generates Modified Mercalli Intensity (MMI) predictions at zip code and county scales. Evaluations on the 2014 Napa and 2019 Ridgecrest earthquakes using USGS ''Did You Feel It? (DYFI)'' reports demonstrate significant alignment, as evidenced by a high correlation of 0.88 and a low RMSE of 0.77 as compared to real reports at the zip code level. Techniques such as RAG and ICL can improve simulation performance, while visual inputs notably enhance accuracy compared to structured numerical data alone. These findings show the promise of LLMs in simulating disaster impacts that can help strengthen pre-event planning.

Problem

• Using LLMs to simulate earthquake impacts for disaster preparedness
• Leveraging multimodal data for accurate Modified Mercalli Intensity predictions
• Improving simulation performance with RAG, ICL, and visual inputs

Innovation

• Uses LLMs for disaster impact simulation
• Integrates multimodal geospatial and socioeconomic data
• Employs RAG and ICL to boost accuracy