Editing IPCC:AR6/WGII/Chapter-16 (section)

=== 16.2.2 Methods and Data for Impact Attribution Including Recent Advances ===

<div id="h2-6-siblings" class="h2-siblings"></div>

By definition, the counterfactual baseline required for impact attribution cannot be observed. However, it may be approximated by impact model simulations forced by a stationary climate, for example derived by de-trending the observed climate ( [[#Diffenbaugh--2017|Diffenbaugh et al., 2017]] ; [[#Mengel--2021|Mengel et al., 2021]] ), while other relevant drivers (e.g., land use changes or application of pesticides) of changes in the system of interest (e.g., a bird population) evolve according to historical conditions. To attribute to anthropogenic climate forcing, the anthropogenic trends in climate are estimated from a range of different climate models and subtracted from the observed climate (e.g., [[#Abatzoglou--2016|Abatzoglou and Williams, 2016]] , for changes in the extent of forest fires or [[#Diffenbaugh--2019|Diffenbaugh and Burke, 2019]] , for effects on economic inequality) or the ‘no anthropogenic climate forcing’ baseline is directly derived from a large ensemble of climate model simulations not accounting for anthropogenic forcings (e.g., Kirchmeier-Young et al.., 2019b, for the extent of forest fires). In any case, it has to be demonstrated that the applied impact models are able to explain the observed changes in natural, human or managed systems by, for example, reproducing the observations when forced by observed changes in climate-related systems and other relevant drivers.

In a situation where an influence of other direct human drivers can be excluded (e.g., by restriction to remote areas not affected by direct human interventions), the ‘no climate-change’ baseline can also be approximated by data from early observational periods with no or minor levels of climate change. In particular, the contribution of climate change to the observed changes in ecosystems is often also determined by a ‘multiple lines of evidence’ approach where the baseline is not formally quantified but the observed changes are identified as a signal of climate change compared with a no-climate-change situation based on process understanding from, for example, palaeo data and laboratory or field experiments in combination with individual long-term observational records and the large-scale spatial or temporal pattern of observed changes that can hardly be explained by alternative drivers ( [[#Parmesan--2013|Parmesan et al., 2013]] ).

To date, explicit accounting for direct human or natural influences is often hampered by an incomplete understanding of the processes and limited observational data. There are, however, first studies demonstrating the potential of detailed process-based or empirical modelling that explicitly accounts for known variations in direct human or natural drivers and separate their effects from the ones induced by changes in the climate-related systems. Examples are Butler et al. (2018) for the separation of growing season adjustments from within growing season climate effects on US crop yields; [[#Wang--2019|Wang and Hijmans (2019)]] separating effects of shifts in land use from climate effects; [[#Jongman--2015|Jongman et al. (2015)]] ; [[#Formetta--2019|Formetta and Feyen (2019)]] and Tanoue et al. (2016) for the separation of changes in exposure and vulnerability from climate effects on river floods; [[#Kirchmeier-Young--2019b|Kirchmeier-Young et al. (2019b)]] for wildfire attribution; and Venter et al. (2018) for the attribution of ecosystem structural changes to climate change versus other disturbances.

There also has been significant progress in the compilation of fragmented and distributed observational data (e.g., [[#Cohen--2018|Cohen et al., 2018]] , for phenological ecosystem changes; [[#Poloczanska--2013|Poloczanska et al., 2013]] , for distributional shifts in marine ecosystems; [[#Andela--2019|Andela et al., 2019]] , with the new global fire atlas including information about individual fire size, duration, speed and direction) as well as in the regional disaggregation (e.g., [[#Ray--2015|Ray et al., 2015]] , for crop yields ) allowing for the identification of an overall picture of the impacts of progressing climate change. Given the ever-increasing body of literature on observed changes in natural, human and managed systems, there also is a first machine learning approach for an automated identification of relevant literature that could complement or support expert assessments as the one provided here ( [[#Callaghan--2021|Callaghan et al., 2021]] ).

<div id="16.2.3" class="h2-container"></div>

<span id="observed-impacts-of-changes-in-climate-related-systems"></span>