Picture: Sun comes out from clouds. Credit: Pixabay (Pexels).

The field of extreme event attribution, the analysis of extreme weather and climate events to estimate the relative influence of different forcings caused by climate change only began this century. However, event attribution has grown quickly in recent years and, as more studies have been performed, more lessons have been learnt about how to approach these studies and traps to avoid.

A 2020 study goes through the step-by-step approach to event attribution employed by the World Weather Attribution team. In our recent study, published in Climatic Change, we discuss the choices taken at each step, which may affect the final outcome and its usefulness. Here are just a few of the issues we raise:

In the beginning, a choice on whether to even analyse a particular event has to be made. Studies are usually initiated based on their impacts and the availability of observational data to characterise the event and evaluate a model’s ability to capture it. This in itself leads to biases in the choice of events that will be examined in detail and means the cataloguing of event attribution analyses is challenging.

Extreme events can also be characterised in different ways using meteorological variables or impact-based measures and the choice made here may influence whether an impact model, such as a hydrological model, is required.

Following this, observational and model analysis may then be conducted. Long high-quality observational data is needed for an attribution study so that the trend in the real-world data can be analysed and the links to observed climate change assessed. Climate models must be evaluated against this observational data and those which cannot capture the behaviour of the variable of interest then the model should be removed from analysis.

At the same time, different evaluation techniques with more or less focus on the tail of the distribution where extremes reside, or on the processes that deliver extreme events, may lead to different results. An approach that considers a variety of model setups (e.g. fully-coupled or prescribed ocean conditions) is preferable for gaining greater confidence in the outcomes of the analysis. The attribution analysis itself may then be conducted, but it should account for model biases that are common for extremes.

The communication of the result is critical and different audiences should be catered for. A framework including a full methodological document allows reproducibility of results, but a shorter press release is needed to tailor the message towards journalists and other non-scientific audiences.

Overall, we suggest that event attribution studies can be most useful when designed carefully to account for the methodological choices that can be made along the way. We have just highlighted a few issues that people may run into performing such studies, but more detailed discussion can be found in the study and on the World Weather Attribution website.

Paper: van Oldenborgh, G.J., van der Wiel, K., Kew, S. et al. Pathways and pitfalls in extreme event attribution. Climatic Change 166, 13 (2021). https://doi.org/10.1007/s10584-021-03071-7