When rainfall ‘misbehaves’: Drawing conclusions from attribution analyses for communities at risk.

Extreme rainfall events continue to cause severe flooding and landslides across regions with complex topography and string climate variability, particularly in the Global South. Using recent extreme rainfall and flooding events in Australia, Colombia, Venezuela, Mexico, Pakistan, Sri Lanka, Malaysia  and Indonesia in 2025, we illustrate the roles of climate variability, human-induced climate change, d socio-economic and environmental drivers in shaping meteorological extremes and their impacts. These events are not meteorologically rare in today's climate, yet they produce disproportionate impacts due to high exposure, land use changes and social vulnerability.

Attribution analyses often reveal substantial disagreement among observational datasets, including satellite-based products, reanalyses and station records- in the magnitude, when highly localised heavy rainfall events are smoothed over in the gridded datasets, and in the direction of rainfall trends. Climate models also exhibit large inter-model spread and have limited skill in representing localised rainfall processes, complex topography and interactions with modes of climate variability such as the El Niño Southern Oscillation and the Indian Ocean Dipole. This is illustrated in the case of the extreme rainfall event in Mexico, where discrepancies among observational datasets, particularly prior to the satellite era due to sparse in-situ observations, combined with the influence of natural variability and mixed model trends, limit the ability to confidently detect or attribute long-term changes. In some instances, observational evidence indicates that short-duration rainfall extremes are becoming less likely, contradicting the physical expectation that a warmer atmosphere can hold more moisture. This apparent contradiction is explained by the timing of events within the seasonal cycle and the geographical extent over which trends are assessed. For example, the extreme rainfall event in New South Wales, Australia occurred in a region that represents both a geographical and climatic transition zone. Areas to the south exhibit a robust drying trend during the cooler months, while regions to the north show mixed rainfall trends. In addition, the event took place in May, a period of seasonal transition when shifts in atmospheric circulation become particularly important. Climate models struggle to represent these transitional regimes, contributing to discrepancies between observed and modelled trends in short-duration rainfall extremes. Consequently, while observational evidence suggests intensification of short-duration rainfall extremes in several regions, models often fail to reproduce these signals consistently, limiting confidence in quantitative attribution.

Overall, our findings show that uncertainty in extreme rainfall attribution is often shaped by limitations in observational coverage and model representation that challenges a straightforward attribution. This underscores the broader need to improve and maintain ground based observational networks, alongside the development of higher resolution models and improved representation of key physical processes in both operational forecasts and climate models. At the same time, robust assessment of climate change impacts requires integrating multiple lines of evidence, including physical understanding of atmospheric processes, regional meteorological context, and existing literature on rainfall trends, in order to properly contextualise results and avoid both overstating and understating the role of climate change in shaping extreme rainfall and associated risks.