On the importance of tail assumptions in climate extreme event attribution
Mengran Li, Daniela Castro-Camilo · 2025
Read On the importance of tail assumptions in climate extreme event attribution by Mengran Li, Daniela Castro-Camilo (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-14019.txt.
Introduction
Mengran Li and Daniela Castro-Camilo warn that the math chosen for the far tail of weather distributions can swing climate attribution verdicts. Their paper appeared in two thousand twenty-five. Attribution science asks how much human influence raised the odds of an extreme event. Extremes live in the tail, where data are thin and models differ.
Working inside a causal counterfactual frame, they compare three models of joint extremes. One locks the tail dependence in place, one bends flexibly, and one blends between regimes. Simulations with planted misspecification show the choice can mislead badly. Real cases cover European winter peaks and United States rainfall. Tail assumptions, they conclude, are not a technical footnote. Careful model choice and checking are essential before blaming or clearing the climate.
Catalog record
| CREATOR | Mengran Li, Daniela Castro-Camilo |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 67 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-14019
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
On the importance of tail assumptions in climate extreme event attribution Mengran Li, Daniela Castro-Camilo Abstract Extreme weather events are becoming more frequent and intense, posing serious threats to human life, biodiversity, and ecosystems. A key objective of extreme event attribution (EEA) is to assess whether and to what extent anthropogenic climate change influences such events. Central to EEA is the accurate statistical characterization of atmospheric extremes, which are inherently multivariate or spatial due to their measurement over high-dimensional grids. Within the counterfactual causal inference framework of Pearl, we evaluate how tail assumptions affect attribution conclusions by comparing three multivariate modeling approaches for estimating causation metrics. These include: (i) the multivariate generalized Pareto distribution, which imposes an invariant tail dependence structure; (ii) the factor copula model of Castro-Camilo and Huser, 2020 , which offers flexible subasymptotic behavior; and (iii) the model of Huser and Wadsworth, 2019 , which smoothly transitions between different forms of extremal dependence. We assess the implications of these modeling choices in both simulated scenarios (under varying forms of model misspecification) and real data applications, using weekly winter maxima over Europe from the Météo-France CNRM model and daily precipitation from the ACCESS-CM2 model over the U.S. Our findings highlight that tail assumptions critically shape causality metrics in EEA. Misspecification of the extremal dependence structure can lead to substantially different and potentially misleading attribution conclusions, underscoring the need for careful model selection and evaluation when quantifying the influence of climate change on extreme events. K eywords Extreme value theory \cdot Extreme event attribution \cdot Multivariate generalized Pareto distribution \cdot Subasymptotic extreme value models 1 Introduction In environmental sciences, extreme events are typically defined as instances where a weather or climate variable exceeds a specified threshold. Statistically, such events correspond to values in the tails of probability distributions. As such, their probability of occurrence is small, but their correct estimation is crucial for adequate risk assessment. In our current changing climate, it is no surprise that extreme events have become a source of increasing concern due to their potential for substantial impacts on human life and the natural environment. In its 2022 Global Risks Report, the World Economic Forum identified extreme weather events, climate inaction, and biodiversity loss as among the most pressing global risks, noting that over half of global GDP depends on nature and urging immediate action to halt nature loss through sustainable practices ( World Economic Forum, 2022 ) . These concerns are underscored by the increasing intensity and frequency of climate-related extreme events ( McPhillips et al., 2018 ) . The IPCC’s Sixth Assessment Report further affirms that human-induced greenhouse gas emissions have contributed to a rise in the frequency and/or intensity of certain weather and climate extremes since pre-industrial times ( Masson-Delmotte et al., 2021 ) . In recent years, there has been a growing interest in extreme event attribution (EEA), the science of quantifying the influence of human activities on the probability of observed extreme weather or climate events ( van Oldenborgh et al., 2021 ) . Put simply, EEA seeks to determine whether human actions have made an event more likely or more severe. In the context of climate change, EEA helps answer questions about the extent to which anthropogenic forcing has altered the probability of particular types of extreme events. A common approach to EEA involves comparing the probability of an extreme event under the factual scenario, representing observed climatic conditions, with that under a counterfactual scenario in which anthropogenic emissions and land-use changes had not occurred ( Pearl, 2009 ) . The counterfactual represents a potential outcome: the state of the world that would have occurred in the absence of the causal factor ( Shadish et al., 2002 ) . Formally, let Y(1) denote the potential outcome under treatment T=1 (e.g., presence of human influence), and Y(0) the potential outcome under treatment T=0 (e.g., absence of human influence). Estimating the causal effect of T on Y lies at the heart of causal inference, and requires a set of identifying assumptions. We adopt the Stable Unit Treatment Value Assumption (SUTVA), which includes (i) consistency : the observed outcome equals the potential outcome under the treatment actually received, and (ii) no interference : one unit’s potential outcomes are unaffected by the treatment status of other units, and treatment is consistently defined across all units. Additionally, we assume: (iii) unconfoundedness , meaning that (Y(0),Y(1)) are independent of T given a set of observed covariates \boldsymbol{X} , and (iv) positivity , which requires that for all values of \boldsymbol{X} , there is a non-zero probability of receiving either treatment ( 0<P(T=1\mid\boldsymbol{X})<1 ). Together, these assumptions allow for valid estimation of causal effects from observational data. In the context of extreme event attribution, let Y=\mathbb{I}(Z>v) denote the binary indicator of an extreme event occurring, where Z is a climatological index and v is a high threshold. The treatment T represents the presence ( T=1 ) or absence ( T=0 ) of anthropogenic forcing. We define the probabilities p_{0}=P(Y(0)=1) and p_{1}=P(Y(1)=1) as the probabilities of the extreme event under the counterfactual and factual scenarios, respectively. Within Pearl’s causal framework, various notions of causation—explored in detail in Section 2 —can be expressed in terms of p_{0} and p_{1} , following the work of Hannart et al., 2016 . It is important to note that these are fundamental quantities, but because we cannot observe repeated realizations of the world, estimates of p_{0} and p_{1} typically rely on numerical simulations from climate models under factual and counterfactual scenarios. Although EEA is a relatively recent area of study, a range of statistical models have been developed to estimate and assess the variability of the probabilities p_{0} and p_{1} . In climatology, non-parametric approaches are commonly employed, modeling the number of threshold exceedances as a binomial count process, where a “success” is recorded when the observed value surpasses a predefined threshold. Risk ratios are then derived by comparing exceedance counts under factual and counterfactual scenarios ( Paciorek et al., 2018 ) . Paciorek et al., 2018 also introduced a formal statistical framework for quantifying sampling uncertainty in risk ratio estimation; however, these methods can become computationally demanding when dealing with rare events defined by high thresholds ( Naveau et al., 2020 ) . Parametric models based on distributional assumptions provide a useful alternative to non-parametric methods. Though they require selecting a specific distribution, these models generally offer better statistical performance and greater robustness to small estimated probabilities. Extreme-value theory (EVT) is a particularly fitting framework for EEA because it focuses on the stochastic behavior of extreme events in the tails of probability distributions, and it offers tools to assess the probability of events more extreme than any previously observed. However, EVT models have not been widely applied to EEA. In a univariate setting, Van Der Wiel et al., 2017 applied the generalized extreme-value (GEV) distribution to annual maxima of 3-day precipitation averages in southern Louisiana. This method estimates the probability of an event with magnitude at least as large as those observed annually, using both observational and model-based data. These estimates are then used in rapid attribution studies, such as for flood-inducing extreme precipitation. Additionally, Gonzalez et al., 2025 introduced a