Improving extreme temperature definitions until they are wrong

"Doesn't this look a bit strange?" 

It began with an innocent question during one of our Master's colloquia. And it could have ended there. "We were just following an approach from the literature". And who could argue against following the literature?

But it bugged me. During a long train ride, I began to think about the issue again. 10 hours and many papers later, I was only more confused: was it really that obvious, and why had no one picked up on it before? But sometimes the most obvious things are the most wicked, and after a few conversations with knowledgeable colleagues, I was sure we were in for an unexpected surprise. 

A commonly used approach to defining heat extremes is as exceedances of percentile-based thresholds that follow the seasonal cycle. Such relative extremes are then expected to be evenly distributed throughout the year. For example, over the 30-year period 1961-1990, we expect three (or 10%) of January 1s to exceed a 90th percentile threshold defined for the same period - and the same for all other days of the year. In a recent study, we show that there are many cases where this does not hold, not even close (Brunner and Voigt 2024).

Here, we tell the story of how this blunder spread in the literature out of the desire to improve extreme thresholds. We show that seemingly innocent changes can sometimes have unintended consequences and that taking the time to check the obvious can help avoid mistakes in science. 

 

Brunner L. and Voigt A. (2024): Pitfalls in diagnosing temperature extremes, Nature Communications, https://doi.org/10.1038/s41467-024-46349-x