ClimaEmpact: Domain-Aligned Small Language Models and Datasets for Extreme Weather Analytics

TL;DR

The paper tackles the scarcity of fine-grained, domain-specific data for extreme weather analytics by introducing ClimaEmpact, a framework that combines the ExtremeWeatherNews dataset with ExtremeWeatherReasoning-Aware Alignment (EWRA). EWRA transfers structured reasoning from large language models to small language models via a two-stage curriculum and synthetic alignment data (ExtremeAlign), enabling robust domain-specific inferences across vulnerability/impact/emergency assessment, topic labeling, and emotion analysis. Empirical results show EWRA outperforms standard supervised fine-tuning and other reasoning-based baselines, particularly on larger 3B-instruction models, with improved ranking alignment and explanation quality. The work releases valuable resources (ExtremeWeatherNews and ExtremeAlign) and demonstrates practical potential for near-real-time extreme weather analytics, with implications for disaster response and policy planning.

Abstract

Accurate assessments of extreme weather events are vital for research and policy, yet localized and granular data remain scarce in many parts of the world. This data gap limits our ability to analyze potential outcomes and implications of extreme weather events, hindering effective decision-making. Large Language Models (LLMs) can process vast amounts of unstructured text data, extract meaningful insights, and generate detailed assessments by synthesizing information from multiple sources. Furthermore, LLMs can seamlessly transfer their general language understanding to smaller models, enabling these models to retain key knowledge while being fine-tuned for specific tasks. In this paper, we propose Extreme Weather Reasoning-Aware Alignment (EWRA), a method that enhances small language models (SLMs) by incorporating structured reasoning paths derived from LLMs, and ExtremeWeatherNews, a large dataset of extreme weather event-related news articles. EWRA and ExtremeWeatherNews together form the overall framework, ClimaEmpact, that focuses on addressing three critical extreme-weather tasks: categorization of tangible vulnerabilities/impacts, topic labeling, and emotion analysis. By aligning SLMs with advanced reasoning strategies on ExtremeWeatherNews (and its derived dataset ExtremeAlign used specifically for SLM alignment), EWRA improves the SLMs' ability to generate well-grounded and domain-specific responses for extreme weather analytics. Our results show that the approach proposed guides SLMs to output domain-aligned responses, surpassing the performance of task-specific models and offering enhanced real-world applicability for extreme weather analytics.

Figures (5)

• Figure 1: An overview of EWRA. The process begins with web scraping news articles to construct the ExtremeWeatherNews, followed by pre-processing. In Step 1, the task is formulated for an LLM. Step 2 introduces the ExtremeAlign (Alignment Data Construction), which structures reasoning-based outputs. Finally, in Step 3, a domain-specific Small Language Model (SLM) is trained using EWRA to improve reasoning alignment for extreme weather analysis.
• Figure S1: One-shot prompt for vulnerability/impact/emergency statement assessment
• Figure S2: One-shot prompt for emotion analysis.
• Figure S3: One-shot prompt for topic/subtopic/keyword labeling.
• Figure S4: ClimaEmpact Online Dashboard for Extreme Weather Analysis