Editing IPCC:AR6/WGI/Chapter-1 (section)

=== 1.4.3 Sources of Uncertainty in Climate Simulations ===

<div id="h2-22-siblings" class="h2-siblings"></div>

When evaluating and analysing simulations of the physical climate system, several different sources of uncertainty need to be considered (e.g., [[#Hawkins--2009|Hawkins and Sutton, 2009]] ; [[#Lehner--2020|Lehner et al., 2020]] ). Broadly, these sources are: uncertainties in radiative forcings (both those observed in the past and those projected for the future); uncertainty in the climate response to particular radiative forcings; internal and natural variations of the climate system (which may be somewhat predictable); and interactions among these sources of uncertainty.

Ensembles of climate simulations ( [[#1.5.4.2|Section 1.5.4.2]] ), such as those produced as part of the sixth phase of the Coupled Model Intercomparison Project (CMIP6), can be used to explore these different sources of uncertainty and estimate their magnitude. Relevant experiments with climate models include both historical simulations constrained by past radiative forcings, and projections of future climate which are constrained by specified drivers, such as GHG concentrations, emissions, or radiative forcings. (The term ‘prediction’ is usually reserved for estimates of the future climate state which are also constrained by the observed initial conditions of the climate system, analogous to a weather forecast.)

<div id="1.4.3.1" class="h3-container"></div>

<span id="sources-of-uncertainty"></span>
==== 1.4.3.1 Sources of Uncertainty ====

<div id="h3-15-siblings" class="h3-siblings"></div>

<div id="1.4.3.1.1" class="h4-container"></div>

<span id="radiative-forc-ing-uncertainty"></span>
===== 1.4.3.1.1 Radiative forcing uncertainty =====

<div id="h4-1-siblings" class="h4-siblings"></div>

Future radiative forcing is uncertain due to as-yet-unknown societal choices that will determine future anthropogenic emissions; this is considered ‘scenario uncertainty’. The RCP and SSP scenarios, which form the basis for climate projections assessed in this Report, are designed to span a plausible range of future pathways ( [[#1.6|Section 1.6]] ) and can be used to estimate the magnitude of scenario uncertainty, but the real world may also differ from any one of these example pathways.

Uncertainties also exist regarding past emissions and radiative forcings. These are especially important for simulations of paleoclimate time periods, such as the Pliocene, Last Glacial Maximum or the last millennium, but are also relevant for the CMIP historical simulations of the instrumental period since 1850. In particular, historical radiative forcings due to anthropogenic and natural aerosols are less well constrained by observations than the GHG radiative forcings. There is also uncertainty in the size of large volcanic eruptions (and in the location for some that occurred before around 1850), and the amplitude of changes in solar activity, before satellite observations. The role of historical radiative forcing uncertainty was considered previously ( [[#Knutti--2002|Knutti et al., 2002]] ; [[#Forster--2013|Forster et al., 2013]] ) but, since AR5, specific simulations have been performed to examine this issue, particularly for the effects of uncertainty in anthropogenic aerosol radiative forcing (e.g., [[#Jiménez-de-la-Cuesta--2019|Jiménez-de-la-Cuesta and Mauritsen, 2019]] ; [[#Dittus--2020|Dittus et al., 2020]] ).

<div id="1.4.3.1.2" class="h4-container"></div>

<span id="climate-respo-nse-uncertainty"></span>

===== 1.4.3.1.2 Climate response uncertainty =====

<div id="h4-2-siblings" class="h4-siblings"></div>

Under any particular scenario ( [[#1.6.1|Section 1.6.1]] ), there is uncertainty in how the climate will respond to the specified emissions or radiative forcing combinations. A range of climate models is often used to estimate the range of uncertainty in our understanding of the key physical processes and to define the ‘model response uncertainty’ (Sections [[#1.5.4|1.5.4]] and [[IPCC:Wg1:Chapter:Chapter-4#4.2.5|4.2.5]] ). However, this range does not necessarily represent the full ‘climate response uncertainty ''’'' in how the climate may respond to a particular radiative forcing or emissions scenario. This is because, for example, the climate models used in CMIP experiments have structural uncertainties not explored in a typical multi-model exercise (e.g., [[#Murphy--2004|Murphy et al., 2004]] ) and are not entirely independent of each other ( [[#1.5.4.8|Section 1.5.4.8]] ; [[#Masson--2011|Masson and Knutti, 2011]] ; [[#Abramowitz--2019|Abramowitz et al., 2019]] ); there are small spatial-scale features which cannot be resolved; and long time-scale processes or tipping points are not fully represented. [[#1.4.4|Section 1.4.4]] discusses how some of these issues can still be considered in a risk assessment context. For some metrics, such as equilibrium climate sensitivity (ECS), the CMIP6 model range is found to be broader than the ''very likely'' range assessed by combining multiple lines of evidence (Sections 4.3.4 and 7.5.6).

<div id="1.4.3.1.3" class="h4-container"></div>

<span id="natural-and-internal-cli-mate-variations"></span>

===== 1.4.3.1.3 Natural and internal climate variations =====

<div id="h4-3-siblings" class="h4-siblings"></div>

Even without any anthropogenic radiative forcing, there would still be uncertainty in projecting future climate because of unpredictable natural factors such as variations in solar activity and volcanic eruptions. For projections of future climate, such as those presented in Chapter 4, the uncertainty in these factors is not normally considered. However, the potential effects on the climate of large volcanic eruptions (Cross-Chapter Box 4.1; [[#Zanchettin--2016|Zanchettin et al., 2016]] ; [[#Bethke--2017|Bethke et al., 2017]] ) and large solar variations ( [[#Feulner--2010|Feulner and Rahmstorf, 2010]] ; [[#Maycock--2015|Maycock et al., 2015]] ) are studied. On longer time scales, orbital effects and plate tectonics also play a role.

Further, even in the absence of any anthropogenic or natural changes in radiative forcing, Earth’s climate fluctuates on time scales from days to decades or longer. These ‘internal’ variations, such as those associated with modes of variability (e.g., ENSO, Pacific Decadal Variability (PDV), or Atlantic Multi-decadal Variability (AMV); Annex IV) are unpredictable on time scales longer than a few years ahead and are a source of uncertainty for understanding how the climate might become in a particular decade, especially regionally. The increased use of ‘large ensembles’ of complex climate model simulations to sample this component of uncertainty is discussed above in [[#1.4.2.1|Section 1.4.2.1]] and further in Chapter 4.

<div id="1.4.3.1.4" class="h4-container"></div>

<span id="interactions-between-variability-and-rad-iative-forcings"></span>

===== 1.4.3.1.4 Interactions between variability and radiative forcings =====

<div id="h4-4-siblings" class="h4-siblings"></div>

It is plausible that there are interactions between radiative forcings and climate variations, such as influences on the phasing or amplitude of internal or natural climate variability ( [[#Zanchettin--2017|Zanchettin, 2017]] ). For example, the timing of volcanic eruptions may influence Atlantic Multi-decadal Variability (e.g., [[#Otterå--2010|Otterå et al., 2010]] ; [[#Birkel--2018|Birkel et al., 2018]] ) or ENSO (e.g., [[#Maher--2015|Maher et al., 2015]] ; [[#Khodri--2017|Khodri et al., 2017]] ; [[#Zuo--2018|Zuo et al., 2018]] ), and anthropogenic aerosols may influence decadal modes of variability in the Pacific (e.g., [[#Smith--2016|Smith et al., 2016]] ). In addition, melting of glaciers and ice caps due to anthropogenic influences has been speculated to increase volcanic activity (e.g., a specific example for Iceland is discussed in [[#Swindles--2018|Swindles et al., 2018]] ).

<div id="1.4.3.2" class="h3-container"></div>

<span id="uncertainty-quantification"></span>

==== 1.4.3.2 Uncertainty Quantification ====

<div id="h3-16-siblings" class="h3-siblings"></div>

Not all of these listed sources of uncertainty are of the same type. For example, internal climate variations are an intrinsic uncertainty that can be estimated probabilistically, and could be more precisely quantified, but cannot usually be reduced. However, advances in decadal prediction offer the prospect of narrowing uncertainties in the trajectory of the climate for a few years ahead ( [[IPCC:Wg1:Chapter:Chapter-4#4.2.3|Section 4.2.3]] ; e.g., [[#Meehl--2014|Meehl et al., 2014]] ; [[#Yeager--2017|Yeager and Robson, 2017]] ).

Other sources of uncertainty, such as model response uncertainty, can in principle be reduced, but are not amenable to a frequency-based interpretation of probability, and Bayesian methods to quantify the uncertainty have been considered instead (e.g., [[#Tebaldi--2004|Tebaldi, 2004]] ; [[#Rougier--2007|Rougier, 2007]] ; [[#Sexton--2012|Sexton et al., 2012]] ). The scenario uncertainty component is distinct from other uncertainties, given that future anthropogenic emissions can be considered as the outcome of a set of societal choices ( [[#1.6.1|Section 1.6.1]] ).

For climate model projections it is possible to approximately quantify the relative amplitude of various sources of uncertainty (e.g., [[#Hawkins--2009|Hawkins and Sutton, 2009]] ; [[#Lehner--2020|Lehner et al., 2020]] ). A range of different climate models are used to estimate the model response uncertainty to a particular emissions pathway, and multiple pathways are used to estimate the scenario uncertainty. The unforced component of internal variability can be estimated from individual ensemble members of the same climate model ( [[#1.5.4.8|Section 1.5.4.8]] ; e.g., [[#Deser--2012|Deser et al., 2012]] ; [[#Maher--2019|Maher et al., 2019]] ).

Figure 1.15 illustrates the relative size of these different uncertainty components using a ‘cascade of uncertainty’ ( [[#Wilby--2010|Wilby and Dessai, 2010]] ), with examples shown for global mean temperature, Northern South American annual temperatures and East Asian summer precipitation changes. For global mean temperature, the role of internal variability is small, and the total uncertainty is dominated by emissions scenario and model response uncertainties. Note that there is considerable overlap between individual simulations for different emissions scenarios, even for the mid-term (2041–2060). For example, the slowest-warming simulation for SSP5-8.5 produces less mid-term warming than the fastest-warming simulation for SSP1-1.9. For the long term, emissions scenario uncertainty becomes dominant.

<div id="_idContainer047" class="_idGenObjectStyleOverride-1"></div>

[[File:5e6738df5d3cb730b505b1733656e44f IPCC_AR6_WGI_Figure_1_15.png]]

'''Figure 1.15 |''' '''The ‘cascade of uncertainties’ in CMIP6 projections.''' Changes in: GSAT '''(left)''' ; Northern South America temperature '''(middle)''' ; and East Asia summer (June–July–August, JJA) precipitation '''(right)''' . These are shown for two time periods: 2041–2060 '''(top)''' and 2081–2100 '''(bottom)''' . The SSP–radiative forcing combination is indicated at the top of each cascade at the value of the multi-model mean for each scenario. This branches downwards to show the ensemble mean for each model, and further branches into the individual ensemble members, although often only a single member is available. These diagrams highlight the relative importance of different sources of uncertainty in climate projections, which varies for different time periods, regions and climate variables. See ( [[#1.4.5|Section 1.4.5]] for the definition of the regions used. Further details on data sources and processing are available in the chapter data table (Table 1.SM.1).

Therelative uncertainty due to internal variability and model uncertainty increases for smaller spatial scales. In the regional example shown in Figure 1.15 for changes in temperature, the same scenario and model combination has produced two simulations which differ by 1°C in their projected 2081–2100 averages due solely to internal climate variability. For regional precipitation changes, emissions scenario uncertainty is often small relative to model response uncertainty. In the example shown in Figure 1.15, the SSPs overlap considerably, but SSP1-1.9 shows the largest precipitation change in the near term, even though global mean temperature warms the least; this is due to differences between regional aerosol emissions projected in this and other scenarios ( [[#Wilcox--2020|Wilcox et al., 2020]] ). These cascades of uncertainty would branch out further if applying the projections to derive estimates of changes in hazard (e.g., [[#Wilby--2010|Wilby and Dessai, 2010]] ; [[#Halsnæs--2018|Halsnæs and Kaspersen, 2018]] ; [[#Hattermann--2018|Hattermann et al., 2018]] ).

<div id="1.4.4" class="h2-container"></div>

<span id="considering-an-uncertain-future"></span>