Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Save It for the "Hot" Day: An LLM-Empowered Visual Analytics System for Heat Risk Management

Haobo Li, Wong Kam-Kwai, Yan Luo, Juntong Chen, Chengzhong Liu, Yaxuan Zhang, Alexis Kai Hon Lau, Huamin Qu, Dongyu Liu

TL;DR

The paper tackles the challenge of heat risk management by bridging numerical climate models with semantically rich news data through an LLM-empowered visual analytics system called Havior. It introduces a two-part pipeline (data preprocessing and human-in-the-loop risk understanding) and novel visuals such as the thermoglyph and news glyph to fuse quantitative signals with contextual narratives. Through a design-driven, expert-informed process, the authors demonstrate that integrating numeric and textual insights improves the identification, assessment, and mitigation of heat risks, enabling both better situational awareness and actionable decision support. The evaluation combines information extraction accuracy, a case study in Hong Kong during the 2022 heatwave, and expert surveys, showing high usability and perceived impact, with clear guidance on limitations and future directions for broader applicability and multimodal data integration.

Abstract

The escalating frequency and intensity of heat-related climate events, particularly heatwaves, emphasize the pressing need for advanced heat risk management strategies. Current approaches, primarily relying on numerical models, face challenges in spatial-temporal resolution and in capturing the dynamic interplay of environmental, social, and behavioral factors affecting heat risks. This has led to difficulties in translating risk assessments into effective mitigation actions. Recognizing these problems, we introduce a novel approach leveraging the burgeoning capabilities of Large Language Models (LLMs) to extract rich and contextual insights from news reports. We hence propose an LLM-empowered visual analytics system, Havior, that integrates the precise, data-driven insights of numerical models with nuanced news report information. This hybrid approach enables a more comprehensive assessment of heat risks and better identification, assessment, and mitigation of heat-related threats. The system incorporates novel visualization designs, such as "thermoglyph" and news glyph, enhancing intuitive understanding and analysis of heat risks. The integration of LLM-based techniques also enables advanced information retrieval and semantic knowledge extraction that can be guided by experts' analytics needs. Our case studies on two cities that faced significant heatwave events and interviews with five experts have demonstrated the usefulness of our system in providing in-depth and actionable insights for heat risk management.

Save It for the "Hot" Day: An LLM-Empowered Visual Analytics System for Heat Risk Management

TL;DR

The paper tackles the challenge of heat risk management by bridging numerical climate models with semantically rich news data through an LLM-empowered visual analytics system called Havior. It introduces a two-part pipeline (data preprocessing and human-in-the-loop risk understanding) and novel visuals such as the thermoglyph and news glyph to fuse quantitative signals with contextual narratives. Through a design-driven, expert-informed process, the authors demonstrate that integrating numeric and textual insights improves the identification, assessment, and mitigation of heat risks, enabling both better situational awareness and actionable decision support. The evaluation combines information extraction accuracy, a case study in Hong Kong during the 2022 heatwave, and expert surveys, showing high usability and perceived impact, with clear guidance on limitations and future directions for broader applicability and multimodal data integration.

Abstract

The escalating frequency and intensity of heat-related climate events, particularly heatwaves, emphasize the pressing need for advanced heat risk management strategies. Current approaches, primarily relying on numerical models, face challenges in spatial-temporal resolution and in capturing the dynamic interplay of environmental, social, and behavioral factors affecting heat risks. This has led to difficulties in translating risk assessments into effective mitigation actions. Recognizing these problems, we introduce a novel approach leveraging the burgeoning capabilities of Large Language Models (LLMs) to extract rich and contextual insights from news reports. We hence propose an LLM-empowered visual analytics system, Havior, that integrates the precise, data-driven insights of numerical models with nuanced news report information. This hybrid approach enables a more comprehensive assessment of heat risks and better identification, assessment, and mitigation of heat-related threats. The system incorporates novel visualization designs, such as "thermoglyph" and news glyph, enhancing intuitive understanding and analysis of heat risks. The integration of LLM-based techniques also enables advanced information retrieval and semantic knowledge extraction that can be guided by experts' analytics needs. Our case studies on two cities that faced significant heatwave events and interviews with five experts have demonstrated the usefulness of our system in providing in-depth and actionable insights for heat risk management.
Paper Structure (37 sections, 8 figures, 1 table)

This paper contains 37 sections, 8 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Visual Analytics of Climate Risk Data
  4. Risk Analysis of Text Data
  5. Steering Documents Retrieval
  6. Informing the Design
  7. Informing the Design
  8. LLM-Empowered Pipeline
  9. Data Preprocessing
  10. Human-in-the-loop Risk Understanding
  11. Numerical Climate Data Understanding
  12. Topic Understanding & Context Analysis
  13. Heat Risk Management
  14. Visual Design of Havior
  15. Temporal View
  16. ...and 22 more sections

Figures (8)

  • Figure 1: The LLM-empowered pipeline contains two parts: data preprocessing (A) and human-in-the-loop risk understanding (B). The data preprocessing involves extracting structural information using LLM (A1) and calculating climate indices (A2-4). In human-in-the-loop risk understanding (B), heterogeneous understandings are integrated through keywords retrieving (Ba), topic modeling (Bb), and RAG (Bc). The interactive analysis process is supported by six views of $\textit{Havior}$ (B1-6) which fulfill the design requirements.
  • Figure 2: The interface of Havior (Heat Savior). The Meteorological Panel (A) facilitates numerical understanding of meteorology, including temporal trends (A1), temporal distribution (A2), and spatial distribution (A3). The "thermoglyph" of Hong Kong (A4) intuitively shows the city-based pattern and correlation between temperature and percentile. The News Panel (B) supports human-in-the-loop news retrieval and enhancement in their semantic understanding, in terms of topic-based hierarchies (B1) and risk-based semantic proximity (B2) of retrieved news. The News List (B3) provides details of structural information in the retrieved news with supportive visual cues. The Summary Panel (C) enables experts to examine the integration of news and numeric risk model (C1), pose contextual questions (C2), and generate risk management reports.
  • Figure 3: The "thermoglyph" in the city gallery for selecting cities. They employ a metaphorical representation. The pattern of color blocks vividly depicts the relationship between temperature and percentile for each city. The black lines connect the current temperature (dashed) or the hovered temperature (bold) to its corresponding percentile.
  • Figure 4: The news topic view displays the hierarchical topics of the retrieved news articles. Left: the first-level topics. Right: the corresponding second-level topics upon clicking a first-level topic. Furthermore, double-clicking on a specific topic enables the filtering of news related to that topic in the subsequent analysis process.
  • Figure 5: We opt for the coxcomb glyph design (A) for the news glyph. Alternative design: target glyph design (B) and pie glyph design (C).
  • ...and 3 more figures