Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Empowering LLM Agents with Geospatial Awareness: Toward Grounded Reasoning for Wildfire Response

Yiheng Chen, Lingyao Li, Zihui Ma, Qikai Hu, Yilun Zhu, Min Deng, Runlong Yu

TL;DR

This work confronts the limitation of traditional LLMs in disaster response by introducing a Geospatial Awareness Layer (GAL) that grounds LLM agents in structured geospatial data via a PostGIS–raster store. GAL converts raw spatial signals into a compact, unit-annotated perception script and augments reasoning with historical analogs through retrieval-augmented generation, all guided by rubric-based CoT prompts to produce auditable, actionable resource decisions. Across fourteen 2020 California wildfires and multiple LLMs, geospatial grounding improves interpretability, stability, and decision quality for daily personnel and budget forecasts, with notable robustness for complex or fragmented fires. The framework generalizes beyond wildfires to other hazards (floods, hurricanes, earthquakes) and provides a reproducible benchmark for geospatially grounded reasoning in disaster response, emphasizing stability, transparency, and operational relevance.

Abstract

Effective disaster response is essential for safeguarding lives and property. Existing statistical approaches often lack semantic context, generalize poorly across events, and offer limited interpretability. While Large language models (LLMs) provide few-shot generalization, they remain text-bound and blind to geography. To bridge this gap, we introduce a Geospatial Awareness Layer (GAL) that grounds LLM agents in structured earth data. Starting from raw wildfire detections, GAL automatically retrieves and integrates infrastructure, demographic, terrain, and weather information from external geodatabases, assembling them into a concise, unit-annotated perception script. This enriched context enables agents to produce evidence-based resource-allocation recommendations (e.g., personnel assignments, budget allocations), further reinforced by historical analogs and daily change signals for incremental updates. We evaluate the framework in real wildfire scenarios across multiple LLM models, showing that geospatially grounded agents can outperform baselines. The proposed framework can generalize to other hazards such as floods and hurricanes.

Empowering LLM Agents with Geospatial Awareness: Toward Grounded Reasoning for Wildfire Response

TL;DR

This work confronts the limitation of traditional LLMs in disaster response by introducing a Geospatial Awareness Layer (GAL) that grounds LLM agents in structured geospatial data via a PostGIS–raster store. GAL converts raw spatial signals into a compact, unit-annotated perception script and augments reasoning with historical analogs through retrieval-augmented generation, all guided by rubric-based CoT prompts to produce auditable, actionable resource decisions. Across fourteen 2020 California wildfires and multiple LLMs, geospatial grounding improves interpretability, stability, and decision quality for daily personnel and budget forecasts, with notable robustness for complex or fragmented fires. The framework generalizes beyond wildfires to other hazards (floods, hurricanes, earthquakes) and provides a reproducible benchmark for geospatially grounded reasoning in disaster response, emphasizing stability, transparency, and operational relevance.

Abstract

Effective disaster response is essential for safeguarding lives and property. Existing statistical approaches often lack semantic context, generalize poorly across events, and offer limited interpretability. While Large language models (LLMs) provide few-shot generalization, they remain text-bound and blind to geography. To bridge this gap, we introduce a Geospatial Awareness Layer (GAL) that grounds LLM agents in structured earth data. Starting from raw wildfire detections, GAL automatically retrieves and integrates infrastructure, demographic, terrain, and weather information from external geodatabases, assembling them into a concise, unit-annotated perception script. This enriched context enables agents to produce evidence-based resource-allocation recommendations (e.g., personnel assignments, budget allocations), further reinforced by historical analogs and daily change signals for incremental updates. We evaluate the framework in real wildfire scenarios across multiple LLM models, showing that geospatially grounded agents can outperform baselines. The proposed framework can generalize to other hazards such as floods and hurricanes.
Paper Structure (31 sections, 1 equation, 6 figures, 3 tables)

This paper contains 31 sections, 1 equation, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Traditional Wildfire Modeling
  4. LLMs for Disaster Response
  5. Methodology
  6. Framework Design
  7. Data Preparation
  8. Geospatial Awareness Layer (GAL)
  9. Event detection and normalization.
  10. Geospatial information retrieval.
  11. Feature consolidation.
  12. GAL-based LLM Reasoning
  13. Perception script.
  14. Analogs scaffold.
  15. Prompted reasoning with rubric-guided CoT.
  16. ...and 16 more sections

Figures (6)

  • Figure 1: An illustration of the research framework.
  • Figure 2: Predictive accuracy across five 2020 California wildfires and hotspot distributions colored by FRP.
  • Figure 3: Ablation of the Geospatial Awareness Layer on daily personnel and daily cost forecasting.
  • Figure 4: Ablation of the RAG module.
  • Figure 5: Feature importance analysis of LLM reasoning across four models.
  • ...and 1 more figures