Abstract
<p>As climate change increases the frequency and severity of extreme weather events (EWEs), people worldwide are beginning to experience these impacts firsthand. While a growing body of research has explored how this direct experience of EWEs affects public perceptions, attitudes, and behaviors, findings remain inconclusive. A central challenge in this literature is the large researcher degrees of freedom in operationalizing EWEs. This inconsistency partly stems from a lack of theoretical and empirical clarity regarding what the public actually perceives as an EWE. Previous work has shown a low correspondence between subjective perceptions and objective meteorological data, highlighting a disconnect between scientific definitions and the experience of laypeople. To address this gap, we investigate how average citizens conceptualize EWEs using a high-quality panel survey conducted in Switzerland. We field an open-ended survey question with experimental variation in its wording and use large language models (LLMs) to classify the content of responses across multiple dimensions – including the type of event, its regional focus, and its relation to climate change. Validating the classification against human coding and benchmarking it against fine-grained objective weather data, we find that respondents draw on a wide range of considerations, that question wording – especially a local cue – strongly shapes responses while explicit climate-change priming does not, and that open-ended responses correspond at least as closely to objective weather data as a targeted closed-ended item. We discuss the promise and limitations of LLM-based measurement for research on climate and extreme weather perceptions.</p>