Finding Extreme Event Analogs Through Information Retrieval: A Methodological Assessment Using the Emergency Events Database

Analog-based methods identify historical climate events that closely resemble a recent target event. They are widely used in climate science to investigate the role of anthropogenic climate change in the severity of current extreme weather events, as well as for impact and risk assessment and adaptation strategies.

Despite their intuitive appeal, the definition of "similarity" between extreme events remains an open and often implicit methodological choice: which variables to consider, how to preprocess them, and how to aggregate them into a distance or similarity measure, can substantially affect which analogs are retrieved and, consequently, the conclusions drawn.

Here, we frame the analog search problem within the framework of Information Retrieval (IR), a well-established field in computer science focused on identifying relevant items from large collections in response to a query. This framing enables a rigorous, pipeline-oriented evaluation of different methodological choices, including variable selection, normalization strategies, dimensionality reduction, and similarity metrics, treating each configuration as a retrieval system whose performance can be systematically assessed. We apply this framework to the Emergency Events Database (EM-DAT), which provides a structured record of historical extreme events. We use EM-DAT both to define the query events and to provide a ground-truth signal for evaluating retrieval quality, i.e., whether retrieved analogs correspond to a disaster of the same type as the query event.

Our results suggest that pipeline configuration has a non-trivial effect on analog quality, with preprocessing choices and variable selection emerging as particularly critical. This work aims to provide a reproducible, quantitative framework for comparing analog-search strategies in the context of extreme event analysis, with potential applications in impact attribution, climate risk assessment and adaptation strategies.

Acknowledgements

This research has been carried out with funding from Ministero dell'Università e della Ricerca under the call Fondo Italiano per la Scienza 2022-2023 (FIS-2) for the project "Mediterranean Extreme Events and Tipping elements in a changing climate on multiple spatiotemporal scales", grant number FIS-2023-00159, CUP: D53C24005450001.