Editing IPCC:AR6/WGI/Chapter-11 (section)

=== 11.2.3 Attribution of Extremes ===

<div id="h2-21-siblings" class="h2-siblings"></div>

Attribution science concerns the identification of causes for changes in characteristics of the climate system (e.g., trends, single extreme events). A general overview and summary of methods of attribution science is provided in the Cross-Working Group Box 1.1. Trend detection using optimal fingerprinting methods is a well-established field, and has been assessed in AR5 (Chapter 10, [[#Bindoff--2013|Bindoff et al., 2013]] ), and [[IPCC:Wg1:Chapter:Chapter-3#3.2.1|Section 3.2.1]] of this Report. There are specific challenges when applying optimal fingerprinting to the detection and attribution of trends in extremes and on regional scales where the lower signal-to-noise ratio is a challenge. In particular, the method generally requires the data to follow a Normal (Gaussian) distribution, which is often not the case for extremes. However, recent studies showed that extremes can be transformed to a Gaussian distribution, for example, by averaging over space, so that optimal fingerprinting techniques can still be used ( [[#Wen--2013|Wen et al., 2013]] ; [[#Zhang--2013|Zhang et al., 2013]] ; [[#Wan--2019|Wan et al., 2019]] ). Non-stationary extreme value distributions, which allow for the detailed detection and attribution of regional trends in temperature extremes, have also been used (Z. [[#Wang--2017|]] [[#Wang--2017|Wang et al., 2017]] a).

Apart from the detection and attribution of trends in extremes, new approaches have been developed to answer the question of whether, and to what extent, external drivers have altered the probability and intensity of an individual extreme event ( [[#NASEM--2016|NASEM, 2016]] ). In AR5, there was an emerging consensus that the role of external drivers of climate change in specific extreme weather events could be estimated and quantified in principle, but related assessments were still confined to particular case studies, often using a single model, and typically focusing on high-impact events with a clear attributable signal.

However, since AR5, the attribution of extreme weather events has emerged as a growing field of climate research with an increasing body of literature (see series of supplements to the annual State of the Climate report ( [[#Peterson--2012|Peterson et al., 2012]] , 2013a; [[#Herring--2014|Herring et al., 2014]] , 2015, 2016, 2018), including the number of approaches to examining extreme events(described in [[#Easterling--2016|Easterling et al., 2016]] ; [[#Otto--2017|Otto, 2017]] ; [[#Stott--2016|Stott et al., 2016]] )). A commonly used approach – often called the risk-based approach in the literature, and referred to here as the ‘probability-based approach’ – produces statements such as ‘anthropogenic climate change made this event type twice as likely ’ or ‘anthropogenic climate change made this event 15% more intense’. This is done by estimating probability distributions of the index characterizing the event in today’s climate, as well as in a counterfactual climate, and either comparing intensities for a given occurrence probability (e.g., 1-in-100-year event) or probabilities for a given magnitude (see FAQ 11.3). There are a number of different analytical methods encompassed in the probability-based approach, building on observations and statistical analyses (e.g., van Oldenborgh et al., 2012), optimal fingerprint methods ( [[#Sun--2014|Sun et al., 2014]] ), regional climate and weather forecast models (e.g., [[#Schaller--2016|Schaller et al., 2016]] ), global climate models (GCMs) (e.g., [[#Lewis--2013|Lewis and Karoly, 2013]] ), and large ensembles of atmosphere-only GCMs (e.g., [[#Lott--2013|Lott et al., 2013]] ). A key component in any event attribution analysis is the level of conditioning on the state of the climate system. In the least conditional approach, the combined effect of the overall warming and changes in the large-scale atmospheric circulation are considered and often utilize fully coupled climate models ( [[#Sun--2014|Sun et al., 2014]] ). Other more conditional approaches involve prescribing certain aspects of the climate system. These range from prescribing the pattern of the surface ocean change at the time of the event (e.g., [[#Hoerling--2013|Hoerling et al., 2013]] , 2014), often using Atmospheric Model Intercomparison Project (AMIP) style global models, where the choice of sea surface temperature and ice patterns influences the attribution results ( [[#Sparrow--2018|Sparrow et al., 2018]] ), to prescribing the large-scale circulation of the atmosphere and using weather forecasting models or methods (e.g., [[#Pall--2017|Pall et al., 2017]] ; [[#Patricola--2018|Patricola and Wehner, 2018]] ; [[#Wehner--2018a|Wehner et al., 2018a]] ). These highly conditional approaches have also been called ‘storylines’ (Cross-Working Group Box 1.1; [[#Shepherd--2016|Shepherd, 2016]] ) and can be useful when applied to extreme events that are too rare to otherwise analyse, or where the specific atmospheric conditions were central to the impact. These methods are also used to enable the use of very high-resolution simulations in cases were lower-resolution models do not simulate the regional atmospheric dynamics well ( [[#Shepherd--2016|Shepherd, 2016]] ; [[#Shepherd--2018|Shepherd et al., 2018]] ). However, the imposed conditions limit an overall assessment of the anthropogenic influence on an event, as the fixed aspects of the analysis may also have been affected by climate change. For instance, the specified initial conditions in the highly conditional hindcast attribution approach often applied to tropical cyclones (e.g., [[#Takayabu--2015|Takayabu et al., 2015]] ; [[#Patricola--2018|Patricola and Wehner, 2018]] ) permit only a conditional statement about the magnitude of the storm if similar large-scale meteorological patterns could have occurred in a world without climate change, thus precluding any attribution statement about the change in frequency if used in isolation. Combining conditional assessments of changes in the intensity with a multi-model approach does allow for the latter as well ( [[#Shepherd--2016|Shepherd, 2016]] ).

The outcome of event attribution is dependent on the definition of the event ( [[#Leach--2020|Leach et al., 2020]] ), as well as the framing ( [[#Otto--2016|Otto et al., 2016]] ; [[#Christidis--2018|Christidis et al., 2018]] ; [[#Jézéquel--2018|Jézéquel et al., 2018]] ) and uncertainties in observations and modelling. Observational uncertainties arise in estimating the magnitude of an event as well as its rarity ( [[#Angélil--2017|Angélil et al., 2017]] ). Results of attribution studies can also be very sensitive to the choice of climate variables ( [[#Sippel--2014|Sippel and Otto, 2014]] ; [[#Wehner--2016|Wehner et al., 2016]] ). Attribution statements are also dependent on the spatial (Uhe et al., 2016; [[#Cattiaux--2018|Cattiaux and Ribes, 2018]] ; Kirchmeier‐Young et al., 2019) and temporal ( [[#Harrington--2017|Harrington, 2017]] ; [[#Leach--2020|Leach et al., 2020]] ) extent of event definitions, as events of different scales involve different processes (W. [[#Zhang--2020|]] [[#Zhang--2020|]] [[#Zhang--2020|]] [[#Zhang--2020|Zhang et al., 2020]] ) and large-scale averages generally yield higher attributable changes in magnitude or probability due to the smoothing out of noise. In general, confidence in attribution statements for large-scale heat and lengthy extreme precipitation events have higher confidence than shorter and more localized events, such as extreme storms, an aspect also relevant for determining the emergence of signals in extremes or the confidence in projections (see also Cross-Chapter Box Atlas.1).

The reliability of the representation of the event in question in the climate models used in a study is essential ( [[#Angélil--2016|Angélil et al., 2016]] ; [[#Herger--2018|Herger et al., 2018]] ). Extreme events characterized by atmospheric dynamics that stretch the capabilities of current-generation models ( [[IPCC:Wg1:Chapter:Chapter-10#10.3.3.4|Section 10.3.3.4]] ; [[#Shepherd--2014|Shepherd, 2014]] ; [[#Woollings--2018|Woollings et al., 2018]] ) limit the applicability of the probability-based approach of event attribution. The lack of model evaluation, in particular in early event attribution studies, has led to criticism of the emerging field of attribution science as a whole ( [[#Trenberth--2015|Trenberth et al., 2015]] ) and of individual studies ( [[#Angélil--2017|Angélil et al., 2017]] ). In this regard, the storyline approach ( [[#Shepherd--2016|Shepherd, 2016]] ) provides an alternative option that does not depend on the model’s ability to represent the circulation reliably. In addition, several ways of quantifying statistical uncertainty ( [[#Paciorek--2018|Paciorek et al., 2018]] ) and model evaluation ( [[#Lott--2016|Lott and Stott, 2016]] ; [[#Philip--2018b|Philip et al., 2018b]] , 2020) have been employed to evaluate the robustness of event attribution results. For the unconditional probability-based approach, multi-model and multi-approach (e.g., combining observational analyses and model experiments) methods have been used to improve the robustness of event attribution ( [[#Hauser--2017|Hauser et al., 2017]] ; [[#Otto--2018a|Otto et al., 2018a]] ; [[#Philip--2018b|Philip et al., 2018b]] , 2019, 2020; [[#van%20Oldenborgh--2018|van Oldenborgh et al., 2018]] ; [[#Kew--2019|Kew et al., 2019]] ).

In the regional tables provided in [[#11.9|Section 11.9]] , the different lines of evidence from event attribution studies and trend attributions are assessed alongside one another to provide an assessment of the human contribution to observed changes in extremes in all AR6 regions.

<div id="11.2.4" class="h2-container"></div>

<span id="projecting-changes-in-extremes-as-a-function-of-global-warming-levels"></span>