This work stands apart: nearby in the archive are studies on weather, climate, and machine learning, but there are almost no direct neighbors on attribution. Adjacent studies examine how to distinguish noise from signal in climate data, how to model precipitation, and how ecosystems respond to natural rhythms. This paper approaches from a different angle: it does not build a new method but tests the reliability of a widely used approach to extreme event attribution.
The authors replicate the World Weather Attribution methodology: they fit extreme value distributions (GEV or GPD) to observations, allowing the parameters to depend on a four-year smoothed global mean temperature. They then apply the same procedure to an 1800-year pre-industrial simulation of the CESM climate model, where anthropogenic forcing is absent by construction. Additionally, they compare the fit quality with simpler normal and lognormal distributions, test the justification for adding the temperature dependence using a likelihood ratio test, and assess parameter uncertainty via bootstrap.
In the pre-industrial simulation for Australian heat, the median value of the sensitivity parameter to global temperature was 4.9 °C per degree of warming, even though there is no anthropogenic signal. For Siberia, this parameter ranged from 1 to 4 °C per degree. This means that internal variability, such as El Niño, can create a false appearance of a strong link between local extremes and global warming. In observations for Australia, the parameter estimate of about 2 °C per degree falls within the upper range of model values, indicating a possible bias. For Madagascar, there were too few data: only 14 points, and no significant relationship could be detected.
Unlike neighboring studies that search for a signal in noise or build forecasts, this paper shows that the attribution tool itself can fail due to natural climate variability. It does not refute the link between warming and extreme weather, but it requires considering internal modes of variability as additional factors. This changes the interpretation of many published attributions and underscores the need for a multi-method approach.
The topic could develop toward incorporating indices of internal variability, such as El Niño or the Pacific Decadal Oscillation, into statistical attribution models. Another direction is systematic testing of methodologies on large model ensembles with controlled forcings to separate the human contribution from natural fluctuations.
The work will affect the practice of operational attribution of extreme events and the interpretation of results in climatology, insurance, and adaptation policy.
The next step is to check whether the identified biases persist when using other climate models and when including additional covariates describing regional variability.
For climatology, the question of the time of emergence of the anthropogenic signal in extreme precipitation and droughts, where natural variability is particularly large, remains unresolved. This work is directly related to the problem of separating the forced response from internal fluctuations in short observational records.
Verkhoyansk, where 38 °C was recorded in 2020, is considered one of the cold poles of the Northern Hemisphere: in winter, temperatures there drop below −60 °C.
In Kim Stanley Robinson's novel 'The Ministry for the Future', heat in India becomes a catalyst for global action, but the attribution of a specific event remains a matter of debate, as in this paper.