The physical climate system comprises all processes that combine to form weather and climate. The early chapters of this report broadly organize their assessments according to overarching realms: the atmosphere, the biosphere, the cryosphere (surface areas covered by frozen water, such as glaciers and ice sheets), and the ocean. Elsewhere in the report, and in previous IPCC assessments, the land is also used as an integrating realm that includes parts of the biosphere and the cryosphere. These overarching realms have been studied and measured in increasing detail by scientists, institutions and the general public since the 18th century, throughout the era of instrumental observation (Section 1.3). Today, observations include those taken by numerous land surface stations, ocean surface measurements from ships and buoys, underwater instrumentation, satellite and surface-based remote sensing, and in situ atmospheric measurements from aeroplanes and balloons. These instrumental observations are combined with paleoclimate reconstructions and historical documentations to produce a highly detailed picture of the past and present state of the whole climate system, and to allow assessments about rates of change across the different realms (Chapter 2 and Section 1.5).

Figure 1.4 documents that the climate system is undergoing a comprehensive set of changes. It shows a selection of key indicators of change through the instrumental era that are assessed and presented in the subsequent chapters of this report. Annual mean values are shown as stripes, with colours indicating their value. The transitions from one colour to another over time illustrate how conditions are shifting in all components of the climate system. For these particular indicators, the observed changes go beyond the yearly and decadal variability of the climate system. In this Report, this is termed an ‘emergence’ of the climate signal (Section 1.4.2 and FAQ 1.2).

Warming of the climate system is most commonly presented through the observed increase in global mean surface temperature (GMST). Taking a baseline of 1850–1900, GMST change until present (2011–2020) is 1.09°C [0.95 to 1.20] °C (Section 2.3 and Cross-Chapter Box 2.3). This evolving change has been documented in previous assessment reports, with each reporting a higher total global temperature change (Section 1.3 and Cross-Chapter Box 1.2). The total change in global surface air temperature (GSAT) (Section 1.4.1 and Cross-Chapter Box 2.3) attributable to anthropogenic activities is assessed to be consistent with the observed change in GSAT (Section 3.3). ¹

Similarly, atmospheric concentrations of a range of GHGs are increasing. Carbon dioxide (CO₂, shown in Figure 1.4 and Figure 1.5a, found in AR5 and earlier reports to be the current strongest driver of anthropogenic climate change), has increased from 285.5 ± 2.1 ppm in 1850 to 409.9 ± 0.4 ppm in 2019; concentrations of methane (CH₄), and nitrous oxide (N₂O) have increased as well (Sections 2.2 and 5.2, and Annex V). These observed changes are assessed to be in line with known anthropogenic and natural emissions, when accounting for observed and inferred uptake by land, ocean and biosphere respectively (Section 5.2), and are a key source of anthropogenic changes to the global energy balance (or radiative forcing; Sections 2.2 and 7.3).

The hydrological (or water) cycle is also changing and is assessed to be intensifying, through a higher exchange of water between the surface and the atmosphere (Sections 2.3 and 8.3). The resulting regional patterns of changes to precipitation are, however, different from surface temperature change, and interannual variability is larger, as illustrated in Figure 1.4. Annual land area mean precipitation in the Northern Hemisphere temperate regions has increased, while the subtropical dry regions have experienced a decrease in precipitation in recent decades (Section 2.3).

The cryosphere is undergoing rapid changes, with increased melting and loss of frozen water mass in most regions. This includes all frozen parts of the globe, such as terrestrial snow, permafrost, sea ice, glaciers, freshwater ice, solid precipitation, and the ice sheets covering Greenland and Antarctica (Chapter 9; SROCC, IPCC, 2019b). Figure 1.4 illustrates how, globally, glaciers have been increasingly losing mass for the last fifty years. The total glacier mass in the most recent decade (2010–2019) was the lowest since the beginning of the 20th century (Sections 2.3 and 9.5).

The global ocean has warmed unabatedly since at least 1970 (Sections 1.3, 2.3 and 9.2; SROCC, IPCC, 2019b). Figure 1.4 shows how the averaged ocean heat content is steadily increasing, with a total increase of [0.28 to 0.55] yottajoule (YJ; 10²⁴ joule) between 1971 and 2018 (Section 9.2). In response to this ocean warming, as well as to the loss of mass from glaciers and ice sheets, the global mean sea level (GMSL) has risen by 0.20 [0.15 to 0.25] metres between 1900 and 2018. GMSL rise has accelerated since the late 1960s (see Section 9.6).

Overall, the changes in these selected climatic indicators have progressed beyond the range of natural year-to-year variability (Chapters 2, 3, 8 and 9, and Sections 1.2.1.2 and 1.4.2). The indicators presented in Figure 1.4 document a broad set of concurrent and emerging changes across the physical climate system. All indicators shown here, along with many others, are further presented in the coming chapters, together with a rigorous assessment of the supporting scientific literature. Later chapters (Chapters 10, 11, 12 and Atlas) present similar assessments at the regional level, where observed changes do not always align with the global mean picture shown here.

Paleoclimate archives (e.g., ice cores, corals, marine and lake sediments, speleothems, tree rings, borehole temperatures, soils) permit the reconstruction of climatic conditions before the instrumental era. This establishes an essential long-term context for the climate change of the past 150 years and the projected changes in the 21st century and beyond (Chapter 3; IPCC, 2013a; Masson-Delmotte et al., 2013). Figure 1.5 shows reconstructions of three key indicators of climate change over the past 800,000 years (800 kyr)²– atmospheric CO₂ concentrations, global mean surface temperature (GMST) and global mean sea level (GMSL) – comprising at least eight complete glacial–interglacial cycles (EPICA Community Members, 2004; Jouzel et al., 2007), which are largely driven by oscillations in the Earth’s orbit and consequent feedbacks on multi-millennial time scales (Berger, 1978; Laskar et al., 1993). The dominant cycles – recurring approximately every 100 kyr – can be found imprinted in the natural variations of these three key indicators. Before industrialisation, atmospheric CO₂ concentrations varied between 174 ppm and 300 ppm, as measured directly in air trapped in ice at Dome Concordia, Antarctica (Bereiter et al., 2015; Nehrbass-Ahles et al., 2020). Relative to 1850–1900 CE, the reconstructed GMST changed in the range of –6°C to +1°C across these glacial–interglacial cycles (see Chapter 2, Section 2.3.1 for an assessment of different paleo-reference periods). GMSL varied between about –130 m during the coldest glacial maxima and +5 to +25 m during the warmest interglacial periods (Chapter 2; Spratt and Lisiecki, 2016). They represent the amplitudes of natural, global-scale climate variations over the last 800 kyr prior to the influence of human activity. Further climate information from a variety of paleoclimatic archives is assessed in Chapters 2, 5, 7 and 9.

Paleoclimatic information also provides a long-term perspective on rates of change of these three key indicators. In high-resolution reconstructions from polor ice cores, the rate of increase in atmospheric CO₂ observed over 1919–2019 CE is one order of magnitude higher than the fastest CO₂ fluctuations documented during the Last Glacial Maximum and the last deglacial transition (Marcott et al., 2014, see Chapter 2, Section 2.2.3.2.1). Current multi-decadal GMST exhibit a higher rate of increase than over the past 2 kyr (Section 2.3.1.1.2; PAGES 2k Consortium, 2019), and in the 20th century GMSL rise was faster than during any other century over the past 3 kyr (Section 2.3.3.3).

Paleoclimate reconstructions also shed light on the causes of these variations, revealing processes that need to be considered when projecting climate change. The paleorecords show that sustained changes in global mean temperature of a few degrees Celsius are associated with increases in sea level of several tens of metres (Figure 1.5). During two extended warm periods (interglacials) of the last 800 kyr, sea level is estimated to have been at least six metres higher than today (Chapter 2; Dutton et al., 2015). During the last interglacial, sustained warmer temperatures in Greenland preceded the peak of sea level rise (Figure 5.15 in Masson-Delmotte et al., 2013). The paleoclimate record therefore provides substantial evidence directly linking warmer GMST to substantially higher GMSL.

GMST will remain above present-day levels for many centuries even if net CO₂ emissions are reduced to zero, as shown in simulations with coupled climate models (Section 4.7.1; Plattner et al., 2008; Section 12.5.3 in Collins et al., 2013; Zickfeld et al., 2013; MacDougall et al., 2020). Such persistent warm conditions in the atmosphere represent a multi-century commitment to long-term sea level rise, summer sea ice reduction in the Arctic, substantial ice-sheet melting, potential ice-sheet collapse, and many other consequences in all components of the climate system (Section 9.4 and Figure 1.5; Clark et al., 2016; Pfister and Stocker, 2016; H. Fischer et al., 2018).

Paleoclimate records also show centennial- to millennial-scale variations, particularly during the ice ages, which indicate rapid or abrupt changes of the Atlantic Meridional Overturning Circulation (AMOC; Section 9.2.3.1) and the occurrence of a ‘bipolar seesaw’ (opposite-phase surface temperature changes in both hemispheres; Section 2.3.3.4.1; Stocker and Johnsen, 2003; EPICA Community Members, 2006; WAIS Divide Project Members et al., 2015; Lynch-Stieglitz, 2017; Pedro et al., 2018; Weijer et al., 2019). This process suggests that instabilities and irreversible changes could be triggered if critical thresholds are passed (Section 1.4.4.3). Several other processes involving instabilities are identified in climate models (Drijfhout et al., 2015), some of which may now be close to critical thresholds (Section 1.4.4.3; see also Chapters 5, 8 and 9 regarding tipping points; Joughin et al., 2014).

Based on Figure 1.5, the reconstructed, observed and projected ranges of changes in the three key indicators can be compared. By the first decade of the 20th century, atmospheric CO₂ concentrations had already moved outside the reconstructed range of natural variation over the past 800 kyr. On the other hand, GMST and GMSL were higher than today during several interglacials of that period (Sections 2.3.1 and 2.3.3, and Figure 2.34). Projections for the end of the 21st century, however, show that GMST will have moved outside of its natural range within the next few decades, except for the strong mitigation scenarios (Section 1.6). There is a risk that GMSL may potentially leave the reconstructed range of natural variations over the next few millennia (Section 9.6.3.5; Clark et al., 2016; SROCC, IPCC, 2019b). In addition, abrupt changes can not be excluded (Section 1.4.4.3).

An important time period in the assessment of anthropogenic climate change is the last 2 kyr. Since AR5, new global datasets have been produced that aggregate aggregating local and regional paleorecords (PAGES 2k Consortium, 2013, 2017, 2019; McGregor et al., 2015; Tierney et al., 2015; Abram et al., 2016; Hakim et al., 2016; Steiger et al., 2018; Brönnimann et al., 2019b). Before the global warming that began around the mid-19th century (Abram et al., 2016), a slow cooling in the Northern Hemisphere from roughly 1450–1850 CE is consistently recorded in paleoclimate archives (PAGES 2k Consortium, 2013; McGregor et al., 2015). While this cooling, primarily driven by an increased number of volcanic eruptions (Section 3.3.1; PAGES 2k Consortium, 2013; Owens et al., 2017; Brönnimann et al., 2019b), shows regional differences, the subsequent warming over the past 150 years exhibits a global coherence that is unprecedented in the last 2 kyr (Neukom et al., 2019).

The rate, scale and magnitude of anthropogenic changes in the climate system since the mid-20th century suggested the definition of a new geological epoch: the Anthropocene (Crutzen and Stoermer, 2000; Steffen et al., 2007), referring to an era in which human activity is altering major components of the Earth system and leaving measurable imprints that will remain in the permanent geological record (Figure 1.5; IPCC, 2018). These alterations include not only climate change itself, but also chemical and biological changes in the Earth system such as rapid ocean acidification due to uptake of anthropogenic CO₂, massive destruction of tropical forests, a worldwide loss of biodiversity and the sixth mass extinction of species (Hoegh-Guldberg and Bruno, 2010; Ceballos et al., 2017; IPBES, 2019). According to the key messages of the last global assessment of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES, 2019), climate change is a ‘direct driver that is increasingly exacerbating the impact of other drivers on nature and human well-being’, and ‘the adverse impacts of climate change on biodiversity are projected to increase with increasing warming.’

The IPCC First Assessment Report (FAR, IPCC, 1990a) provided the scientific background for the establishment of the UNFCCC (UNFCCC, 1992), which committed parties to negotiate ways to ‘prevent dangerous anthropogenic interference with the climate system’ (the ultimate objective of the UNFCCC). The Second Assessment Report (SAR, IPCC, 1996) informed governments in negotiating the Kyoto Protocol (1997), the first major agreement focusing on mitigation under the UNFCCC. The Third Assessment report (TAR, IPCC, 2001a) highlighted the impacts of climate change and the need for adaptation, and introduced the treatment of new topics such as policy and governance in IPCC reports. The Fourth and Fifth Assessment Reports (AR4, IPCC, 2007a; AR5, IPCC, 2013a) provided the scientific background for the second major agreement under the UNFCCC: the Paris Agreement (2015), which entered into force in 2016.

Parties to the PA commit to the goal of limiting global average temperature increase to ‘well below 2°C above pre-industrial levels, and to pursue efforts to limit the temperature increase to 1.5°C in order to significantly reduce the risks and impacts of climate change’. InAR6, as in many previous IPCC reports, observations and projections of changes in global temperature are expressed relative to 1850–1900 as an approximation for pre-industrial levels (Cross-Chapter Box 1.2).

The PA further addresses mitigation (Article 4) and adaptation to climate change (Article 7), as well as loss and damage (Article 8), through the mechanisms of finance (Article 9), technology development and transfer (Article 10), capacity-building (Article 11) and education (Article 12). To reach its long-term temperature goal, the PA recommends ‘achieving a balance between anthropogenic emissions by sources and removals by sinks of greenhouse gases in the second half of this century’, a state commonly described as ‘net zero’ emissions (Article 4) (Section 1.6 and Box 1.4). Each Party to the PA is required to submit a Nationally Determined Contribution (NDC) and pursue, on a voluntary basis, domestic mitigation measures with the aim of achieving the objectives of its NDC (Article 4).

Numerous studies of the NDCs submitted since adoption of the PA in 2015 (Fawcett et al., 2015; UNFCCC, 2015, 2016; Lomborg, 2016; Rogelj et al., 2016, 2017; Benveniste et al., 2018; Gütschow et al., 2018; UNEP, 2019) conclude that they are insufficient to meet the Paris temperature goal. In the present IPCC Sixth Assessment Cycle, a Special Report on Global Warming of 1.5°C (SR1.5, IPCC, 2018) found, with high agreement, that current NDCs ‘are not in line with pathways that limit warming to 1.5°C by the end of the century.’ The PA includes a ratcheting mechanism designed to increase the ambition of voluntary national pledges over time. Under this mechanism, NDCs will be communicated or updated every five years. Each successive NDC will represent a ‘progression beyond’ the ‘then current’ NDC and reflect the ‘highest possible ambition’ (Article 4). These updates will be informed by a five-yearly periodic review including the Structured Expert Dialogue (SED), as well as a ‘global stocktake’, to assess collective progress toward achieving the PA long-term goals. These processes will rely upon the assessments prepared during the IPCC Sixth Assessment Cycle (e.g., Cross-Chapter Box 1.1; Schleussner et al., 2016b).

Since AR5, the formal dialogue between the scientific and policy communities has been strengthened through a new science– policy interface, the Structured Expert Dialogue (SED). The SED was established by UNFCCC to support the work of its two subsidiary bodies, the Subsidiary Body for Scientific and Technological Advice (SBSTA) and the Subsidiary Body for Implementation (SBI). The first SED aimed to ‘ensure the scientific integrity of the first periodic review’ of the UNFCCC, the 2013–2015 review. The Mandate of the periodic review is to ‘assess the adequacy of the long-term (temperature) goal in light of the ultimate objective of the convention’ and the ‘overall progress made towards achieving the long-term global goal, including a consideration of the implementation of the commitments under the Convention.’

The SED of the first periodic review (2013–2015) provided an important opportunity for face-to-face dialogue between decision makers and experts on review themes, based on ‘the best available scientific knowledge, including the assessment reports of the IPCC.’ That SED was instrumental in informing the long-term global goal of the PA and in providing the scientific argument for the consideration of limiting warming to 1.5°C warming (UNFCCC, 2015; Fischlin, 2017). The SED of the second periodic review, initiated in the second half of 2020, focuses on, among other things, ‘enhancing Parties’ understanding of the long-term globalgoal and the scenarios towards achieving it in the light of the ultimate objective of the Convention’. The second SED provides a formal venue for the scientific and the policy communities to discuss the requirements and benchmarks to achieve the ‘long-term temperature goal’ (LTTG) of 1.5°C and well below 2°C global warming. The discussions also concern the associated timing of net zero emissions targets and the different interpretations of the PA LTTG, including the possibility of overshooting the 1.5° C warming level before returning to it by means of negative emissions (e.g., Section 1.6; Schleussner and Fyson, 2020). The second periodic review is planned to continue until November 2022 and its focus includes the review of the progress made since the first review, while minimising ‘possible overlaps’ and profiting from ‘synergies with the global stocktake’.

Many interactions among environmental problems and development are addressed in the United Nations 2030 Agenda for Sustainable Development and its Sustainable Development Goals. The 2030 Agenda, supported by the finance-oriented Addis Ababa Action Agenda (UN DESA, 2015), calls on nations to ‘take the bold and transformative steps which are urgently needed to shift the world onto a sustainable and resilient path.’ The 2030 Agenda recognizes that ‘climate change is one of the greatest challenges of our time and its adverse impacts undermine the ability of all countries to achieve sustainable development.’ SDG 13 deals explicitly with climate change, establishing several targets for adaptation, awareness-raising and finance. Climate and climate change are also highly relevant to most other SDGs, and UNFCCC is acknowledged as the main forum to negotiate the global response to climate change. For example, both long-lived GHGs (through mitigation decisions), and SLCFs (through air quality), are relevant to SDG 11 (sustainable cities and communities). Chapter 6 assesses the effects of SLCFs on climate and the implications of changing climate for air quality, including opportunities for mitigation relevant to the SDGs (Box 6.2). Also, the UN Conference on Housing and Sustainable Development established a New Urban Agenda (United Nations, 2017) envisaging cities as part of the solutions for sustainable development, climate change adaptation and mitigation.

The Sendai Framework for Disaster Risk Reduction is a non-binding agreement to reduce risks associated with disasters of all scales, frequencies and onset rates caused by natural or human-made hazards, including climate change. The SFDRR outlines targets and priorities for action including ‘understanding disaster risk’, along the dimensions of vulnerability, exposure of persons and assets, and hazard characteristics. Chapter 12 assesses climate information relevant to regional impact and risk assessment, with a focus on climate hazards and other aspects of climate that influence society and ecosystems and makes the link with Working Group II. AR6 adopts a consistent risk- and solution-oriented framing (Cross-Chapter Box 1.3) that calls for a multidisciplinary approach and cross-Working Group coordination in order to ensure integrative discussions of major scientific issues associated with integrative risk management and sustainable solutions (IPCC, 2017).

Efforts to address climate change take place alongside and in the context of other major environmental problems, such as biodiversity loss. IPBES, established in 2012, builds on the IPCC model of a science–policy interface and assessment. The Platform’s objective is to ‘strengthen the science–policy interface for biodiversity and ecosystem services for the conservation and sustainable use of biodiversity, long-term human well-being and sustainable development’ (UNEP, 2012). The SROCC (IPCC, 2019b) and SRCCL (IPCC, 2019a) assessed the relations between changes in biodiversity and in the climate system. The rolling work programme of IPBES up to 2030 will address interlinkages among biodiversity, water, food and health. This assessment will use a nexus approach to examine interlinkages between biodiversity and the above-mentioned issues, including climate change mitigation and adaptation. Furthermore, IPBES and IPCC will directly collaborate on biodiversity and climate change under the rolling work programme.

Addressing climate change alongside other environmental problems, while simultaneously supporting sustainable socio-economic development, requires a holistic approach. Since AR5, there is increasing attention on the need for coordination among previously independent international agendas, and a recognition that climate change, disaster risk, economic development, biodiversity conservation and human well-being are tightly interconnected. The current COVID-19 pandemic provides an example of the need for such interconnection, with its widespread impacts on economy, society and environment (e.g., Shan et al., 2021). Cross-Chapter Box 6.1 assesses the consequences of the COVID-19 lockdowns for emissions of GHGs and SLCFs, and related implications for the climate. Another example of the interconnected nature of these issues is the close link between SLCF emissions, climate change and air quality concerns (Chapter 6). Emissions of halocarbons have previously been successfully regulated under the Montreal Protocol and its Kigali Amendment. This has been achieved in an effort to reduce ozone depletion that has also modulated other anthropogenic climate influence (Estrada et al., 2013; Wu et al., 2013). In the process, emissions of some SLCFs were jointly regulated to reduce environmental and health impacts from air pollution (e.g., Gothenburg Protocol; Reis et al., 2012). Considering the recognized importance of SLCFs in climate change processes, the IPCC decided in May 2019 to approve that the IPCC Task Force on National Greenhouse Gas Inventories produces an IPCC Methodology Report on SLCFs to develop guidance for national SLCF inventories.

The evolving governance context since AR5 challenges the IPCC to provide policymakers and other actors with information relevant for both adaptation to and mitigation of climate change, and for the loss and damage induced.

Responses to climate change are facilitated when leaders, policymakers, resource managers and their constituencies share a basic understanding of the causes, effects, and possible future course of climate change (SR1.5, IPCC, 2018; SRCCL, IPCC, 2019a). Achieving shared understanding is complicated, since scientific knowledge interacts with pre-existing conceptions of weather and climate that have built up in diverse world cultures over centuries, and which are often embedded in strongly held values and beliefs stemming from ethnic or national identities, traditions, religions, and lived relationships to weather, land and sea (Van Asselt and Rotmans, 1996; Rayner and Malone, 1998; Hulme, 2009, 2018; Green et al., 2010; Jasanoff, 2010; Orlove et al., 2010; Nakashima et al., 2012; Shepherd and Sobel, 2020).These diverse, more local understandings can both contrast with and enrich the planetary-scale analyses of global climate science (high confidence).

Political cultures also give rise to variation in how climate science knowledge is interpreted, used and challenged (Leiserowitz, 2006; Oreskes and Conway, 2010; Brulle et al., 2012; Dunlap and Jacques, 2013; Mahony, 2014, 2015; Brulle, 2019). A meta-analysis of 87 studies carried out between 1998 and 2016 (62 USA national, 16 non-USA national, 9 cross-national) found that political orientation and political party identification were the second most important predictors of views on climate change after environmental values (McCright et al. 2016). Ruiz et al. (2020) systematically reviewed 34 studies of non-US nations or clusters of nations and 30 studies of the USA alone. They found that in the non-US studies, ‘changed weather’ and ‘socio-altruistic values’ were the most important drivers of public attitudes. For the USA case, by contrast, political affiliation and the influence of corporations were most important. Widely varying media treatment of climate issues also affects public responses (Section 1.2.3.4). In summary, environmental and socio-altruistic values are the most significant influences on public opinion about climate change globally, while political views, political party affiliation, and corporate influence also had strong effects, especially in the USA (high confidence).

Furthermore, climate change itself is not uniform. Some regions face steady, readily observable change, while others experience high variability that masks underlying trends (Section 1.4.1); mostregions are subject to hazards, but some may also experience benefits, at least temporarily (Chapters 11, 12 and Atlas). This non-uniformity may lead to wide variation in public climate change awareness and risk perceptions at multiple scales (Howe et al., 2015; Lee et al., 2015). For example, short-term temperature trends, such as cold spells or warm days, have been shown to influence public concern (Hamilton and Stampone, 2013; Zaval et al., 2014; Bohr, 2017).

Given these manifold influences and the highly varied contexts of climate change communication, special care is required when expressing findings and uncertainties, including IPCC assessments that inform decision making. Throughout the IPCC’s history, all three Working Groups have sought to explicitly assess and communicate scientific uncertainty (Le Treut et al., 2007; Cubasch et al., 2013). Over time, the IPCC has developed and revised a framework to treat uncertainties consistently across assessment cycles, reports, and Working Groups through the use of calibrated language (Moss and Schneider, 2000; IPCC, 2005). Since its First Assessment Report (FAR; IPCC, 1990a), the IPCC has specified terms and methods for communicating authors’ expert judgments (Mastrandrea and Mach, 2011). During the AR5 cycle, this calibrated uncertainty language was updated and unified across all Working Groups (Mastrandrea et al., 2010, 2011). Box 1.1 summarizes this framework as it is used in AR6.

Box 1.1 | Treatment of Uncertainty and Calibrated Uncertainty Language in AR6

The AR6 follows the approach developed for AR5 (Box 1.1, Figure 1), as described in the ‘Guidance Notes for Lead Authors of the IPCC Fifth Assessment Report on Consistent Treatment of Uncertainties’ (Mastrandrea et al., 2010). The uncertainty Guidance Note used in AR6 clarifies the relationship between the qualitative description of confidence and the quantitative representation of uncertainty expressed by the likelihood scale. The calibrated uncertainty language emphasizes traceability of the assessment throughout the process. Key chapter findings presented in each chapter’s Executive Summary are supported in the chapter text by a summary of the underlying literature that is assessed in terms of evidence and agreement, confidence, and also likelihood, if applicable.

In all three Working Groups, author teams evaluate underlying scientific understanding and use two metrics to communicate the degree of certainty in key findings. These metrics are:

1. confidence: a qualitative measure of the validity of a finding, based on the type, amount, quality and consistency of evidence (e.g., data, mechanistic understanding, theory, models, expert judgment) and the degree of agreement.
2. Likelihood: a quantitative measure of uncertainty in a finding, expressed probabilistically (e.g., based on statistical analysis of observations or model results, or both, and expert judgement by the author team or from a formal quantitative survey of expert views, or both).

Throughout IPCC reports, the calibrated language indicating a formal confidence assessment is clearly identified byitalics (e.g., medium confidence). Where appropriate, findings can also be formulated as statements of fact without uncertainty qualifiers.

Box.1.1, Figure 1 (adapted from Mach et al., 2017) shows the idealized step-by-step process by which IPCC authors assess scientific understanding and uncertainties. It starts with the evaluation of the available evidence and agreement (steps 1–2). The following summary terms are used to describe the available evidence: limited, medium, orrobust; and the degree of agreement: low, medium, or high. Generally, evidence is most robust when there are multiple, consistent, independent lines of high-quality evidence.

If the author team concludes that there is sufficient evidence and agreement, the level of confidence can be evaluated. In this step, assessments of evidence and agreement are combined into a single metric (steps 3–5). The assessed level of confidence is expressed using five qualifiers: very low, low, medium, high, and very high. Step 4 depicts how summary statements for evidence and agreement relate to confidence levels. For a given evidence and agreement statement, different confidence levels can be assigned depending on the context, but increasing levels of evidence and degrees of agreement correlate with increasing confidence. When confidence in a finding is assessed to be low, this does not necessarily mean that confidence in its opposite is high, and vice versa. Similarly, low confidence does not imply distrust in the finding; instead, it means that the statement is the best conclusion based on currently available knowledge. Further research and methodological progress may change the level of confidence in any finding in future assessments.

Ifthe expert judgement of the author team concludes that there is sufficient confidence and quantitative/probabilistic evidence, assessment conclusions can be expressed with likelihood statements (steps 5–6). Unless otherwise indicated, likelihood statements are related to findings for which the authors’ assessment of confidence is highorvery high. Terms used to indicate the assessed likelihood of an outcome include: virtually certain: 99–100% probability, very likely : 90–100%, likely : 66–100%, about as likely as not : 33–66%, unlikely : 0–33%, very unlikely : 0–10%, exceptionally unlikely : 0–1%. Additional terms (extremely likely : 95–100%, more likely than not >50–100%, and extremely unlikely 0–5%) may also be used when appropriate.

Likelihood can indicate probabilities for single events or broader outcomes. The probabilistic information may build from statistical or modelling analyses, other quantitative analyses, or expert elicitation. The framework encourages authors, where appropriate, to present probability more precisely than can be done with the likelihood scale, for example with complete probability distributions or percentile ranges, including quantification of tails of distributions, which are important for risk management (Sections 1.2.2 and 1.4.4; Mach et al., 2017). In some instances, multiple combinations of confidence and likelihood are possible to characterize key findings

Box 1.1

Box 1.1, Figure 1 | The IPCC AR6 approach for characterizing understanding and uncertainty in assessment findings. This diagram illustrates the step-by-step process authors use to evaluate and communicate the state of knowledge in their assessment (Mastrandrea et al., 2010). Authors present evidence/agreement, confidence, or likelihood terms with assessment conclusions, communicating their expert judgments accordingly. Example conclusions drawn from Report are presented in the box at the bottom of the figure. Figure adapted from Mach et al. (2017) .

For example, avery likely statement might be made with high confidence, whereas a likely statement might be made with very high confidence. In these instances, the author teams consider which statement will convey the most balanced information to the reader.

Throughout this WGI Report, unless stated otherwise, uncertainty is quantified using 90% uncertainty intervals. The 90% uncertainty interval, reported in square brackets [x to y], is estimated to have a 90% likelihood of covering the value that is being estimated. The range encompasses the median value and there is an estimated 10% combined likelihood of the value being below the lower end of the range (x) and above its upper end (y). Often the distribution will be considered symmetric about the corresponding best estimate (as in the illustrative example in the figure), but this is not always the case. In this report, an assessed 90% uncertainty interval is referred to as a ‘very likely range’. Similarly, an assessed 66% uncertainty interval is referred to as a ‘likely range’.

Considerable critical attention has focused on whether applying the IPCC framework effectively achieves consistent treatment of uncertainties and clear communication of findings to users (Shapiro et al., 2010; Adler and Hirsch Hadorn, 2014). Specific concerns include, for example, the transparency and traceability of expert judgements underlying the assessment conclusions (Oppenheimer et al., 2016) and the context-dependent representations and interpretations of probability terms (Budescu et al., 2009, 2012; Janzwood, 2020). Budescu et al. (2014) surveyed 25 samples in 24 countries (a total of 10,792 individual responses), finding that even when shown IPCC uncertainty guidance, lay readers systematically misunderstood IPCC likelihood statements. When presented with a ‘high likelihood’ statement, they understood it as indicating a lower likelihood than intended by the IPCC authors. Conversely, they interpreted ‘low likelihood’ statements as indicating a higher likelihood than intended. In another study, British lay readers interpreted uncertainty language somewhat differently from IPCC guidance, but Chinese lay people reading the same uncertainty language translated into Chinese differed much more in their interpretations (Harris et al., 2013). Further, even though it is objectively more probable that wide uncertainty intervals will encompass true values, wide intervals were interpreted by lay people as implying subjective uncertainty or lack of knowledge on the part of scientists (Løhre et al., 2019). Mach et al. (2017) investigated the advances and challenges in approaches to expert judgment in AR5. Their analysis showed that the shared framework increased the overall comparability of assessment conclusions across all Working Groups and topics related to climate change, from the physical science basis to resulting impacts, risks, and options for response. Nevertheless, many challenges in developing and communicating assessment conclusions persist, especially for findings drawn from multiple disciplines and Working Groups, for subjective aspects of judgements, and for findings with substantial uncertainties (Adler and Hirsch Hadorn, 2014). In summary, the calibrated language cannot entirely prevent misunderstandings, including a tendency to systematically underestimate the probability of the IPCC’s higher-likelihood conclusions and overestimate the probability of the lower-likelihood ones (high confidence). However, a consistent and systematic approach across Working Groups to communicate the assessment outcomes is an important characteristic of the IPCC.

Some suggested alternatives are impractical, such as always including numerical values along with calibrated language (Budescu et al., 2014). Others, such as using positive instead of negative expressions of low-to-medium probabilities, show promise but were not proposed in time for adoption in AR6 (Juanchich et al., 2020). This report therefore retains the same calibrated language used in AR5 (Box 1.1). Like previous reports, AR6 also includes FAQs that express its chief conclusions in plain language designed for lay readers.

The framework for communicating uncertainties does not allow for indicating cases where ‘deep uncertainty’ is identified in the assessment (Adler and Hirsch Hadorn, 2014). The definition of deep uncertainty in IPCC assessments has been described in the context of SROCC (IPCC, 2019b; Box 5 in Abram et al., 2019): a situation of deep uncertainty exists when experts or stakeholders do not know or cannot agree on: (i) appropriate conceptual models that describe relationships among key driving forces in a system; (ii) the probability distributions used to represent uncertainty about key variables and parameters; and/or (iii) how to weigh and value desirable alternative outcomes (Cross-Chapter Box 1.2 and Annex VII: Glossary; Abram et al., 2019). Since AR5, ‘storylines’ or ‘narratives’ approaches have been used to address issues related to deep uncertainty, for example low-likelihood events that would have high impact if they occurred, to better inform risk assessment and decision making (Section 1.4.4). Chapter 9 (Section 9.2.3) notes deep uncertainty in long-term projections for sea level rise, and in processes related to marine ice-sheet instability and marine ice cliff instability.

As noted above, values – fundamental attitudes about what is important, good, and right – play critical roles in all human endeavours, including climate science. In AR5, Chapters 3 and 4 of the WGIII Assessment addressed the role of cultural, social and ethical values in climate change mitigation and sustainable development (Fleurbaey et al., 2014; Kolstad et al., 2014). These values include widely accepted concepts of human rights, enshrined in international law, that are relevant to climate impacts and policy objectives (Hall and Weiss, 2012; Peel and Osofsky, 2018; Setzer and Vanhala, 2019). Specific values – human life, subsistence, stability, and equitable distribution of the costs and benefits of climate impacts and policies – are explicit in the texts of the UNFCCC and the PA (Breakey et al., 2016; Dooley and Parihar, 2016). Here we address the role of values in how scientific knowledge is created, verified and communicated. Chapters 10, 12 and Cross-Chapter Box 12.2 address how the specific values and contexts of users can be addressed in the co-production of climate information.

The epistemic (knowledge-related) values of science include explanatory power, predictive accuracy, falsifiability, replicability, and justification of claims by explicit reasoning (Popper, 1959; Kuhn, 1977). These are supported by key institutional values, including openness, ‘organized scepticism’, and objectivity or ‘disinterestedness’ (Merton, 1973), operationalized as well-defined methods, documented evidence, publication, peer review, and systems for institutional review of research ethics (COSEPUP, 2009; Elliott, 2017). In recent decades, open data, open code and scientific cyber-infrastructure (notably the Earth System Grid Federation, a partnership of climate modelling centers dedicated to supporting climate research by providing secure, web-based, distributed access to climate model data) have facilitated scrutiny from a larger range of participants, and FAIR data stewardship principles – making data Findable, Accessible, Interoperable and Reusable (FAIR) – are being mainstreamed in many fields (Wilkinson et al., 2016). Climate science norms and practices embodying these scientific values and principles include the publication of data and model code, multiple groups independently analysing the same problems and data, model intercomparison projects (MIPs), explicit evaluations of uncertainty, and comprehensive assessments by national academies of science and the IPCC.

The formal Principles Governing IPCC Work (1998, amended 2003, 2006, 2012, 2013) specify that assessments should be ‘comprehensive, objective, open and transparent.’ The IPCC assessment process seeks to achieve these goals in several ways: by evaluating evidence and agreement across all relevant peer-reviewed literature, especially that published or accepted since the previous assessment; by maintaining a traceable, transparent process that documents the reasoning, data and tools used in the assessment; and by maximizing the diversity of participants, authors, experts, reviewers, institutions and communities represented, across scientific discipline, geographical location, gender, ethnicity, nationality and other characteristics. The multi-stage review process is critical to ensure an objective, comprehensive and robust assessment, with hundreds of scientists, other experts and governments providing comments to a series of drafts before the report is finalized.

Social values are implicit in many choices made during the construction, assessment and communication of climate science information (Heymann et al., 2017; Skelton et al., 2017). Some climate science questions are prioritized for investigation, or given a specific framing or context, because of their relevance to climate policy and governance. One example is the question of how the effects of a 1.5°C global warming would differ from those of a 2°C warming, an assessment specifically requested by Parties to the PA. The SR1.5 (2018) explicitly addressed this issue ‘within the context of sustainable development; considerations of ethics, equity and human rights; and the problem of poverty’ (Chapters 1 and 5; see also Hoegh-Guldberg et al., 2019) following the outcome of the approval of the outline of the Special Report by the IPCC during its 44th Session (Bangkok, Thailand, 17–20 October 2016). Likewise, particular metrics are sometimes prioritized in climate model improvement efforts because of their practical relevance for specific economic sectors or stakeholders. Examples include reliable simulation of precipitation in a specific region, or attribution of particular extreme weather events to inform rebuilding and future policy (Chapters 8 and 11; Intemann, 2015; Otto et al., 2018; James et al., 2019). Sectors or groups whose interests do not influence research and modelling priorities may thus receive less information in support of their climate-related decisions (Parker and Winsberg, 2018).

Recent work also recognizes that choices made throughout the research process can affect the relative likelihood of false alarms (overestimating the probability and/or magnitude of hazards) or missed warnings (underestimating the probability and/or magnitude of hazards), known respectively as Type I and Type II errors. Researchers may choose different methods depending on which type of error they view as most important to avoid, a choice that may reflect social values (Douglas, 2009; Knutti, 2018; Lloyd and Oreskes, 2018). This reflects a fundamental trade-off between the values of reliability and informativeness. When uncertainty is large, researchers may choose to report a wide range as very likely , even though it is less informative about potential consequences. By contrast, high-likelihood statements about a narrower range may be more informative, yet also prove less reliable if new evidence later emerges that widens the range. Furthermore, the difference between narrower and wider uncertainty intervals has been shown to be confusing to lay readers, who often interpret wider intervals as less certain (Løhre et al., 2019).

In AR6, ‘climate information’ refers to specific information about the past, current or future state of the climate system that is relevant for mitigation, adaptation and risk management. Cross-Chapter Box 1.1 is an example of climate information at the global scale. It provides climate change information with potential relevance for the global stocktake, and indicates where in AR6 this information may be found.

Responding to national and regional policymakers’ needs for tailored information relevant to risk assessment and adaptation, AR6 emphasizes assessment of regional information more than earlier reports. Here the phrase ‘regional climate information’ refers to predefined reference sets of land and ocean regions; various typological domains (such as mountains or monsoons); temporal frames including baseline periods as well as near term (2021–2040), medium term (2041–2060) and long term (2081–2100); and global warming levels (Chapters 10 and 12, Sections 1.4.1 and 1.4.5, and Atlas). Regional climate change information is constructed from multiple lines of evidence including observations, paleoclimate proxies, reanalyses, attribution of changes and climate model projections from both global and regional climate models (Sections 1.5.3 and 10.2–10.4). The constructed regional information needs to take account of user context and values for risk assessment, adaptation and policy decisions (Sections 1.2.3 and 10.5).

As detailed in Chapter 10, scientific climate information often requires ‘tailoring’ to meet the requirements of specific decision-making contexts. In a study of the UK Climate Projections 2009 (UKCP09) project, researchers concluded that climate scientists struggled to grasp and respond to users’ information needs because they lacked experience interacting with users, institutions and scientific idioms outside the climate science domain (Porter and Dessai, 2017). Economic theory predicts the value of ‘polycentric’ approaches to climate change informed by specific global, regional and local knowledge and experience (Ostrom, 1996, 2012). This is confirmed by numerous case studies of extended, iterative dialogue among scientists, policymakers, resource managers and other stakeholders to produce mutually understandable, usable, task-related information and knowledge, policymaking and resource management around the world (Lemos and Morehouse, 2005; Lemos et al., 2012, 2014, 2018; see Vaughan and Dessai, 2014 for a critical view). The SR1.5 (2018) assessed that ‘education, information, and community approaches, including those that are informed by indigenous knowledge and local knowledge, can accelerate the wide-scale behaviour changes consistent with adapting to and limiting global warming to 1.5°C. These approaches are more effective when combined with other policies and tailored to the motivations, capabilities and resources of specific actors and contexts (high confidence).’ These extended dialogic co-production and education processes have thus been demonstrated to improve the quality of both scientific information and governance (high confidence) (Section 10.5 and Cross Chapter Box 12.2).

Since AR5, climate services have increased at multiple levels (local, national, regional and global) to aid decision-making of individuals and organizations and to enable preparedness and early climate change action. These services include appropriate engagement from users and providers, are based on scientifically credible information and producer and user expertise, have an effective access mechanism, and respond to the users’ needs (Glossary; Hewitt et al., 2012). A Global Framework for Climate Services (GFCS) was established in 2009 by the World Meteorological Organization (WMO) in support of these efforts (Hewitt et al., 2012; Lúcio and Grasso, 2016). Climate services are provided across sectors and time scales, from sub-seasonal to multi-decadal, and support co-design and co-production processes that involve climate information providers, resource managers, planners, practitioners and decision makers (Brasseur and Gallardo, 2016; Trenberth et al., 2016; C.D. Hewitt et al., 2017). For example, they may provide high-quality data on temperature, rainfall, wind, soil moisture and ocean conditions, as well as maps, risk and vulnerability analyses, assessments, and future projections and scenarios. These data and information products may be combined with non-meteorological data, such as agricultural production, health trends, population distributions in high-risk areas, road and infrastructure maps for the delivery of goods, and other socio-economic variables, depending on users’ needs (WMO, 2020a). Cross-Chapter Box 12.2 illustrates the diversity of climate services with three examples from very different contexts.

The current landscapeof climate services is assessed in detail in Chapter 12 (Section 12.6), with a focus on multi-decadal time scales relevant for climate change risk assessment. Other information relevant to improving climate services for decision-making includes the assessment of methods to construct regional information (Chapter 10), as well as projections at the regional level (Atlas) relevant for impact and risk assessment in different sectors (Chapter 12).

Climate services focus on users with specific needs for climate information, but most people learn about climate science findings from media coverage. Since AR5, research has expanded on how mass media report climate change and how their audiences respond (Dewulf, 2013; Jaspal and Nerlich, 2014; Jaspal et al., 2014). For example, in five European Union (EU) countries, television coverage of AR5 used ‘disaster’ and ‘opportunity’ as its principal themes, but virtually ignored the ‘risk’ framing introduced by AR5 WGII (Painter, 2015) and now extended by the AR6 (Cross-Chapter Box 1.3). Other studies show that people react differently to climate change news when it is framed as a catastrophe (Hine et al., 2016), as associated with local identities (Sapiains et al., 2016), or as a social justice issue (Howell, 2013). Similarly, audience segmentation studies show that responses to climate change vary between groups of people with different, although not necessarily opposing, views on this phenomenon (e.g., Maibach et al., 2011; Sherley et al., 2014; Detenber et al., 2016). In Brazil, two studies have shown the influence of mass media on the high level of public climate change concern in that country (Rodasand Di Giulio, 2017; Dayrell, 2019). In the USA, analyses of television network news show that climate change receives minimal attention, is most often framed in a political context, and largely fails to link extreme weather events to climate change using appropriate probability framing (Hassol et al., 2016). However, recent evidence suggests that Climate Matters (an Internet resource to help US television weather forecasters link weather to climate change trends) may have had a positive effect on public understanding of climate change (Myers et al., 2020). Also, some media outlets have recently adopted and promoted terms and phrases stronger than the more neutral ‘climate change’ and ‘global warming’, including ‘climate crisis’, ‘global heating’, and ‘climate emergency’ (Zeldin-O’Neill, 2019). Google searches on those terms, and on ‘climate action’, increased 20-fold in 2019, when large social movements such as School Strikes forClimate gained worldwide attention (Thackeray et al., 2020). We thus assess that specific characteristics of media coverage play a major role in climate understanding and perception (high confidence), including how IPCC assessments are received by the general public.

Since AR5, social media platforms have dramatically altered the mass-media landscape, bringing about a shift from uni-directional transfer of information and ideas to more fluid, multi-directional flows (Pearce et al., 2019). A survey covering 18 Latin American countries (StatKnows-CR2, 2019) found that the main sources of information about climate change mentioned were the Internet (52% of mentions), followed by social media (18%). There are well-known challenges with social media, such as misleading or false presentations of scientific findings, incivility that diminishes the quality of discussion around climate change topics, and ‘filter bubbles’ that restrict interactions to those with broadly similar views (Anderson and Huntington, 2017). However, at certain moments (such as at the release of the AR5 WGI report), Twitter studies have found that more mixed, highly-connected groups existed, within which members were less polarized (Pearce et al., 2014; Williams et al., 2015). Thus, social media platforms may in some circumstances support dialogic or co-production approaches to climate communication. Because the contents of IPCC reports speak not only to policymakers, but also to the broader public, the character and effects of media coverage are important considerations across Working Groups.

Natural variations in both weather and longer time scale phenomena can temporarily mask or enhance any anthropogenic trends (e.g., Deser et al., 2012; Kay et al., 2015). These effects are more important on small spatial and temporal scales but can also occur on the global scale (Cross-Chapter Box 3.1).

Since AR5, many studies have examined the role of internal variability through the use of ‘large ensembles’. Each such ensemble consists of many different simulations by a single climate model for the same time period and using the same radiative forcings. These simulations differ only in their phasing of the internal climate variations (also see Section 1.5.4.2). A set of illustrative examples using one such large ensemble (Maher et al., 2019) demonstrates how variability can influence trends on decadal time scales (Figure 1.13). The long-term anthropogenic trends in this set of climate indicators are clearly apparent when considering the ensemble as a whole (grey shading), and all the individual ensemble members have very similar trends for ocean heat content (OHC), which is a robust estimate of the total energy stored in the climate system (e.g., Palmer and McNeall, 2014). However, the individual ensemble members can exhibit very different decadal trends in global surface air temperature (GSAT), UK summer temperatures, and Arctic sea ice variations. More specifically, for a representative 11-year period, both positive and negative trends can be found in all these surface indicators, even though the long-term trend is for increasing temperatures and decreasing sea ice. Periods in which the long-term trend is substantially masked or enhanced for more than 20 years are also visible in these regional examples. This highlights the fact that observations are expected to exhibit short-term trends which are larger or smaller than the long-term trend or that differ from the average projected trend from climate models, especially on continental spatial scales or smaller (Cross-Chapter Box 3.1). The actual observed trajectory can be considered as one realization of many possible alternative worlds that experienced different weather; this is also demonstrated by the construction of ‘observation-based large ensembles’, which are alternate possible realizations of historical observations that retain the statistical properties of observed regional weather (e.g., McKinnon and Deser, 2018).

Figure 1.13 | Simulated changes in various climate indicators under historical and RCP4.5 scenarios using the MPI ESM Grand Ensemble. The grey shading shows the 5–95% range from the 100-member ensemble. The coloured lines represent individual example ensemble members, with linear trends for the 2011–2021 period indicated by the dashed lines. Changes in ocean heat content (OHC) over the top 2000 m represents the integrated signal of global warming (left). The top row shows surface air temperature-related indicators (annual GSAT change and UK summer temperatures) and The bottom row shows Arctic sea ice-related indicators (annual ice volume and September sea ice extent). For smaller regions and for shorter time-period averages the variability increases and simulated short-term trends can temporarily mask or enhance anthropogenic changes in climate. Data from Maher et al. (2019). Further details on data sources and processing are available in the chapter data table (Table 1.SM.1).

Open figure

In the 1930s it was noted that temperatures were increasing at both local and global scales (Figure 1.8; Kincer, 1933; Callendar, 1938). At the time it was unclear whether the observed changes were part of a longer-term trend or a natural fluctuation; the ‘signal’ had not yet clearly emerged from the ‘noise’ of natural variability. Numerous studies have since focused on the emergence of changes in temperature using instrumental observations (e.g., Madden and Ramanathan, 1980; Wigley and Jones, 1981; Mahlstein et al., 2011, 2012; Lehner and Stocker, 2015; Lehner et al., 2017) and paleo-temperature data (e.g., Abram et al., 2016).

Since the IPCC Third’s Assessment Report in 2001, the observed signal of climate change has been unequivocally detected at the global scale (Section 1.3), and this signal is increasingly emerging from the noise of natural variability on smaller spatial scales and in a range of climate variables (FAQ 1.2). In this Report emergence of a climate change signal or trend refers to when a change in climate (the ‘signal’) becomes larger than the amplitude of natural or internal variations (defining the ‘noise’). This concept is often expressed as a ‘signal-to-noise’ ratio (S/N) and emergence occurs at a defined threshold of this ratio (e.g., S/N >1 or 2). Emergence can be estimated using observations and/or model simulations and can refer to changes relative to a historical or modern baseline (Section 12.5.2 and Glossary). The concept can also be expressed in terms of time (the ‘time of emergence’; Glossary) or in terms of a global warming level (Section 11.2.5; Kirchmeier-Young et al., 2019) and is also used to refer to a time when we can expect to see a response of mitigation activities that reduce emissions of GHGs or enhance their sinks (emergence with respect to mitigation; Section 4.6.3.1). Whenever possible, emergence should be discussed in the context of a clearly defined level of S/N or other quantification, such as ‘the signal has emerged at the level of S/N >2’, rather than as a simple binary statement. For an extended discussion, see Chapter 10 (Section 10.4.3).

Related to the concept of emergence is the detection of change (Chapter 3). Detection of change is defined as the process of demonstrating that some aspect of the climate, or a system affected by climate, has changed in some defined statistical sense, often using spatially aggregating methods that try to maximize S/N, such as ‘fingerprints’ (e.g., Hegerl et al., 1996), without providing a reason for that change. An identified change is detected in observations if its likelihood of occurrence by chance due to internal variability alone is determined to be small, for example, <10% (Glossary).

An example of observed emergence in surface air temperatures is shown in Figure 1.14. Both the largest changes in temperature and the largest amplitude of year-to-year variations are observed in the Arctic, with lower latitudes showing less warming and smaller year-to-year variations. For the six example regions shown in Figure 1.14, the emergence of changes in temperature is more apparent in Northern South America, East Asia and Central Africa, than for northern North America or Northern Europe. This pattern was predicted by Hansen et al. (1988) and noted in subsequent observations by Mahlstein et al. (2011) (Sections 10.3.4.3 and 12.5.2). Overall, tropical regions show earlier emergence of temperature changes than at higher latitudes (high confidence).

Figure 1.14 | The observed emergence of changes in temperature. (Top left) The total change in temperature estimated for 2020 relative to 1850–1900 (following Hawkins et al., 2020), showing the largest warming occurring in the Arctic. (Top right) The amplitude of estimated year-to-year variations in temperature. (Middleleft) The ratio of the observed total change in temperature and the amplitude of temperature variability (the ‘signal-to-noise (S/N) ratio’), showing that the warming is most apparent in the tropical regions (also see FAQ 1.2). (Middle right) The global warming level at which the change in local temperature becomes larger than the local year-to-year variability. The bottom panels show time series of observed annual mean surface air temperatures over land in various example regions, as indicated by the boxes in the top-left panel. The 1 and 2 standard deviations (σ) of estimated year-to-year variations for that region are shown by the pink shaded bands. Observed temperature data from Berkeley Earth (Rohde and Hausfather, 2020). Further details on data sources and processing are available in the chapter data table (Table 1.SM.1).

Open figure

Since AR5, the emergence of projected future changes has also been extensively examined, in variables including surface air temperature (Hawkins and Sutton, 2012; Kirtman et al., 2013; Tebaldi and Friedlingstein, 2013), ocean temperatures and salinity (Banks and Wood, 2002), mean precipitation (Giorgi and Bi, 2009; Maraun, 2013), drought (Orlowsky and Seneviratne, 2013), extremes (Diffenbaugh and Scherer, 2011; Fischer et al., 2014; King et al., 2015; Schleussner and Fyson, 2020), and regional sea level change (Lyu et al., 2014). The concept has also been applied to climate change impacts such as effects on crop growing regions (Rojas et al., 2019). In AR6, the emergence of oceanic signals such as regional sea level change and changes in water mass properties is assessed in Chapter 9 (Section 9.6.1.4); emergence of future regional changes is assessed in Chapter 10 (Section 10.4.3); the emergence of extremes as a function of global warming levels is assessed in Chapter 11 (Section 11.2.5); and the emergence of climatic impact-drivers for AR6 regions and many climate variables is assessed in Chapter 12 (Section 12.5.2).

Although the magnitude of any change is important, regions which have a larger signal of change relative to the background variations will potentially face greater risks than other regions, as they will see unusual or novel climate conditions more quickly (Frame et al., 2017). As in Figure 1.14, the signal of temperature change is often smaller in tropical countries, but their lower amplitude of variability means they may experience the effects of climate change earlier than the mid-latitudes. In addition, these tropical countries are often among the most exposed, due to large populations (Lehner and Stocker, 2015), and often more vulnerable (Harrington et al., 2016; Harrington and Otto, 2018; Russo et al., 2019). Higher levels of exposure and vulnerability increase the risk from climate-related impacts (Cross-Chapter Box 1.3). The rate of change is also important for many hazards (e.g., Loarie et al., 2009). Providing more information about changes and variations on regional scales, and the associated attribution to particular causes (Cross-Working Group Box: Attribution), is therefore important for adaptation planning.

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 (Section 4.2.3; e.g., Meehl et al., 2014; 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; Rougier, 2007; 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 (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 and Sutton, 2009; 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 (Section 1.5.4.8; e.g., Deser et al., 2012; Maher et al., 2019).

Figure 1.15 illustrates the relative size of these different uncertainty components using a ‘cascade of uncertainty’ (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.

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 (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).

Open figure

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 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 and Dessai, 2010; Halsnæs and Kaspersen, 2018; Hattermann et al., 2018).

In the AR6, certain low-likelihood outcomes are described and assessed because they may be associated with high levels of risk, and the greatest risks may not be associated with the most likely outcome. The aim of assessing these possible futures is to better inform risk assessment and decision-making. Two types are considered: (i) low-likelihood high-warming (LLHW) scenarios, which describe the climate in a world with very high climate sensitivity; and (ii) low-likelihood, high-impact outcomes that have a low likelihood of occurring, but would cause large potential impacts on societies or ecosystems.

An illustrative example of how low-likelihood outcomes can produce significant additional risks is shown in Figure 1.16. The Reasons for Concern (RFCs) produced by the IPCC AR5 WGII define the additional risks due to climate change at different global warming levels. These have been combined with Chapter 4 assessments of projected global temperature for different emissions scenarios (SSPs; Section 1.6), and Chapter 7 assessments about ECS. For example, even following an intermediate emissions scenario could result in high levels of additional risk if ECS is at the upper end of the very likely range. However, not all possible low-likelihood outcomes relate to ECS, and AR6 considers these issues in more detail than previous IPCC assessment reports (see Table 1.1 and Section 1.4.4.2 for some examples).

Figure 1.16 | Illustrating concepts of low-likelihood outcomes. Left: schematic likelihood distribution consistent with the IPCC AR6 assessments that equilibrium climate sensitivity (ECS) is likely in the range 2.5°C to 4.0°C, and very likely between 2.0°C and 5.0°C (Chapter 7). ECS values outside the assessed very likely range are designated low-likelihood outcomes in this example (light grey). Middle and right-hand columns: additional risks due to climate change for 2020–2090 using the Reasons For Concern (RFCs, see IPCC, 2014b), specifically RFC1 describing the risks to unique and threatened systems and RFC3 describing risks from the distribution of impacts (O’Neill et al., 2017b; Zommers et al., 2020). The projected changes of GSAT used are the 95%, median and 5% assessed ranges from Chapter 4 for each SSP (top, middle and bottom); these are designated High ECS, Mid-range ECS and Low ECS respectively. The ‘burning-ember’ risk spectrum of graduated colours is usually associated with levels of committed GSAT change; instead, this illustration associates the risk spectrum with the GSAT temperature reached in each year from 2020 to 2090. Note that this illustration does not include the vulnerability aspect of each SSP scenario. Further details on data sources and processing are available in the chapter data table (Table 1.SM.1).

Open figure

As societies are increasingly experiencing the impacts of climate change-related events, the climate science community is developing climate information tailored for particular regions and sectors. There is a growing focus on explaining and exploring complex physical chains of events or on predicting climate under various future socio-economic developments. Since AR5, ‘storylines’ or ‘narratives’ approaches have been used to better inform risk assessment and decision-making, to assist understanding of regional processes, and represent and communicate climate projection uncertainties more clearly. The aim is to help build a cohesive overall picture of potential climate change pathways that moves beyond the presentation of data and figures (Glossary; Fløttum and Gjerstad, 2017; Moezzi et al., 2017; Dessai et al., 2018; T.G. Shepherd et al., 2018).

In the broader IPCC context, the term ‘scenario storyline’ refers to a narrative description of one or more scenarios, highlighting their main characteristics, relationships between key driving forces and the dynamics of their evolution (e.g., emissions of short-lived climate forcers assessed in Chapter 6 are driven by ‘scenario storylines’; see Section 1.6). The AR6 WGI is mainly concerned with ‘physical climate storylines’. A physical climate storyline is a self-consistent and plausible physical trajectory of the climate system, or a weather or climate event, on time scales from hours to multiple decades (T.G. Shepherd et al., 2018). This approach can be used to constrain projected changes or specific events on specified explanatory elements such as projected changes of large-scale indicators (Box 10.2). For example, Hazeleger et al. (2015) suggested using ‘tales of future weather’, blending numerical weather prediction with a climate projection to illustrate the potential behaviour of future high-impact events (also see Hegdahl et al., 2020). Several studies describe how possible large changes in atmospheric circulation would affect regional precipitation and other climate variables, and discuss the various climate drivers that could cause such a circulation response (James et al., 2015; Zappa and Shepherd, 2017; Mindlin et al., 2020). Physical climate storylines can also help frame the causal factors of extreme weather events (Shepherd, 2016) and then be linked to event attribution (Section 11.2.2 and Cross-Working Group Box: Attribution).

Storyline approaches can be used to communicate and contextualize climate change information in the context of risk for policymakers and practitioners (Box 10.2; e.g., de Bruijn et al., 2016; Dessai et al., 2018; Scott et al., 2018; Jack et al., 2020). They can also help in assessing risks associated with LLHI events (Weitzman, 2011; Sutton, 2018), because they consider the ‘physically self-consistent unfolding of past events, or of plausible future events or pathways’ (Shepherd et al., 2018), which would be masked in a probabilistic approach. These aspects are important as the greatest risk need not be associated with the highest-likelihood outcome, and in fact will often be associated with low-likelihood outcomes. The storyline approach can also acknowledge that climate-relevant decisions in a risk-oriented framing will rarely be taken on the basis of physical climate change alone; instead, such decisions will normally take into account socio-economic factors as well (Shepherd, 2019).

In the AR6 WGIAssessment Report, these different storyline approaches are used in several places (see Table 1.1). Chapter 4 uses a storyline approach to assess the upper tail of the distribution of global warming levels (the storylines of high global warming levels) and their manifestation in global patterns of temperature and precipitation changes. Chapter 9 uses a storyline approach to examine the potential for, and early warning signals of, a high-end sea level scenario, in the context of deep uncertainty related to our current understanding of the physical processes that contribute to long-term sea level rise. Chapter 10 assesses the use of physical climate storylines and narratives as a way to explore uncertainties in regional climate projections, and to link to the specific risk and decision context relevant to a user, for developing integrated and context-relevant regional climate change information. Chapter 11 uses the term storyline in the framework of extreme event attribution. Chapter 12 assesses the use of a storylines approach with narrative elements for communicating climate (change) information in the context of climate services (Cross-Chapter Box 12.2).

An ‘abrupt change’ is defined in this report as a change that takes place substantially faster than the rate of change in the recent history of the affected component of a system (Glossary). In some cases, abrupt change occurs because the system state actually becomes unstable, such that the subsequent rate of change is independent of the forcing. We refer to this class of abrupt change as a ‘tipping point’, defined as a critical threshold beyond which a system reorganizes, often abruptly and/or irreversibly (Glossary; Lenton et al., 2008). Some of the abrupt climate changes and climate tipping points discussed in this Report could have severe local climate responses, such as extreme temperature, droughts, forest fires, ice-sheet loss and collapse of the thermohaline circulation (Sections 4.7.2, 5.4.9, 8.6 and 9.2.3).

There is evidence of abrupt changes in Earth’s history, and some of these events have been interpreted as tipping points (Dakos et al., 2008). Some of these are associated with significant changes in the global climate, such as deglaciations in the Quaternary (past 2.5 million years) and rapid warming at the Palaeocene–Eocene Thermal Maximum (around 55.5 million years ago; Bowen et al., 2015; Hollis et al., 2019). Such events changed the planetary climate for tens to hundreds of thousands of years, but at a rate that is actually much slower than projected anthropogenic climate change over this century, even in the absence of tipping points.

Such paleoclimate evidence has even fuelled concerns that anthropogenic GHGs could tip the global climate into a permanent hot state (Steffen et al., 2018). However, there is no evidence of such non-linear responses at the global scale in climate projections for the next century, which indicates a near-linear dependence of global temperature on cumulative GHG emissions (Sections 1.3.5, 5.5 and 7.4.3.1). At the regional scale, abrupt changes and tipping points, such as Amazon rainforest dieback and permafrost collapse, have occurred in projections with Earth System Models (Section 4.7.3; Drijfhout et al., 2015; Bathiany et al., 2020). In such simulations, tipping points occur in narrow regions of parameter space (e.g., CO₂ concentration or temperature increase), and for specific climate background states. This makes them difficult to predict using Earth system models (ESMs) relying on parmeterizations of known processes. In some cases, it is possible to detect forthcoming tipping points through time-series analysis that identifies increased sensitivity to perturbations as the tipping point is approached (e.g., ‘critical slowing-down’, Scheffer et al., 2012).

Some suggested climate tipping points prompt transitions from one steady state to another (Figure 1.17). Transitions can be prompted by perturbations such as climate extremes which force the system outside of its current well of attraction in the stability landscape; this is called noise-induced tipping (Figure 1.17a,b; Ashwin et al., 2012). For example, the tropical forest dieback seen in some ESM projections is accelerated by longer and more frequent droughts over tropical land (Good et al., 2013).

Figure 1.17 | Illustration of two types of tipping points: noise-induced (a, b) and bifurcation (c, d). (a) and (c) are example time-series (coloured lines) through the tipping point, with solid-black lines indicating stable climate states (e.g., low or high rainfall) and dashed lines representing the boundary between stable states. (b) and (d) are stability landscapes, which provide an intuitive understanding of the different types of tipping point. The ‘valleys’ represent different climate states the system can occupy, with ‘hilltops’ separating the stable states. The resilience of a climate state is implied by the depth of the valley. The current state of the system is represented by a ball. Both scenarios assume that the ball starts in the left-hand valley (dashed-black lines) and then through different mechanisms dependent on the type of tipping transitions to the right-hand valley (coloured lines). Noise-induced tipping events (a, b), for instance drought events causing sudden dieback of the Amazon rainforest, develop from fluctuations within the system. The stability landscape in this scenario remains fixed and stationary. A series of perturbations in the same direction, or one large perturbation, are required to force the system over the hilltop and into the alternative stable state. Bifurcation tipping events (c, d), such as a collapse of the thermohaline circulation in the Atlantic Ocean under climate change, occur when a critical level in the forcing is reached. Here the stability landscape is subjected to a change in shape. Under gradual anthropogenic forcing the left-hand valley begins to shallow and eventually vanishes at the tipping point, forcing the system to transition to the right-hand valley.

Open figure

Alternatively, transitions from one state to another can occur if a critical threshold is exceeded; this is called ‘bifurcation tipping’ (Figure 1.17c,d; Ashwin et al., 2012). The new state is defined as ‘irreversible’ on a given time scale if the recovery from this state takes substantially longer than the time scale of interest, which is decades to centuries for the projections presented in this report. A well-known example is the modelled irreversibility of the ocean’s thermohaline circulation in response to North Atlantic changes such as freshwater input from rainfall and ice-sheet melt (Rahmstorf et al., 2005; Alkhayuon et al., 2019), which is assessed in detail in Chapter 9 (Section 9.2.3).

The tipping point concept is most commonly framed for systems in which the forcing changes relatively slowly. However, this is not the case for most scenarios of anthropogenic forcing projected for the 21st century. Systems with inertia lag behind rapidly increasing forcing, which can lead to the failure of early warning signals or even the possibility of temporarily overshooting a bifurcation point without provoking tipping (Ritchie et al., 2019).

‘Surprises’ are a class of risk that can be defined as low-likelihood but well-understood events: they are events that cannot be predicted with current understanding. The risk from such surprises can be accounted for in risk assessments (Parker and Risbey, 2015). Examples relevant to climate science include: a series of major volcanic eruptions or a nuclear war, either of which would cause substantial planetary cooling (Robock et al., 2007; Mills et al., 2014); significant 21st century sea level rise due to marine ice sheet instability (MISI; Box 9.4); the potential for collapse of the stratocumulus cloud decks (Schneider et al., 2019) or other substantial changes in climate feedbacks (Section 7.4); and unexpected biological epidemics among humans or other species, such as the COVID-19 pandemic (Cross-Chapter Box 6.1; Forster et al., 2020; Le Quéré et al., 2020). The discovery of the hole in the ozone layerwas also a surprise even though some of the relevant atmospheric chemistry was known at the time. The term ‘unknownunknowns’ (Parker and Risbey, 2015) is also sometimes used in this context to refer to events that cannot be anticipated with presentknowledge or were of an unanticipated nature before they occurred.

The AR5 assessed regional-scale detection and attribution and assessed key regional climate phenomena and their relevance for future regional climate projections. This report shows that past and future climate changes and extreme weather events can be substantial on local and regional scales (Chapters 8–12 and Atlas), where they may differ considerably from global trends, not only in intensity but even in the direction of change (e.g., Fischer et al., 2013).

Although the evolution of global climate trends emerges as the net result of regional phenomena, average or aggregate estimates often do not reflect the intensity, variability and complexity of regional climate changes (Stammer et al., 2018; Shepherd, 2019). A fundamental aspect of the study of regional climate changes is the definition of characteristic climate zones, clusters or regions, across which the emergent climate change signal can be properly analysed and projected (see Atlas). Suitable sizes and shapes of such zones strongly depend not only on the climate variable and process of interest, but also on relevant multi-scale feedbacks.

There are several approaches to the classification of climate regions. When climate observation data was sparse and limited, the aggregation of climate variables was implicitly achieved through the consideration of biomes, giving rise to the traditional vegetation-based classification of Köppen (1936). In the last decades, the substantial increases in climate observations, climate modelling, and data processing capabilities have allowed new approaches to climate classification, for example through interpolation of aggregated global data from thousands of stations (Peel et al., 2007; Belda et al., 2014; Beck et al., 2018) or through data-driven approaches applied to delineate ecoregions that behave in a coherent manner in response to climate variability (Papagiannopoulou et al., 2018). Experience shows that each method has strengths and weaknesses through trade-offs between detail and convenience. For instance, a very detailed classification, with numerous complexly shaped regions derived from a large set of variables, may be most useful for the evaluation of climate models (Rubel and Kottek, 2010; Belda et al., 2015; Beck et al., 2018) and climate projections (Feng et al., 2014; Belda et al., 2016). In contrast, geometrically simple regions are often best suited for regional climate modelling and downscaling (e.g., the Coordinated Regional Climate Downscaling Experiment (CORDEX) domains; Section 1.5.3; Giorgi and Gutowski, 2015).

IPCC’s recognition of the importance of regional climates can be traced back to its First Assessment Report (FAR; IPCC, 1990a), where climate projections for 2030 were presented for five sub-continental regions (see Section 1.3.6 for an assessment of those projections). In subsequent reports, there has been a growing emphasis on the analysis of regional climate, including two special reports: one on regional impacts (IPCC, 1998) and another on extreme events (SREX, IPCC, 2012). A general feature of previous IPCC reports is that the number and coverage of climate regions vary according to the subject and across Working Groups. Such varied definitions have the advantage of optimizing the results for a particular application (e.g., national boundaries are crucial for decision-making, but they rarely delimit distinctive climate regions), whereas variable definitions of regions may have the disadvantage of hindering multidisciplinary assessments and comparisons between studies or Working Groups.

In this Report, regional climate change is primarily addressed through the introduction of four classes of regions (unless otherwise explicitly mentioned and justified). The first two are the unified WGI Reference Sets of (i) Land Regions and (ii) Ocean Regions, which are used throughout the Report. These are supplemented by additional sets of (iii) Typological Regions – used in Chapters 5, 8–12 and Atlas – and (iv) Continental Regions, which are mainly used for linking Chapters 11, 12 and Atlas with Working Group II (Figure 1.18). All four classes of regions are defined and described in detail in the Atlas. Here we summarize their basic features.

Figure 1.18 | Main region types used in this report. (a) AR6 WGI Reference Set of Land and Ocean Regions (Iturbide et al., 2020), consisting of 46 land regions and 15 ocean regions, including 3 hybrid regions (CAR, MED, SEA) that are both land and ocean regions. Abbreviations are explained to the right of the map. Notice that RAR, SPO, NPO and EPO extend beyond the 180º meridian, therefore appearing at both sides of the map (indicated by dashed lines). A comparison with the previous reference regions of AR5 WGI (IPCC, 2013a) is presented in the Atlas. (b) Example of typological regions: monsoon domains (see Chapter 8). Abbreviations are explained to the right of the map. The black contour lines represent the global monsoon zones, while the coloured regions denote the regional monsoon domains. The two stippled regions (EqAmer and SAfri) do receive seasonal rainfall, but their classification as monsoon regions is still under discussion. (c) Continental Regions used mainly in Chapter 12 and the Atlas. Stippled zones define areas that are assessed in both regions (e.g., the Caribbean is assessed as Small Islands and also as part of Central America). Small Islands are ocean regions containing small islands with consistent climate signals and/or climatological coherence.

Open figure

The Reference Sets of Land and Ocean Regions are polygonal, sub-continental domains, defined through a combination of environmental, climatic and non-climatic (e.g., pragmatic, technical, historical) factors, in accordance with the literature and climatological reasoning based on observed and projected future climate. Merging the diverse functions and purposes of the regions assessed in the literature into a common reference set implies a certain degree of compromise between simplicity, practicality and climate consistency. For instance, Spain is fully included in the Mediterranean (MED) Reference Region, but is one of the most climatically diverse countries in the world. Likewise, a careful comparison of panels a and b of Figure 1.18 reveals that the simplified southern boundary of the Sahara (SAH) Reference Region slightly overlaps the northern boundary of the West African Monsoon Typological Region. As such, the resulting Reference Regions are not intended to precisely represent climates, but rather to provide simple domains suitable for regional synthesis of observed and modelled climate and climate change information (Iturbide et al., 2020). In particular, CMIP6 model results averaged over Reference Regions are presented in the Atlas.

The starting point for defining the AR6 Reference Sets of Land Regions was the collection of 26 regions introduced in SREX (IPCC, 2012). The SREX collection was then revised, reshaped, complemented and optimized to reflect the recent scientific literature and observed climate-change trends, giving rise to the novel AR6 Reference Set of 46 Land Regions. Additionally, AR6 introduces a new Reference Set of 15 Ocean Regions (including 3 Hybrid Regions that are treated as both: land and ocean), which complete the coverage of the whole Earth (Iturbide et al., 2020).

Particular aspects of regional climate change are described by specialized domains called Typological Regions (Figure 1.18b). These regions cover a wide range of spatial scales and are defined by specificfeatures, called typologies. Examples of typologies include: tropical forests, deserts, mountains, monsoon regions and megacities, among others. Typological Regions are powerful tools to summarize complex aspects of climate defined by a combination of multiple variables. For this reason, they are used in many chapters of AR6 WGI and WGII (e.g., Chapters 8–12 and Atlas).

Finally, consistency with WGII is also pursued in Chapters 11, 12 and the Atlas through the use of a set of Continental Regions (Figure 1.18c), based on the nine continental domains defined in AR5 WGII Part B (Hewitson et al., 2014). These are classical geopolitical divisions of Africa, Asia, Australasia, Europe, North America, Central and South America, plus Small Islands, Polar Regions, and the Ocean. In AR6 WGI, five hybrid zones (Caribbean–Small Islands, East Europe–Asia, European Arctic, North American Arctic, and Northern Central America) are also identified, which are assessed in more than one Continental Region. Additional consistency with WGIII is pursued in Chapter 6 through the use of sub-continental domains which essentially form a subset of the Continental Set of Regions (Figure 1.18c and Section 6.1).

The lockdowns and societal outcomes arising from the COVID-19 pandemic pose a new threat to observing systems. For example, WMO and UNESCO-IOC (Intergovernmental Oceanographic Commission) published a summary of the changes to Earth system observations during COVID-19 (WMO, 2020b). Fewer aircraft flights (down 75–90% in May 2020, depending on region) and ship transits (down 20% in May 2020) mean that onboard observations from those networks have reduced in number and frequency (James et al., 2020; Ingleby et al., 2021). Europe has deployed more radiosonde soundings to account for the reduction in data from air traffic. Fewer ocean observing buoys were deployed during 2020, and reductions have been particularly prevalent in the tropics and Southern Hemisphere. The full consequences of the pandemic, and responses to it, will come to light over time. Estimates of the effect of the reduction in aircraft data assimilation on weather forecasting skill are small (James et al., 2020; Ingleby et al., 2021), potentially alleviating concerns about veracity of future atmospheric reanalyses of the COVID-19 pandemic period.

Surface-based networks have reduced in their coverage or range of variables measured due to COVID-19 and other factors. Over land, several factors, including the ongoing transition from manual to automatic observations of weather, have reduced the spatial coverage of certain measurement types, including rainfall intensity, radiosonde launches and pan evaporation, posing unique risks to datasets used for climate assessment (WMO, 2017; Lin and Huybers, 2019). Ship-based measurements, which are important for ocean climate and reanalyses through time (Smith et al., 2019), have been in decline due to the number of ships contributing observations. There has also been a decline in the number of variables recorded by ships, but an increase in the quality and time-resolution of others (e.g., sea level pressure, Kent et al., 2019).

Certain satellite frequencies are used to detect meteorological features that are vital to climate change monitoring. These can be disturbed by certain radio communications (Anterrieu et al., 2016), although scientists work to remove noise from the signal (Oliva et al., 2016). For example, water vapour in the atmosphere naturally produces a weak signal at 23.8 gigahertz (GHz), which is within the range of frequencies of the 5G cellular communications network (Liu et al., 2021). Concern has been raised about potential leakage from 5G network transmissions into the operating frequencies of passive sensors on existing weather satellites, which could adversely influence their ability to remotely observe water vapour in the atmosphere (Yousefvand et al., 2020).

Threats to observational capacity also include the loss of natural climate archives that are disappearing as a direct consequence of warming temperatures. Ice-core records from vulnerable alpine glaciers in the tropics (Permana et al., 2019) and the mid-latitudes (Gabrielli et al., 2016; Winski et al., 2018; Moreno et al., 2021) document more frequent melt layers in recent decades, with glacial retreat occurring at a rate and geographic scale that is unusual in the Holocene (Solomina et al., 2015). The scope and severity of coral bleaching and mortality events have increased in recent decades (Hughes et al., 2018), with profound implications for the recovery of coral climate archives from new and existing sites. An observed increase in the mortality of larger, long-lived trees over the last century is attributed to a combination of warming, land-use change, and disturbance (e.g., McDowell et al., 2020). The ongoing loss of these natural, high-resolution climate archives endanger an end in their coverage over recent decades, given that many of the longest monthly- to annually-resolved paleoclimate records were collected in the 1960s to 1990s (e.g., the PAGES2K database as represented in PAGES 2k Consortium, 2017). This gap presents a barrier to the calibration of existing decades-to-centuries-long records needed to constrain past temperature and hydrology trends and extremes.

Historical archives of weather and climate observations contained in ships’ logs, weather diaries, observatory logbooks and other sources of documentary data also risk being lost, for example to natural disasters or accidental destruction. These archives include measurements of temperature (air and sea surface), rainfall, surface pressure, wind strength and direction, sunshine amount, and many other variables back into the 19th century. While internationally coordinated data-rescue efforts are focused on recovering documentary sources of past weather and climate data (e.g., Allan et al., 2011), no such coordinated efforts exist for vulnerable paleoclimate archives. Furthermore, oral traditions about local and regional weather and climate from indigenous peoples represent valuable sources of information, especially when used in combination with instrumental climate data (Makondo and Thomas, 2018), but are in danger of being lost as indigenous knowledge-holders pass away.

In summary, while the quantity, quality and diversity of climate system observations have grown since AR5, the loss or potential loss of several critical components of the observational network is also evident (high confidence).

Extensive improvements have been made in global atmospheric reanalyses since AR5. The growing demand for high-resolution data has led to the development of higher-resolution atmospheric reanalyses, such as the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2; Gelaro et al., 2017) and ERA5 (Hersbach et al., 2020). There is a focus on ERA5 here because it has been assessed as of high enough quality to present temperature trends alongside more traditional observational datasets (Section 2.3.1.1) and is also used in the Interactive Atlas.

Atmospheric reanalyses that were assessed in AR5 are still being used in the literature, and results from ERA-Interim (about 80 km resolution, production stopped in August 2019; Dee et al., 2011), the Japanese 55-year Reanalysis (JRA-55; Ebita et al., 2011; Kobayashi et al., 2015; Harada et al., 2016) and Climate Forecast System Reanalysis (CFSR; Saha et al., 2010) are assessed in AR6. Some studies still also use the NCEP/NCAR reanalysis, particularly because it extends back to 1948 and is updated in near-real time (Kistler et al., 2001). Older reanalyses have a number of limitations, which have to be accounted for when assessing the results of any study that uses them.

ERA5 provides hourly atmospheric fields at about 31 km resolution on 137 levels in the vertical, as well as land-surface variables and ocean waves. It is available from 1979 onwards and is updated in near-real time, with plans to extend back to 1950. A 10-member ensemble is also available at coarser resolution, allowing uncertainty estimates to be provided (e.g., Section 2.3). MERRA-2 includes many updates from the earlier version, including the assimilation of aerosol observations, several improvements to the representation of the stratosphere, including ozone, and improved representations of cryospheric processes. All of these improvements increase the usefulness of these reanalyses (Section 7.3; Hoffmann et al., 2019).

Models of atmospheric composition and emissions sources and sinks allow the forecast and reanalysis of constituents such as O₃, carbon monoxide (CO), nitrogen oxides (NOx) and aerosols. The Copernicus Atmosphere Monitoring Service (CAMS) reanalysis shows improvement against earlier atmospheric composition reanalyses, giving greater confidence for its use to study trends and evaluate models (Section 7.3; e.g., Inness et al., 2019).

The intercomparison of reanalyses with each other, or with earlier versions, is often done for particular variables or aspects of the simulation. ERA5 is assessed as the most reliable reanalysis for climate trend assessment (Section 2.3). Compared to ERA-Interim, the ERA5 forecast model and assimilation system, as well as the availability of improved reprocessing of observations, resulted in relatively smaller errors when compared to observations, including a better representation of global energy budgets, radiative forcing from volcanic eruptions (e.g., Mt. Pinatubo: Allan et al., 2020), the partitioning of surface energy (Martens et al., 2020), and wind (Kaiser-Weiss et al., 2015, 2019; Borsche et al., 2016; Scherrer, 2020). In ERA5, higher resolution means a better representation of Lagrangian motion convective updrafts, gravity waves, tropical cyclones, and other meso- to synoptic-scale features of the atmosphere (Hoffmann et al., 2019; Martens et al., 2020). Low-frequency variability is found to be generally well represented and, from 10 hPa downwards, patterns of anomalies in temperature match those from the ERA-Interim, MERRA-2 and JRA-55 reanalyses. Inhomogeneities in the water cycle have also been reduced (Hersbach et al., 2020).

Precipitation is not usually assimilated in reanalyses and, depending on the region, reanalysis precipitation can differ from observations by more than the observational error (Zhou and Wang, 2017; Sun et al., 2018; Alexander et al., 2020; Bador et al., 2020), although these studies did not include ERA5. Assimilation of radiance observations from microwave imagers which, over ice-free ocean surfaces, improve the analysis of lower-tropospheric humidity, cloud liquid water and ocean-surface wind speed have resulted in improved precipitation outputs in ERA5 (Hersbach et al., 2020). Global averages of other fields, particularly temperature, from ERA-Interim and JRA-55 reanalyses continue to be consistent over the last 20 years with surface observational data sets that include the polar regions (Simmons and Poli, 2015), although biases in precipitation and radiation can influence temperatures regionally (Zhou et al., 2018). The global average surface temperature from MERRA-2 is far cooler in recent years than temperatures derived from ERA-Interim and JRA-55, which may be due to the assimilation of aerosols and their interactions (Section 2.3).

A number of regional atmospheric reanalyses (Section 10.2.1.2) have been developed, such as COSMO-REA (Wahl et al., 2017), and the Australian Bureau of Meteorology Atmospheric high-resolution Regional Reanalysis for Australia (BARRA; Su et al., 2019). Regional reanalyses can add value to global reanalyses due to the lower computational requirements, and can allow multiple numerical weather prediction models to be tested (e.g., Kaiser-Weiss et al., 2019). There is some evidence that these higher-resolution reanalyses better capture precipitation variability than global lower-resolution reanalyses (Jermey and Renshaw, 2016; Cui et al., 2017). They are further assessed in Section 10.2.1.2 and used in the Interactive Atlas.

In summary, the improvements in atmospheric reanalyses, and the greater number of years since the routine ingestion of satellite data began, relative to AR5, mean that there is increased confidence in using atmospheric reanalysis products alongside more standard observation-based datasets in AR6 (high confidence).

Although reanalyses such as ERA5 take advantage of new observational datasets and present a great improvement in atmospheric reanalyses, the issues introduced by the evolving observational network remain. Sparse input reanalyses, where only a limited set of reliable and long-observed records are assimilated, address these issues, with the limitation of fewer observational constraints. These efforts are sometimes called centennial-scale reanalyses. One example is the atmospheric 20th century Reanalysis (Compo et al., 2011; Slivinski et al., 2021) which assimilates only surface and sea-level pressure observations, and is constrained by time-varying observed changes in atmospheric constituents, prescribed sea surface temperatures and sea ice concentration, creating a reconstruction of the weather over the whole globe every three hours for the period 1806–2015. The ERA-20C atmospheric reanalysis (covering 1900–2010; Poli et al., 2016) also assimilates marine wind observations, and CERA-20C is a centennial-scale reanalysis that assimilates both atmospheric and oceanic observations for the 1901–2010 period (Laloyaux et al., 2018). These centennial-scale reanalyses are often run as ensembles that provide an estimate of the uncertainty in the simulated variables over space and time. Slivinski et al. (2021) conclude that the uncertainties in surface circulation fields in version 3 of the 20th century Reanalysis are reliable and that there is also skill in its tropospheric reconstruction over the 20th century. Long-term changes in other variables, such as precipitation, also agree well with direct observation-based datasets (Sections 2.3.1.3 and 8.3.2.8).

Since AR5, ocean reanalyses have improved due to: increased model resolution (Zuo et al., 2017; Lellouche et al., 2018; Heimbach et al., 2019); improved physics (Storto et al., 2019); improvements in the atmospheric forcing from atmospheric reanalyses (see Section 1.5.2.1.3); and improvements in the data quantity and quality available for assimilation (e.g., Lellouche et al., 2018; Heimbach et al., 2019), particularly due to Argo observations (Annex I; Zuo et al., 2019).

The first Ocean Reanalyses Intercomparison project (ORA-IP; Balmaseda et al., 2015) focussed on the uncertainty in key climate indicators, such as ocean heat content (Palmer et al., 2017), thermosteric sea level (Storto et al., 2017, 2019), salinity (Shi et al., 2017), sea ice extent (Chevallier et al., 2017), and the AMOC (Karspeck et al., 2017). Reanalysis uncertainties occur in areas of inhomogeneous or sparse observational data sampling, such as for the deep ocean, the Southern Ocean, and western boundary currents (Lellouche et al., 2018; Storto et al., 2019). Intercomparisons have also been dedicated to specific variables such as mixed-layer depths (Toyoda et al., 2017), eddy kinetic energy, globally (Masina et al., 2017) and in the polar regions (Uotila et al., 2019). Karspeck et al. (2017) found disagreement in the AMOC variability and strength in reanalyses over observation-sparse periods, whereas Jackson et al. (2019) reported a lower spread in AMOC strength across an ensemble of ocean reanalyses of the recent period (1993–2010), linked to improved observation availability for assimilation. Reanalyses also have a larger spread of ocean heat uptake than data-only products and can produce spurious overestimates of heat uptake (Palmer et al., 2017), which is important in the context of estimating climate sensitivity (Storto et al., 2019). The ensemble approach for ocean reanalyses provides another avenue for estimating uncertainties across ocean reanalyses (Storto et al., 2019).

While there are still limitations in their representation of oceanic features, ocean reanalyses add value to products based only on observation, and are used to inform assessments in AR6 (Chapters 2, 3, 7 and 9). Reanalyses of the atmosphere or ocean alone may not account for important atmosphere–ocean coupling, motivating the development of coupled reanalyses (Laloyaux et al., 2018; Schepers et al., 2018; Penny et al., 2019), but these are not assessed in AR6.

Longer reanalyses that extend further back in time than the beginning of the instrumental record are being developed. They include the complete integration of paleoclimate archives and newly available early instrumental data into extended reanalysis datasets. Such integration leverages ongoing development of climate models that can simulate paleoclimate records in their units of analysis (i.e., oxygen isotope composition, tree ring width, etc.), in many cases using physical climate variables as input for so-called proxy system models (Evans et al., 2013; Dee et al., 2015). Ensemble Kalman filter data assimilation approaches allow for combining paleoclimate data and climate model data to generate annually resolved fields (Last Millenium Reanalysis, Hakim et al., 2016; Tardif et al., 2019) or even monthly fields (Franke et al., 2017). This allows for a greater understanding of decadal variability (Parsons and Hakim, 2019) and greater certainty around the full range of the frequency and severity of climate extremes. This, in turn, allows for better-defined detection of change. It also helps to identify the links between biogeochemical cycles, ecosystem structure and ecosystem functioning, and to provide initial conditions for further model experiments or downscaling (Chapter 2).

The developments in reanalyses described above mean that they are now used across a range of applications. In AR6, reanalyses provide information for fields and in regions where observations are limited. There is growing confidence that modern reanalyses can provide another line of evidence in describing recent temperature trends (Tables 2.4 and 2.5). As their spatial resolution increases, the exploration of fine-scale extremes in both space and time becomes possible (e.g., wind; Kaiser-Weiss et al., 2015). Longer reanalyses can be used to describe the change in the climate over the last 100 to 1000 years. Reanalyses have been used to help post-process climate model output, and drive impact models; however, they are often bias adjusted first (Cross-Chapter Box 10.2; e.g., Weedon et al., 2014). Copernicus Climate Change Service (C3S) provides a bias-adjusted dataset for global land areas based on ERA5 called WFDE5 (Cucchi et al., 2020) which, combined with ERA5 information over the ocean (W5E5; Lange, 2019), is used as the AR6 Interactive Atlas reference for the bias adjustment of model output.

The growing interest in longer-term climate forecasts (from seasonal to multi-year and decadal) means that reanalyses are now more routinely being used to develop the initial state for these forecasts, such as for the Decadal Climate Prediction Project (DCPP; Boer et al., 2016). Ocean reanalyses are now being used routinely in the context of climate monitoring, (e.g., the Copernicus Marine Environment Monitoring Service Ocean State Report; von Schuckmann et al., 2019).

In summary, reanalyses have improved since AR5 and can increasingly be used as a line of evidence in assessments of the state and evolution of the climate system (high confidence). Reanalyses provide consistency across multiple physical quantities, and information about variables and locations that are not directly observed. Since AR5, new reanalyses have been developed with various combinations of increased resolution, extended records, more consistent data assimilation, estimation of uncertainty arising from the range of initial conditions, and an improved representation of the atmosphere or ocean system. While noting their remaining limitations, this Report uses the most recent generation of reanalysis products alongside more standard observation-based datasets.

Earth system models are mathematical formulations of the natural laws that govern the evolution of climate-relevant systems: atmosphere, ocean, cryosphere, land, and biosphere, as well as the carbon cycle (Flato, 2011). They build on the fundamental laws of physics (e.g., Navier–Stokes or Clausius–Clapeyron equations) or empirical relationships established from observations and, when possible, they are constrained by fundamental conservation laws (e.g., mass and energy). The evolution of climate-relevant variables is computed numerically using high-performance computers (André et al., 2014; Balaji et al., 2017), on three-dimensional discrete grids (Staniforth and Thuburn, 2012). The spatial (and temporal) resolution of these grids in both the horizontal and vertical directions determines which processes need to be parameterized or whether they can be explicitly resolved. Developments since AR5 in model resolution, parameterizations and modelling of the land and ocean biosphere and of biogeochemical cycles are discussed below.

1.5.3.1.1 Model gridsand resolution

The horizontal resolution and the number of vertical levels in ESMs is generally higher in CMIP6 than in CMIP5 (Figure 1.19). Global models with finer horizontal grids better represent many aspects of the circulation of the atmosphere (Gao et al., 2020; Schiemann et al., 2020) and ocean (Bishop et al., 2016; Storkey et al., 2018), bringing improvements in the simulation of the global hydrological cycle (Roberts et al., 2018). CMIP6 includes a dedicated effort (HighResMIP, Haarsma et al., 2016) to explore the effect of higher horizontal resolution, such as ~50 km, ~25 km and even ~10 km (Section 1.5.4.2 and Annex II, Table AII.6). Improvements are documented in the highest-resolution coupled models used for HighResMip (Hewitt et al., 2017; Roberts et al., 2019). Flexible grids allowing spatially variable resolution in the atmosphere (McGregor, 2015; Giorgetta et al., 2018) and in the ocean (Wang et al., 2014; Petersen et al., 2019) are more widely used than at the time of the AR5.

Figure 1.19 | Resolution of the atmospheric and oceanic components of global climate models participating in CMIP5, CMIP6 and HighResMIP: (a, b) horizontal resolution (km), and (c, d) number of vertical levels. Darker-colour circles indicate high-top models (in which the top of the atmosphere is above 50 km). The crosses are the median values. These models are documented in Annex II. Note that duplicated models in a modelling group are counted as one entry when their horizontal and vertical resolutions are the same. For HighResMIP, one atmosphere–ocean coupled model with the highest resolution from each modelling group is used. The horizontal resolution (rounded to 10 km) is the square root of the surface area of the Earth divided by the number of grid points, or the area of the ocean surface divided by the number of surface ocean grid points, for the atmosphere and ocean, respectively.

Open figure

The number of vertical levels in the atmosphere of global models has increased (Figure 1.19), partly to enable simulations to include higher levels in the atmosphere and better represent stratospheric processes (Charlton-Perez et al., 2013; Kawatani et al., 2019). Half the modelling groups now use ‘high-top’ models with a top level above the stratopause (a pressure of about 1 hPa). The number of vertical levels in the ocean models has also increased in order to achieve finer resolution over the water column and especially in the upper mixed layer and to better resolve the diurnal cycle (Section 3.5 and Annex II; Bernie et al., 2008).

Despite the documented progress of higher resolution, the model evaluation carried out in subsequent chapters shows that improvements between CMIP5 and CMIP6 remain modest at the global scale (Section 3.8.2; Bock et al., 2020). Lower resolution alone does not explain all model biases, for example, a low blocking frequency (Davini and D’Andrea, 2020) or a wrong shape of the Intertropical Convergence Zone (Tian and Dong, 2020). Model performance depends on model formulation and parameterizations as much as on resolution (Chapters 3, 8 and 10).

When developing climate models, choices have to be made in a number of areas. Besides model formulation and resolution, parameterizations of unresolved processes also involve many choices as, for each of these, several parameters can be set. The acceptable range for these parameters is set by mathematical consistency (e.g., convergence of a numerical scheme), physical considerations (e.g., energy conservation), observations, or a combination of factors. Model developers choose a set of parameters that both falls within this range and mimics observations of individual processes or their statistics.

An initial set of such choices is usually made by (often extensive) groups of modellers working on individual components of the Earth system (e.g., ocean, atmosphere, land or sea ice). As components are assembled to build an ESM, the choices are refined so that the simulated climate best represents a number of pre-defined climate variables, or ‘tuning targets’. When these are met the model is released for use in intercomparisons such as CMIP. Tuning targets can be one of three types: mean climate; regional phenomena and features; or historical trends (Hourdin et al., 2017). One example of such a goal is that when the simulated climate system receives energy from the sun in accordance with what we observe today, the resulting mean equilibrium temperature should also be close to observations. Whether tuning should be performed to facilitate accurate simulation of long-term trends such as changes in global mean temperature over the historical era, or rather be performed for each process independently such that all collective behaviour is emergent, is an open question (Schmidt et al., 2017; Burrows et al., 2018).

Each modelling group has its own strategy and, after AR5, a survey was conducted to understand the tuning approach used in 23 CMIP5 modelling centres. The results are discussed in Hourdin et al. (2017), which stresses that the behaviour of ESMs depends on the tuning strategy. An important recommendation is that the calibration steps that lead to particular model tuning should be carefully documented. In CMIP6 each modelling group now describes the three levels of tuning, both for the complete ESM and for the individual components (available at https://explore.es-doc.organd in the published model descriptions, Annex II: Models). The most important global tuning target for CMIP6 models is the net top-of-the-atmosphere (TOA) heat flux and its radiative components. Other global targets include: the decomposition of the energy fluxes at TOA into a clear sky component and a component due to the radiative effect of clouds, global mean air and ocean temperature, sea ice extent, sea ice volume, glacial mass balance, and the global root mean square error of precipitation. The TOA heat flux balance is achieved using a diversity of approaches, usually unique to each modelling group. Adjustments are made for parameters associated with uncertain or poorly constrained processes (Schmidt et al., 2017), for example the aerosol indirect effects, adjustments to ocean albedo, marine dimethyl sulfide (DMS) parameterization, or cloud properties (Mauritsen and Roeckner, 2020).

Regional tuning targets include: the AMOC, the Southern Ocean circulation, and temperature profiles in ocean basins (Golaz et al., 2019; Sellar et al., 2019); regional land properties and precipitations (Mauritsen et al., 2019; Yukimoto et al., 2019) ; latitudinal distribution of radiation (Boucher et al., 2020); spatial contrasts in TOA radiative fluxes or surface fluxes; and stationary waves in the Northern Hemisphere (Schmidt et al., 2017; Yukimoto et al., 2019).

Even with some core commonalities of approaches to model tuning, practices can differ, such as the use of initial drift from initialized forecasts, the explicit use of the transient observed record for the historical period, or the use of the present-day radiative imbalance at the TOA as a tuning target rather than an equilibrated pre-industrial balance. The majority of CMIP6 modelling groups report that they do not tune their model for the observed trends during the historical period (23 out of 29 groups), nor for ECS (25 out of 29). ECS and TCR are thus emergent properties for a large majority of models. The effect of tuning on model skill and ensemble spread in CMIP6 is further discussed in Section 3.3.

The need for accurate climate information at the regional scale is increasing (Section 10.1). High-resolution global climate models, such as those taking part in HighResMIP, provide more detailed information at the regional scale (Roberts et al., 2018). However, due to the large computational resources required by these models, only a limited number of simulations per model are available. In addition to CMIP global models, regional information can be derived using regional climate models (RCMs) and downscaling techniques, presented in Chapter 10 and the Atlas. RCMs are dynamical models, similar to GCMs, that simulate a limited region and are forced with boudary conditions from a global simulation, often correcting for biases (Section 10.3, Cross-Chapter Box 10.2 and Annex II). This approach allows the use of a higher resolution within the chosen domain, and thus better represents important drivers of regional climate such as mountain ranges, land management and urban effects. RCMs resolving atmospheric convection explicitly are now included in intercomparisons (Coppola et al., 2020) and are used in Chapters 10, 11 and 12. Other approaches, such as statistical downscaling, are also used to generate regional climate projections (Section 10.3; Maraun and Widmann, 2018).

The number of climate centres or consortia that carry out global climate simulations and projections has grown from 11 in the first CMIP to 19 in CMIP5 and 28 for CMIP6 (Section 1.5.4.2 and Annex II). Regional climate models participating in the Coordinated Regional Downscaling Experiment (CORDEX) are more diverse than the global ESMs (Section 1.5.4.3 and Annex II) and engage an even wider international community (Figure 1.20).

Figure 1.20 | World map showing the increased diversity of modelling centres contributing to CMIP and CORDEX. Climate models are often developed by international consortia. One such consortium, EC-Earth, is shown as an example under the label 8 EU Cities (involving SMHI, Sweden; KNMI, The Netherlands; DMI, Denmark; AEMET, Spain; Met Éireann, Ireland; CNR‐ISAC, Italy; Instituto de Meteorologia, Portugal; and FMI, Finland). There are too many such collaborations to display all of them on this map. More complete information about institutions contributing to CORDEX and CMIP6 is found in Annex II.

Open figure

Earth system models of intermediate complexity(EMICs) complement the model hierarchy and fill the gap between conceptual, simple climate models and complex GCMs or ESMs (Claussen et al., 2002). EMICs are simplified; they include processes in a more parameterized, rather than explicitly calculated, form and generally have lower spatial resolution compared to the complex ESMs. As a result, EMICs require much less computational resource and can be integrated for many thousands of years without supercomputers (Hajima et al., 2014). The range of EMICs used in climate change research is highly heterogeneous, ranging from zonally averaged or mixed-layer ocean models coupled to statistical-dynamical models of the atmosphere, to low-resolution three-dimensional ocean models coupled to simplified dynamical models of the atmosphere. An increasing number of EMICs include interactive representations of the global carbon cycle, with varying levels of complexity and numbers of processes considered (Plattner et al., 2008; Zickfeld et al., 2013; MacDougall et al., 2020). Given the heterogeneity of the EMIC community, modellers tend to focus on specific research questions and develop individual models accordingly. As for any type of models assessed in this Report, the set of EMICs undergoes thorough evaluation and fit-for-purpose testing before being applied to address specific climate aspects.

EMICs have been used extensively in past IPCC reports, providing long-term integrations on paleoclimate and future time scales, including stabilization pathways and a range of commitment scenarios, with perturbed physics ensembles and sensitivity studies, or with simulations targeting the uncertainty in global climate–carbon cycle systems (e.g., Meehl et al., 2007b; Collins et al., 2013). More recently, a number of studies have pointed to the possibility of systematically different climate responses to external forcings in EMICs and complex ESMs (Frölicher and Paynter, 2015; Pfister and Stocker, 2017, 2018) that need to be considered in the context of this report. For example, Frölicher and Paynter (2015) showed that EMICs have a higher simulated realized warming fraction (i.e., the TCR/ECS ratio) than CMIP5 ESMs and speculated that this may bias the temperature response to zero carbon emissions. But, in a recent comprehensive multi-model analysis of the zero CO₂ emissions commitment, MacDougall et al. (2020) did not find any significant differences between EMICs and ESMs in committed temperatures 90 years after halting emissions. While some EMICs contribute to parts of the CMIP6-endorsed MIPs, a coordinated EMICs modelling effort similar to those carried out for AR4 (Plattner et al., 2008) and AR5 (Eby et al., 2013; Zickfeld et al., 2013) is not in place for IPCC AR6; however, EMICs are assessed in a number of chapters. For example, Chapters 4 and 5 use EMICs in the assessment of long-term climate change beyond 2100 (Section 5.5); zero-emissions commitments, overshoot and recovery (Section 4.7); consequences of CO₂ removal (CDR) on the climate system and the carbon cycle (Sections 4.6 and 5.6); and long-term carbon cycle–climate feedbacks (Section 5.4).

Physical emulators and simple climate modelsmake up a broad class of heavily parametrized models designed to reproduce the responses of the more complex, process-based models, and provide rapid translations of emissions, via concentrations and radiative forcing, into probabilistic estimates of changes to the physical climate system. The main application of emulators is to extrapolate insights from ESMs and observational constraints to a larger set of emissions scenarios (Cross-Chapter Box 7.1). The computational efficiency of various emulating approaches opens new analytical possibilities, given that ESMs take a lot of computational resources for each simulation. The applicability and usefulness of emulating approaches are however constrained by their skill in capturing the global mean climate responses simulated by the ESMs (mainly limited to global mean or hemispheric land/ocean temperatures) and by their ability to extrapolate skilfully outside the calibrated range.

The terms ‘emulator’ and ‘simple climate model’ (SCM) are different, although they are sometimes used interchangeably. SCM refers to a broad class of lower-dimensional models of the energy balance, radiative transfer, carbon cycle, or a combination of such physical components. SCMs can also be tuned to reproduce the calculations of climate-mean variables of a given ESM, assuming that their structural flexibility can capture both the parametric and structural uncertainties across process-oriented ESM responses. When run in this setup, they are termed emulators. Simple climate models do not have to be run in ‘emulation’ mode, though, as they can also be used to test consistency across multiple lines of evidence with regard to ranges in ECS, TCR, TCRE and carbon cycle feedbacks (Chapters 5 and 7). Physical emulation can also be performed with very simple parameterizations (‘one-or-few-line climate models’), statistical methods like neural networks, genetic algorithms, or other artificial intelligence approaches, where the emulator behaviour is explicitly tuned to reproduce the response of a given ESM or model ensemble (Chapters 4, 5 and 7).

Current emulators and SCMs include the generic impulse response model outlined in Chapter 8 of AR5 (AR5-IR; Supplementary Material 8.SM.11 of Myhre et al., 2013), two-layer models (Held et al., 2010; Rohrschneider et al., 2019; Nicholls et al., 2020), and higher-complexity approaches that include upwelling, diffusion and entrainment in the ocean component (e.g., MAGICC Version 5.3 (Raper et al., 2001; Wigley et al., 2009); Version 6/7 (Meinshausen et al., 2011a); OSCAR(Gasser et al., 2017); CICERO SCM(Skeie et al., 2017); FaIR (Millar et al., 2017a; Smith et al., 2018); and a range of statistical approaches (Schwarber et al., 2019; Beusch et al., 2020b). An example of recent use of an emulator approach is an early estimate of the climate implications of the COVID-19 lockdowns (Cross-Chapter Box 6.1; Forster et al., 2020).

Since AR5, simplified climate models have been developed further, and their use is increasing. Different purposes motivating development include: being as simple as possible for teaching purposes (e.g., a two-layer energy balance model); being as comprehensive as possible to allow for propagation of uncertainties across multiple Earth system domains (MAGICC and others); or focusing on higher-complexity representation of specific domains (e.g., OSCAR). The common theme motivating many models is to improve parameterizations that reflect the latest findings in complex ESM interactions – such as the nitrogen cycle addition to the carbon cycle, or tropospheric and stratospheric ozone exchange – with the aim of emulating their global mean temperature response. Also, within the simple models that have a rudimentary representation of spatial heterogeneity (e.g., four-box simple climate models), the ambition is to represent heterogeneous forcers such as black carbon more adequately (Stjern et al., 2017), provide an appropriate representation of the forcing–feedback framework (e.g., Sherwood et al., 2015), investigate new parameterizations of ocean heat uptake, and implement better representations of volcanic aerosol-induced cooling (Gregory et al., 2016a).

MAGICC (Wigley et al., 2009; Meinshausen et al., 2011a) and FaIR (Smith et al., 2018) were used in IPCC SR1.5 (IPCC, 2018) to categorize mitigation pathways into classes of scenarios that peak near 1.5°C, overshoot 1.5°C, or stay below 2°C. The SR1.5 (Rogelj et al., 2018b) concluded that there was high agreement on the relative temperature response of pathways, butmedium agreement on the precise absolute magnitude of warming, introducing a level of imprecision in the attribution of a single pathway to a given category.

In this Report, there are two notable uses of simple climate models. One is the connection between the assessed range of ECS in Chapter 7, and the projections of future global surface air temperature (GSAT) change in Chapter 4, which is done via a two-layer model based on Held et al. (2010). It is also used as input to sea level projections in Chapter 9. The other usage is the transfer of Earth system assessment knowledge to WGIII, via a set of models (MAGICC, FaIR, CICERO-SCM) specifically tuned to represent the WGI assessment. For an overview of the uses, and an assessment of the related Reduced Complexity Model Intercomparison Project (RCMIP), see Nicholls et al. (2020) and Cross-Chapter Box 7.1.

A key issue addressed in this Report is whether climate models are adequate or ‘fit’ for purposes of interest, that is, whether they can be used to successfully answer particular research questions, especially about the causes of recent climate change and the future evolution of climate (e.g., Parker, 2009; Notz, 2015; Knutti, 2018; Winsberg, 2018). Assessment of a model’s fitness-for-purpose can be informed both by how the model represents relevant physical processes and by relevant performance metrics (Baumberger et al., 2017; Parker, 2020). The processes and metrics that are most relevant can vary with the question of interest. For example, a question about changes in deep-ocean circulation compared with a question about changes in regional precipitation (Notz, 2015; Gramelsberger et al., 2020). New model-evaluation tools (Section 1.5.4.5) and emergent constraint methodologies (Section 1.5.4.7) can also aid the assessment of fitness-for-purpose, especially in conjunction with process understanding (Klein and Hall, 2015; Knutti, 2018). The broader availability of large model ensembles may allow for novel tests of fitness that better account for natural climate variability (Section 1.5.4.2). Fitness-for-purpose of models used in this Report is discussed in Chapter 3 (Section 3.8.4) for the global scale, in Chapter 10 (Section 10.3) for regional climate, and in the other chapters for the process level.

Typical strategies for enhancing the fitness-for-purpose of a model include increasing resolution in order to explicitly simulate key processes, improving relevant parameterizations, and careful tuning. Changes to a model that enhance its fitness for one purpose can sometimes decrease its fitness for others, by upsetting a pre-existing balance of approximations. When it is unclear whether a model is fit for a purpose of interest, there is often a closely related purpose for which the evidence of fitness is clearer. For example, it might be unclear whether a model is fit for providing highly accurate projections of precipitation changes in a region, but reasonable to think that the model is fit for providing projections of precipitation changes that cannot yet be ruled out (Parker, 2009). Such information about plausible or credible changes can be useful to inform adaptation. Note that challenges associated with assessing models’ fitness-for-purpose need not prevent reaching conclusions with high confidence if there are multiple other lines of evidence supporting those same conclusions.

A key approach in climate science is the comparison of results from multiple model simulations with each other and against observations. These simulations have typically been performed by separate models with consistent boundary conditions and prescribed emissions or radiative forcings, as in the Coupled Model Intercomparison Project phases (CMIP, Meehl et al., 2000, 2007a; Taylor et al., 2012; Eyring et al., 2016). Such multi-model ensembles (MMEs) have proven highly useful in sampling and quantifying model uncertainty, within and between generations of climate models. They also reduce the influence on projections of the particular sets of parametrizations and physical components simulated by individual models. The primary usage of MMEs is to provide a well-quantified model range, but when used carefully they can also increase confidence in projections (Knutti et al., 2010). Presently, however, many models also share provenance (Masson and Knutti, 2011) and may have common biases that should be acknowledged when presenting and building on MME-derived conclusions (Section 1.5.4.6; Boé, 2018; Abramowitz et al., 2019).

Since AR5, an increase in computing power has made it possible to investigate simulated internal variability and to provide robust estimates of forced model responses, using large initial condition ensembles (ICEs), also referred to as single model initial condition large ensembles (SMILEs). Examples using GCMs or ESMs that support assessments in AR6 include the CESM Large Ensemble (Kay et al., 2015), the MPI Grand Ensemble (Maher et al., 2019), and the CanESM2 large ensembles (Kirchmeier-Young et al., 2017). Such ensembles employ a single GCM or ESM in a fixed configuration, but starting from a variety of different initial states. In some experiments, these initial states only differ slightly. As the climate system is chaotic, such tiny changes in initial conditions lead to different evolutions for the individual realizations of the system as a whole. Other experiments start from a set of well-separated ocean initial conditions to sample the uncertainty in the circulation state of the ocean and its role in longer-time scale variations. These two types of ICEs have been referred to as ‘micro’ and ‘macro’ perturbation ensembles respectively (Hawkins et al., 2016). In support of this Report, most models contributing to CMIP6 have produced ensembles of multiple realizations of their historical and scenario simulations (Chapters 3 and 4).

Recently, the ICE technique has been extended to atmosphere-only simulations (Mizuta et al., 2017), single-forcer influences such as volcanic eruptions (Bethke et al., 2017), regional modelling (Mote et al., 2015; Fyfe et al., 2017; Schaller et al., 2018; Leduc et al., 2019), and to attribution of extreme weather events using crowdsourced computing (climateprediction.net ; Massey et al., 2015).

ICEs can also be used to evaluate climate model parameterizations, if models are initialized appropriately (Phillips et al., 2004; Williams et al., 2013), mostly within the framework of seamless weather and climate predictions (e.g., Palmer et al., 2008; Hurrell et al., 2009; Brown et al., 2012). Initializing an atmospheric model in hindcast mode and observing the biases as they develop permits testing of the parameterized processes, by starting from a known state rather than one dominated by quasi-random short-term variability (Williams et al., 2013; Ma et al., 2014; Vannière et al., 2014). However, single-model initial-conditions ensembles cannot cover the same degrees of freedom as a multi-model ensemble, because model characteristics substantially affect model behaviour (Flato et al., 2013).

A third common modelling technique is the perturbed parameter ensemble (PPE; note that the abbreviation also sometimes refers to the sub-category ‘perturbed physics ensemble’). These methods are used to assess uncertainty based on a single model, with individual parameters perturbed to reflect the full range of their uncertainty (Murphy et al., 2004; Knutti et al., 2010; Lee et al., 2011; Shiogama et al., 2014). Statistical methods can then be used to detect which parameters are the main causes of uncertainty across the ensemble. PPEs have been used frequently in simpler models, such as EMICs, and are being applied to more complex models. A caveat of PPEs is that the estimated uncertainty will depend on the specific parameterizations of the underlying model and may well be an underestimation of the ‘true’ uncertainty. It is also challenging to disentangle forced responses from internal variability using a PPE alone.

Together, the three ensemble methods (MMEs, ICEs, PPEs) allow investigation of climate model uncertainty arising from internal variability, initial and internal boundary conditions, model formulations and parameterizations (Parker, 2013). Figure 1.21 illustrates the different ensemble types. Recent studies have also started combining multiple ensemble types or using ensembles in combination with statistical analytical techniques. For example, Murphy et al. (2018) combine MMEs and PPEs to give a fuller assessment of modelling uncertainty. Wagman and Jackson (2018) use PPEs to evaluate the robustness of MME-based emergent constraints. Sexton et al. (2019) study the robustness of ICE approaches by identifying parameters and processes responsible for model errors at the two different time scales.

Figure 1.21 | Illustration of common types of model ensemble, simulating the time evolution of a quantity Q (such as global mean surface temperature). (a) Multi-model ensemble, where each model has its own realization of the processes affecting Q, and its own internal variability around the baseline value (dashed line). The multi-model mean (black) is commonly taken as the ensemble average. (b) Initial condition ensemble, where several realizations from a single model are compared. These differ only by minute (‘micro’) perturbations to the initial conditions of the simulation, such that over time, internal variability will progress differently in each ensemble member. (c) Perturbed physics ensemble, which also compares realizations from a single model, but where one or more internal parameters that may affect the simulations of Q are systematically changed to allow for a quantification of the impact of those quantities on the model results. Additionally, each parameter set may be taken as the starting point for an initial condition ensemble. In this figure, each set has three ensemble members.

Open figure

Overall, we assess that increases in computing power and the broader availability of larger and more varied ensembles of model simulations have contributed to better estimations of uncertainty in projections of future change (high confidence). Note, however, that despite their widespread use in climate science today, the cost of the ensemble approach in human and computational resources, and the challenges associated with the interpretation of multi-model ensembles, has been questioned (Palmer and Stevens, 2019; Touzé-Peiffer et al., 2020).

The Coupled Model Intercomparison Project (CMIP) provides a framework to compare the results of different GCMs or ESMs performing similar experiments. Since its creation in the mid-1990s, it has evolved in different phases, involving all major climate modelling centres in the world (Figure 1.20). The results of these phases have played a key role in previous IPCC reports, and the present Report assesses a range of results from CMIP5 that were not published until after the AR5, as well as the first results of the 6th phase of CMIP (CMIP6; Eyring et al., 2016). The CMIP6 experiment design is somewhat different from previous phases. It now consists of a limited set of DECK (Diagnostic, Evaluation and Characterization of Klima) simulations and an historical simulation that must be performed by all participating models, as well as a wide range of CMIP6-Endorsed model intercomparison projects (MIPs) covering specialized topics (Figure 1.22; Eyring et al., 2016). Each MIP activity consists of a series of model experiments, documented in the literature (Table 1.3) and in an online database (es-doc.org; Annex II; Pascoe et al., 2020).

Figure 1.22 | Structure of CMIP6, the 6th phase of the Coupled Model Intercomparison Project . The centre shows the common DECK (Diagnostic, Evaluation and Characterization of Klima) and historical experiments that all participating models must perform. The outer circles show the topics covered by the endorsed (red) and other MIPs (orange). See Table 1.3 for explanation of the MIP acronyms. Figure is adapted from Eyring et al. (2016).

Open figure

Table 1.3 | CMIP6-Endorsed MIPs, their key references, and where they are used or referenced throughout this Report.

The CMIP DECK simulations form the basis for a range of assessments and projections in the following chapters. As in CMIP5, they consist of: a ‘pre-industrial’ control simulation (piControl, where ‘pre-industrial’ is taken as fixed 1850 conditions in these experiments); an idealized, abrupt quadrupling of CO₂ concentrations relative to piControl (to estimate equilibrium climate sensitivity); a 1% per year increase in CO₂ concentrations relative to piControl (to estimate the transient climate response); and a transient simulation with prescribed sea-surface temperatures for the period 1979–2014 (termed ‘AMIP’ for historical reasons). In addition, all participating models perform a historical simulation for the period 1850–2014. For the latter, common CMIP6 forcings are prescribed (Cross-Chapter Box 1.4, Table 2). Depending on the model setup, these include emissions and concentrations of short-lived species (Hoesly et al., 2018; Gidden et al., 2019), long-lived GHGs (Meinshausen et al., 2017), biomass burning emissions (van Marle et al., 2017), global gridded land-use forcing data (Ma et al., 2020), solar forcing (Matthes et al., 2017), and stratospheric aerosol data from volcanoes (Zanchettin et al., 2016). The methods for generating gridded datasets are described in Feng et al. (2020). For AMIP simulations, common sea surface temperatures (SSTs) and sea ice concentrations (SICs) are prescribed. For simulations with prescribed aerosol abundances (i.e., not calculated from emissions), optical properties and fractional changes in cloud droplet effective radius are generally prescribed in order to provide a more consistent representation of aerosol forcing relative to earlier CMIP phases (Fiedler et al., 2017; Stevens et al., 2017). For models without ozone chemistry, time-varying gridded ozone concentrations and nitrogen deposition are also provided (Checa-Garcia et al., 2018).

Beyond the DECK and the historical simulations, the CMIP6-Endorsed MIPs aim to investigate how models respond to specific forcings, their potential systematic biases, their variability, and their responses to detailed future scenarios such as the Shared Socio-economic Pathways (SSPs; Section 1.6). Table 1.3 lists the 23 CMIP6-Endorsed MIPs and key references. Results from a range of these MIPs, and many others outside of the most recent CMIP6 cycle, will be assessed in the following chapters (also shown in Table 1.3). References to all the CMIP6 datasets used in the report are found in Annex II, Table AII.10.

The Coordinated Regional Downscaling Experiment (CORDEX; Gutowski Jr. et al., 2016) is an intercomparison project for regional models and statistical downscaling techniques, coordinating simulations on common domains and under common experimental conditions in a similar way to the CMIP effort. Dynamical and statistical downscaling techniques can provide higher-resolution climate information than is available directly from global climate models (Section 10.3). These techniques require evaluation and quantification of their performance before they can be considered appropriate as usable regional climate information or be used in support of climate services. CORDEX simulations have been provided by a range of regional downscaling models for 14 regions, together covering much of the globe (Figure Atlas.7), and they are used extensively in the AR6 WGI Atlas (Atlas.1.4 and Annex II).

In support of AR6, CORDEX has undertaken a new experiment (CORDEX-CORE) in which regional climate models downscale a common set of global model simulations, performed at a coarser resolution, to a spatial resolution spanning from 12–25 km over most of the CORDEX domains (Box Atlas.1). CORDEX-CORE represents an improved level of coordinated intercomparison of downscaling models (Remedio et al., 2019).

For the first time in CMIP, a range of comprehensive evaluation tools are now available that can run alongside the commonly used distributed data platform – Earth System Grid Federation (ESGF; see Annex II) – to produce comprehensive results as soon as the model output is published to the CMIP archive.

For instance, the Earth System Model Evaluation Tool (ESMValTool; Eyring et al., 2020; Lauer et al., 2020; Righi et al., 2020) is used by a number of chapters. It is an open-source community software tool that includes a large variety of diagnostics and performance metrics relevant for coupled Earth system processes, such as for the mean, variability and trends, and it can also examine emergent constraints (Section 1.5.4.7). ESMValTool also includes routines provided by the WMO Expert Team on Climate Change Detection and Indices for the evaluation of extreme events (Min et al., 2011; Sillmann et al., 2013) and diagnostics for key processes and variability. Another example of an evaluation tool is the CLIVAR 2020 ENSO metrics package (Planton et al., 2021).

These tools are used in several chapters of this report for the creation of the figures that show CMIP results. Together with the Interactive Atlas, they allow for traceability of key results, and an additional level of quality control on whether published figures can be reproduced. It also provides the capability to update published figures with, as much as possible, the same set of models in all figures, and to assess model improvements across different phases of CMIP (Section 3.8.2).

These new developments are facilitated by the definition of common formats for CMIP model output (Balaji et al., 2018) and the availability of reanalyses and observations in the same format as CMIP output (obs4MIPs; Ferraro et al., 2015). The tools are also used to support routine evaluation at individual model centres and simplify the assessment of improvements in individual models or generations of model ensembles (Eyring et al., 2019). Note, however, that while tools such as ESMValTool can produce an estimate of overall model performance, dedicated model evaluation still needs to be performed when analysing projections for a particular purpose, such as assessing changing hazards in a given region. Such evaluation is discussed in the next section, and in greater detail in later chapters of this Report.

Techniques used for evaluating process-based climate models against observations were assessed in AR5 (Flato et al., 2013), and have progressed rapidly since (Eyring et al., 2019). The most widely used technique is to compare climatologies (long-term averages of specific climate variables) or time series of simulated (process-based) model output with observations, considering the observational uncertainty. A further approach is to compare the results of process-based models with those from statistical models. In addition to a comparison of climatological means, trends and variability, AR5 already made use of a large set of performance metrics for a quantitative evaluation of the models.

Since AR5, a range of studies has investigated model agreement with observations well beyond large-scale mean climate properties (e.g., Bellenger et al., 2014; Covey et al., 2016; Pendergrass and Deser, 2017; Goelzer et al., 2018; Beusch et al., 2020a), providing information on the performance of recent model simulations across multiple variables and components of the Earth system (e.g., Anav et al., 2013; Guan and Waliser, 2017). Based on such studies, this Report assesses model improvements across different CMIP DECK, CMIP6 historical and CMIP6-Endorsed MIP simulations, and of differences in model performance between different classes of models, such as high- versus low-resolution models (see e.g., Section 3.8.2).

In addition, process- or regime-oriented evaluation of models has been expanded since AR5. By focusing on processes, causes of systematic errors in the models can be identified and insights can be gained as to whether a mean state or trend is correctly simulated and for the right reasons. This approach is commonly used for the evaluation of clouds (e.g., Williams and Webb, 2009; Konsta et al., 2012; Bony et al., 2015; Dal Gesso et al., 2015; Jin et al., 2017), dust emissions (e.g., Parajuli et al., 2016; Wu et al., 2016) as well as aerosol–cloud (e.g., Gryspeerdt and Stier, 2012) and chemistry–climate (SPARC, 2010) interactions. Process-oriented diagnostics have also been used to evaluate specific phenomena such as the El Niño–Southern Oscillation (ENSO; Guilyardi et al., 2016), the Madden–Julian Oscillation (MJO; Ahn et al., 2017; Jiang et al., 2018), Southern Ocean clouds (Hyder et al., 2018), monsoons (Boo et al., 2011; James et al., 2015) and tropical cyclones (Kim et al., 2018).

Instrument simulators provide estimates of what a satellite would see if looking down on the model-simulated planet, and improve the direct comparison of modelled variables such as clouds, precipitation and upper tropospheric humidity with observations from satellites (e.g., Kay et al., 2011; Klein et al., 2013; Cesana and Waliser, 2016; Konsta et al., 2016; Jin et al., 2017; Chepfer et al., 2018; Swales et al., 2018; Zhang et al., 2018). Within the framework of the Cloud Feedback Model Intercomparison Project (CFMIP) contribution to CMIP6 (Webb et al., 2017), a new version of the Cloud Feedback Model Intercomparison Project Observational Simulator (COSP; Swales et al., 2018) has been released which makes use of a collection of observation proxies or satellite simulators. Related approaches in this rapidly evolving field include simulators for Arctic Ocean observations (Burgard et al., 2020) and measurements of aerosol observations along aircraft trajectories (Watson-Parris et al., 2019).

In this Report, model evaluation is performed in the individual chapters, rather than in a separate chapter as was the case for AR5. This applies to the model types discussed above, and also to dedicated models of subsystems that are not (or not yet) part of usual climate models, for example, glacier or ice-sheet models (Annex II). Further discussions are found in Chapter 3 (attribution), Chapter 5 (carbon cycle), Chapter 6 (short-lived climate forcers), Chapter 8 (water cycle), Chapter 9 (ocean, cryosphere and sea level), Chapter 10 (regional scale information) and the Atlas (regional models).

An emergent constraint is the relationship between an uncertain aspect of future climate change and an observable feature of the Earth System, evident across an ensemble of models (Allen and Ingram, 2002; Mystakidis et al., 2016; Wenzel et al., 2016; Hall et al., 2019; Winkler et al., 2019). Complex Earth system models (ESMs) simulate variations on time scales from hours to centuries, telling us how aspects of the current climate relate to its sensitivity to anthropogenic forcing. Where an ensemble of different ESMs displays a relationship between a short-term observable variation and a longer-term sensitivity, an observation of the short-term variation in the real world can be converted, via the model-based relationship, into an ‘emergent constraint’ on the sensitivity. This is shown schematically in Figure 1.23 (see Glossary; Eyring et al., 2019).

Figure 1.23 | The principle of emergent constraints. An ensemble of models (blue dots) defines a relationship between an observable mean, trend or variation in the climate (x-axis) and an uncertain projection, climate sensitivity or feedback (y-axis). An observation of the x-axis variable can then be combined with the model-derived relationship to provide a tighter estimate of the climate projection, sensitivity or feedback on the y-axis. Figure adapted from Eyring et al. (2019).

Open figure

Emergent constraints use the spread in model projections to estimate the sensitivities of the climate system to anthropogenic forcing, providing another type of ensemble-wide information that is not readily available from simulations with one ESM alone. As emergent constraints depend on identifying those observable aspects of the climate system that are most related to climate projections, they also help to focus model evaluation on the most relevant observations (Hall et al., 2019). However, there is a chance that indiscriminate data-mining of the multi-dimensional outputs from ESMs could lead to spurious correlations (Caldwell et al., 2014; Wagman and Jackson, 2018) and less-than-robust emergent constraints on future changes (Bracegirdle and Stephenson, 2013). To avoid this, emergent constraints need to be tested ‘out of sample’ on parts of the dataset that were not included in its construction (Caldwell et al., 2018) and should also always be based on sound physical understanding and mathematical theory (Hall et al., 2019). Their conclusions should also be reassessed when a new generation of MMEs becomes available, such as CMIP6. As an example, Chapter 7 (Section 7.5.4) discusses and assesses recent studies where equilibrium climate sensitivities (ECS) diagnosed in a multi-model ensemble are compared with the same models’ estimates of an observable quantity, such as post-1970s global warming or tropical sea surface temperatures of past climates like the Last Glacial Maximum or the Pliocene. Assessments of other emergent constraints appear throughout later chapters, such as Chapter 4 (Section 4.2.5), Chapter 5 (Section 5.4.6) and Chapter 7 (Section 7.5.4).

Assessments of climate model ensembles have commonly assumed that each individual model is of equal value (‘model democracy’) and when combining simulations to estimate the mean and variance of quantities of interest, they are typically unweighted (Haughton et al., 2015). This practice has been noted to diminish the influence of models exhibiting a good match with observations (Tapiador et al., 2020). However, exceptions to this approach exist, notably AR5 projections of sea ice, which only selected a few models which passed a model performance assessment (Collins et al., 2013), and more studies on this topic have appeared since AR5 (e.g., Eyring et al., 2019). Ensembles are typically sub-selected by removing either poorly performing model simulations (McSweeney et al., 2015) or model simulations that are perceived to add little additional information, typically where multiple simulations have come from the same model. They may also be weighted based on model performance.

Several recent studies have attempted to quantify the effect of various strategies for selection or weighting of ensemble members based on some set of criteria (Haughton et al., 2015; Olonscheck and Notz, 2017; Sanderson et al., 2017). Model weighting strategies have been further employed since AR5 to reduce the spread in climate projections for a given scenario by using weights based on one or more model performance metrics (Wenzel et al., 2016; Knutti et al., 2017; Sanderson et al., 2017; Lorenz et al., 2018; Liang et al., 2020). However, models may share representations of processes, parameterization schemes, or even parts of code, leading to common biases. The models may therefore not be fully independent, calling into question inferences derived from multi-model ensembles (Abramowitz et al., 2019). Emergent constraints (Section 1.5.4.5) also represent an implicit weighting technique that explicitly links present performance to future projections (Bracegirdle and Stephenson, 2013).

Concern has been raised about the large extent to which code is shared within the CMIP5 multi-model ensemble (Sanderson et al., 2015a). Boé (2018) showed that a clear relationship exists between the number of components shared by climate models and how similar the simulations are. The resulting similarities in behaviour need to be accounted for in the generation of best-estimate multi-model climate projections. This has led to calls to move beyond equally-weighted multi-model means towards weighted means that take into account both model performance and model independence (Sanderson et al., 2015b, 2017; Knutti et al., 2017). Model independence has been defined in terms of performance differences within an ensemble (Masson and Knutti, 2011; Knutti et al., 2013, 2017, Sanderson et al., 2015a, b, 2017; Lorenz et al., 2018). However, this definition is sensitive to the choice of variable, observational dataset, metric, time period, and region, and a performance-ranked ensemble has been shown to sometimes perform worse than a random selection (Herger et al., 2018a). The adequacy of the constraint provided by the data and experimental methods can be tested using a ‘calibration-validation’ style partitioning of observations into two sets (Bishop and Abramowitz, 2013), or a ‘perfect model approach’ where one of the ensemble members is treated as the reference dataset and all model weights are calibrated against it (Bishop and Abramowitz, 2013; Wenzel et al., 2016; Knutti et al., 2017; Sanderson et al., 2017; Herger et al., 2018a, b). Sunyer et al. (2014) use a Bayesian framework to account for model dependencies and changes in model biases. Annan and Hargreaves (2017) provides a statistical, quantifiable definition of independence that is independent of performance-based measures.

The AR5 quantified uncertainty in CMIP5 climate projections by selecting one realization per model per scenario, and calculating the 5–95% range of the resulting ensemble (Box 4.1) and the same strategy is generally still used in AR6. Broadly, the following chapters take the CMIP6 5–95% ensemble range as the likely uncertainty range for projections, ⁸ with no further weighting or consideration of model ancestry and as long as no universal, robust method for weighting a multi-model projection ensemble is available (Box 4.1). A notable exception to this approach is the assessment of future changes in global surface air temperature (GSAT), which also draws on the updated best estimate and range of equilibrium climate sensitivity assessed in Chapter 7. For a thorough description of the model-weighting choices made in this Report, and the assessment of GSAT, see Chapter 4 (Box 4.1). Model selection and weighting in downscaling approaches for regional assessment is discussed in Chapter 10 (Section 10.3.4).

The Shared Socio-economic Pathways SSP1 to SSP5 describe a range of plausible trends in the evolution of society over the 21st century. They were developed in order to connect a wide range of research communities (Nakicenovic et al., 2014) and consist of two main elements: a set of qualitative, narrative storylines describing societal futures (O’Neill et al., 2017a) and a set of quantified measures of development at aggregated and/or spatially resolved scales. Each pathway is an internally consistent, plausible and integrated description of a socio-economic future, but these socio-economic futures do not account for the effects of climate change, and no new climate policies are assumed. The SSPs’ quantitative projections of socio-economic drivers include population, gross domestic product (GDP) and urbanization (Dellink et al., 2017; Jiang and O’Neill, 2017; Samir and Lutz, 2017). By design, the SSPs differ in terms of the socio-economic challenges they present for climate change mitigation and adaptation (Rothman et al., 2014; Schweizer and O’Neill, 2014) and the evolution of these drivers within each SSP reflects this design. Broadly, the five SSPs represent ‘sustainability’ (SSP1), a ‘middle-of-the-road’ path (SSP2), ‘regional rivalry’ (SSP3), ‘inequality’ (SSP4), and ‘fossil fuel-intensive’ development (SSP5; Cross-Chapter Box 1.4, Figure 1; O’Neill et al., 2017a). More specific information on the SSP framework and the assumptions underlying the SSPs will be provided in the IPCC WGIII report (WGIII Chapter 3; see also Box SPM.1 in SRCCL (IPCC, 2019d)).

The SSP narratives and drivers were used to develop scenarios of energy use, air pollution control, land use, and GHG emissions developments using integrated assessment models (IAMs; Riahi et al., 2017; Rogelj et al., 2018a). An IAM can derive multiple emissions futures for each socio-economic development pathway, assuming no new mitigation policies or various levels of additional mitigation action (in the case of reference scenarios and mitigation scenarios, respectively; Riahi et al., 2017). By design, the evolution of drivers and emissions within the SSP scenarios do not take into account the effects of climate change.

The SSPX-Y scenarios and the RCP scenarios are categorized similarly, by reference to the approximate radiative forcing levels each one entails at the end of the 21st century. For example, the ‘1.9’ in the SSP1-1.9 scenario stands for an approximate radiative forcing level of 1.9 W m^–2 in 2100. The first number (X) in the ‘SSPX-Y’ acronym refers to one of the five shared socio-economic development pathways (Cross-Chapter Box 1.4, Figure 1 and Table 1.4).

Table 1.4 | Overview of different RCP and SSP acronyms as used in this report.

This SSP scenario categorization, focused on end-of-century radiative forcing levels, reflects how scenarios were conceptualized until recently, namely, to reach a particular climate target in 2100 at the lowest cost and irrespective of whether the target was exceeded over the century. More recently, and in particular since IPCC SR1.5 report focused attention on peak warming scenarios (Rogelj et al., 2018b), scenario development started to explicitly consider peak warming, cumulative emissions and the amount of net negative emissions (Rogelj et al., 2018b; Fujimori et al., 2019).

The SSP scenarios can be used for either emissions- or concentration-driven model experiments (Cross-Chapter Box 1.4). ESMs can be run with emissions and concentrations data for GHGs and aerosols and land-use or landcover maps and calculate levels of radiative forcing internally. The radiative forcing labels of the RCP and SSP scenarios, such as ‘2.6’ in RCP2.6 or SSP1-2.6, are thus approximate labels for the year 2100 only. The actual global mean effective radiative forcing varies across ESMs due to different radiative transfer schemes, uncertainties in aerosol–cloud interactions, and different feedback mechanisms, among other reasons. Nonetheless, using approximate radiative forcing labels is advantageous because it establishes a clear categorization of scenarios, with multiple climate forcings and different combinations in those scenarios summarized in a single number. The classifications according to cumulative carbon emissions (Section 1.6.3) and global warming level (Section 1.6.2 and Cross-Chapter Box 7.1 on emulators) complement those forcing labels.

A key advance of the SSP scenarios relative to the RCPs is a wider span of assumptions on future air-quality mitigation measures, and hence emissions of short-lived climate forcers (SLCFs; Rao et al., 2017; Lund et al., 2020). This allows for a more detailed investigation into the relative roles of GHG and SLCF emissions in future global and regional climate change, and hence the implications of policy choices. For instance, SSP1-2.6 builds on an assumption of stringent air-quality mitigation policy, leading to rapid reductions in particle emissions, while SSP3-7.0 assumes slow improvements, with pollutant emissions over the 21st century comparable to current levels (Figure 6.19 and Cross-Chapter Box 1.4, Figure 2).

One limitation of the SSP scenarios used for CMIP6 and in this Report is that they reduce emissions from all the major ozone-depleting substances controlled under the Montreal Protocol (CFCs, halons, and hydrochlorofluorocarbons (HCFCs)) uniformly, rather than representing a fuller range of possible high- and low-emissions futures (UNEP, 2016). Hydrofluorocarbon (HFC) emissions, on the other hand, span a wider range within the SSPs than in the RCPs (Cross-Chapter Box 1.4, Figure 2).

The SSP scenarios and previous RCP scenarios are not directly comparable. First, the gas-to-gas compositions differ; for example, the SSP5-8.5 scenario has higher CO₂ concentrations but lower CH₄ concentrations compared to RCP8.5. Second, the projected 21st-century trajectories may differ, even if they result in the same radiative forcing by 2100. Third, the overall effective radiative forcing (Chapter 7) may differ, and tends to be higher for the SSPs compared to RCPs that share the same nominal stratospheric-temperature-adjusted radiative forcing label. The stratospheric-temperature-adjusted radiative forcings of the SSPs and RCPs, however, remain relatively close, at least by 2100 (Tebaldi et al., 2021). In summary, differences in, for example, CMIP5 RCP8.5 and CMIP6 SSP5-8.5 ESM outputs, are partially due to different scenario characteristics rather than different ESM characteristics only (Section 4.6.2).

When investigating various mitigation futures, WGIII goes beyond the core set of SSP scenarios assessed in WGI (SSP1-1.9, SSP1-2.6, etc.) to consider the characteristics of more than 1000 scenarios (Cross-Chapter Box 7.1). In addition, while staying within the framework of socio-economic development pathways (SSP1 to SSP5), WGIII also considers various mitigation possibilities through so-called illustrative pathways (IPs). These illustrative pathways help to highlight key narratives in the literature concerning various technological, social and behavioural options for mitigation, various timings for implementation, or varying emphasis on different GHG and land-use options. Just as with the SSPX-Y scenarios considered in this Report, these illustrative pathways can be placed in relation to the matrix of SSP families and approximate radiative forcing levels in 2100 (Cross-Chapter Box 1.4, Figure 1; IPCC WGIII, Chapter 3).

No likelihood is attached to the scenarios assessed in this report, and the feasibility of specific scenarios in relation to current trends is best informed by the WGIII contribution to AR6. In the scenario literature, the plausibility of the high emissions levels underlying scenarios such as RCP8.5 or SSP5-8.5 has been debated in light of recent developments in the energy sector (Section 1.6.1.4).

The scenario generation process involves research communities linked to all three IPCC Working Groups (Figure 1.27). It generally starts in the scientific communities associated with WGII and WGIII with the definition of new socio-economic scenario storylines (IPCC, 2000; O’Neill et al., 2014) that are quantified in terms of their drivers – i.e., GDP, population, technology, energy and land use – and their resulting emissions (Riahi et al., 2017). Then, numerous complementation and harmonization steps are necessary for datasets within the WGI and WGIII science communities, including gridding emissions of anthropogenic short-lived forcers, providing open biomass-burning emissions estimates, preparing land-use patterns, aerosol fields, stratospheric and tropospheric ozone, nitrogen deposition datasets, solar irradiance and aerosol optical property estimates, and observed and projected GHG concentration time series (documented for CMIP6 through input4mips; Cross-Chapter Box 1.4, Table 2; Durack et al., 2018).

Figure 1.27 | A simplified illustration of the scenario generation process, involving the scientific communities represented in the three IPCC Working Groups. The circular set of arrows at the top indicates the main set of models and workflows used in the scenario generation process, with the lower level indicating the datasets.

Open figure

Once these datasets are completed, ESMs are run in coordinated model intercomparison projects in the WGI science community, using standardized simulation protocols and scenario data. The most recent example of such a coordinated effort is the CMIP6 exercise (Section 1.5.4; Eyring et al., 2016) with, in particular, ScenarioMIP (O’Neill et al., 2016). The WGI science community feeds back climate information to WGIII via climate emulators (Cross-Chapter Box 7.1) that are updated and calibrated with the ESMs’ temperature responses and other lines of evidence. Next, this climate information is used to compute several high-level global climate indicators (e.g., atmospheric concentrations, global temperatures) for a much wider set of hundreds of scenarios that are assessed as part of the IPCC WGIII Assessment (WGIII Annex C). The outcomes from climate models run under the different scenarios are then used to calculate the evolution of climatic impact-drivers (Chapter 12), and utilized by impact researchers together with exposure and vulnerability information, in order to characterize risk to human and natural systems from future climate change. The climate impacts associated with these scenarios or different warming levels are then assessed as part of WGII reports (Figure 1.27).

Scenario modelling experiments have been a core element of physical climate science since the first transient simulations with a general circulation model in 1988 (Section 1.3; Hansen et al., 1988). Scenarios and modelling experiments assessed in IPCC reports have evolved over time, which provides a ‘history of how the future was seen’. The starting time for the scenarios moves as actual emissions supersede earlier emissions assumptions, while new scientific insights into the range of plausible population trends, behavioural changes and technology options and other key socio-economic drivers of emissions also emerge (see WGIII; Leggett et al., 1992; IPCC, 2000; Moss et al., 2010; Riahi et al., 2017). Many different sets of climate projections have been produced over the past several decades, using different sets of scenarios. Here, we compare those earlier scenarios against the most recent ones.

Figure 1.28 | Comparison of the range of fossil fuel and industrial CO₂ emissions from scenarios used in previous assessments up to AR6. Previous assessments are the IS92 scenarios from 1992 (top), the Special Report on Emissions Scenarios (SRES) scenarios from the year 2000 (second panel), the Representative Concentration Pathway (RCP) scenarios designed around 2010 (third panel) and the Shared Socio-economic Pathways (SSP) scenarios (fourth panel). In addition, historical emissions are shown (black line; Figure 5.5); a more complete set of scenarios is assessed in SR1.5 (bottom); (Huppmann et al., 2018). Further details on data sources and processing are available in the chapter data table (Table 1.SM.1).

Open figure

Climate science research involving scenarios necessarily follows a series of consecutive steps (Figure 1.27). As each step waits for input from the preceding one, delays often occur that result in the impact literature basing its analyses on earlier scenarios than those most current in the climate change mitigation and climate system literature. It is therefore important to provide an approximate comparison across the various scenario generations (Chapter 4, Figure 1.28, and Cross-Chapter Box 1.4, Table 1).

The first widely used set of IPCC emissions scenarios was the IS92 scenarios in 1992 (Leggett et al., 1992). Apart from reference scenarios, IS92 also included a set of stabilization scenarios, the so-called ‘S’ scenarios. Those ‘S’ pathways were designed to lead to CO₂ stabilization levels such as 350 ppm or 450 ppm. By 1996, those latter stabilization levels were complemented in the scientific literature by alternative trajectories that assumed a delayed onset of climate change mitigation action (Figure 1.28; Wigley et al., 1996).

By 2000, the IPCC Special Report on Emissions Scenarios (SRES) produced the SRES scenarios (IPCC, 2000), albeit without assuming any climate policy-induced mitigation. The four broad groups of SRES scenarios (scenario ‘families’) – A1, A2, B1 and B2 – were the first scenarios to emphasize socio-economic scenario storylines, and also first to emphasize other GHGs, land-use change and aerosols. Represented by three scenarios for the high-growth A1 scenario family, those 6 SRES scenarios (A1FI, A1B, A1T, A2, B1, and B2) can still sometimes be found in today’s climate impact literature. The void of missing climate change mitigation scenarios was filled by a range of community exercises, including the so-called ‘post-SRES scenarios’ (Swart et al., 2002).

The RCP scenarios (van Vuuren et al., 2011) then broke new ground by providing low-emissions pathways that implied strong climate change mitigation, including an example with negative CO₂ emissions on a large scale, namely RCP2.6. As shown in Figure 1.28, the upper end of the scenario range has not substantially shifted. Building on the SRES multi-gas scenarios, the RCPs include time series of emissions and concentrations of the full suite of GHGs, aerosols and chemically active gases, as well as land use and land cover (Moss et al., 2010). The word ‘representative’ signifies that each RCP is only one of many possible scenarios that would lead to the specific radiative forcing characteristics. The term ‘pathway’ emphasizes that not only the long-term concentration levels are of interest, but also the trajectory taken over time to reach that outcome (Moss et al., 2010). RCPs usually refer to the concentration pathway extending to 2100, for which IAMs produced corresponding emissions scenarios. Four RCPs produced from IAMs were selected from the published literature and are used in AR5 as well as in this report, spanning approximately the range from below 2°C warming to high (above 4°C) warming best-estimates by the end of the 21st century: RCP2.6, RCP4.5 and RCP6.0 and RCP8.5 (Cross-Chapter Box 1.4, Table 1). Extended Concentration Pathways (ECPs) describe extensions of the RCPs from 2100 to 2300 that were calculated using simple rules generated by stakeholder consultations; these do not represent fully consistent scenarios (Meinshausen et al., 2011b).

By design, theRCP emissions and concentrations pathways were originally developed using particular socio-economic development pathways, but those are no longer considered (Moss et al., 2010). The different levels of emissions and climate change represented in the RCPs can hence be explored against the backdrop of different socio-economic development pathways (SSP1 to SSP5; Section 1.6.1.1 and Cross-Chapter Box 1.4). This integrative SSP-RCP framework (‘SSPX-RCPY’ in Table 1.4) is now widely used in the climate impact and policy analysis literature (e.g., ICONICS, 2021; Green et al., 2020; O’Neill et al., 2020), where climate projections obtained under the RCP scenarios are analysed against the backdrop of various SSPs. Considering various levels of future emissions and climate change for each socio-economic development pathway was an evolution from the previous SRES framework (IPCC, 2000), in which socio-economic and emissions futures were closely aligned.

The new set of scenarios (SSP1-1.9 to SSP5-8.5) now features a higher top level of CO₂ emissions (SSP5-8.5 compared to RCP8.5), although the most significant change is again the addition of a very low climate change mitigation scenario (SSP1-1.9, compared to the previous low scenario, RCP2.6). Also, historically, none of the previous scenario sets featured a scenario that involves a very pronounced peak-and-decline emissions trajectory, but SSP1-1.9 does so now. The full set of nine SSP scenarios now includes a high-aerosol-emissions scenario (SSP3-7.0). The RCPs featured more uniformly low aerosol trajectories across all scenarios (Cross-Chapter Box 1.4, Figure 2). More generally, the SSP scenarios feature a later peak of global emissions for the lower scenarios, simply as a consequence of historical emissions not having followed the trajectory projected by previous low scenarios (Figure 1.28).

Over the last decades, discussions around scenarios have often focussed on whether recent trends make certain future scenarios more or less probable or whether all scenarios are too high or too low. When the SRES scenarios first appeared, the debate was often whether the scenarios were overestimating actual world emissions developments (e.g., Castles and Henderson, 2003). With the strong emissions increase throughout the 2000s, that debate then shifted towards the question of whether the lower future climate change mitigation scenarios were rendered unfeasible (Pielke et al., 2008; van Vuuren and Riahi, 2008). Historical emissions between 2000 and 2010 approximately track the upper half of SRES and RCP projections (Figure 1.28). More generally, the global fossil fuel and industrial CO₂ emissions of recent decades tracked approximately the middle of the projected scenario ranges (Figure 1.28), although with regional differences (Pedersen et al., 2020).

In general, no likelihood is attached to the scenarios assessed in this Report. The use of different scenarios for climate change projections allows the exploration of ‘scenario uncertainty’ (Section 1.4.4; SR1.5; Collins et al., 2013). Scenario uncertainty is fundamentally different from geophysical uncertainties, which result from limitations in the understanding and predictability of the climate system (Smith and Stern, 2011). In scenarios, by contrast, future emissions depend to a large extent on the collective outcome of choices and processes related to population dynamics and economic activity, or on choices that affect a given activity’s energy and emissions intensity (Jones, 2000; Knutti et al., 2008; Kriegler et al., 2012; van Vuuren et al., 2014). Even if identical socio-economic futures are assumed, the associated future emissions still face uncertainties, since different experts and model frameworks diverge in their estimates of future emissions ranges (Ho et al., 2019).

When exploring various climate futures, scenarios with no, or no additional, climate policies are often referred to as ‘baseline’ or ‘reference scenarios’ (Section 1.6.1.1 and Glossary). Among the five core scenarios used most in this report, SSP3-7.0 and SSP5-8.5 are explicit ‘no-climate-policy’ scenarios (Cross-Chapter Box 1.4, Table 1; Gidden et al., 2019), assuming a carbon price of zero. These future ‘baseline’ scenarios are hence counterfactuals that include fewer climate policies compared to ‘business-as-usual’ scenarios – given that ‘business-as-usual’ scenarios could be understood to imply a continuation of existing climate policies. Generally, future scenarios are meant to cover a broad range of plausible futures, due, for example to unforeseen discontinuities in development pathways (Raskin and Swart, 2020), or to large uncertainties in underlying long-term projections of economic drivers (Christensen et al., 2018). However, the likelihood of high-emissions scenarios such as RCP8.5 or SSP5-8.5 is considered low in light of recent developments in the energy sector (Hausfather and Peters, 2020a, b). Studies that consider possible future emissions trends in the absence of additional climate policies, such as the recent IEA 2020 World Energy Outlook ‘stated policy’ scenario (IEA, 2020), project approximately constant fossil fuel and industrial CO₂ emissions out to 2070, approximately in line with the intermediate RCP4.5, RCP6.0 and SSP2-4.5 scenarios (Hausfather and Peters, 2020b) and the 2030 global emissions levels that are pledged as part of the Nationally Determined Contributions (NDCs) under the Paris Agreement (Section 1.2.2; Fawcett et al., 2015; Rogelj et al., 2016; UNFCCC, 2016; IPCC, 2018). On the other hand, the default concentrations aligned with RCP8.5 or SSP5-8.5 and resulting climate futures derived by ESMs could be reached by lower emissions trajectories than RCP8.5 or SSP5-8.5. That is because the uncertainty range on carbon cycle feedbacks includes stronger feedbacks than assumed in the default derivation of RCP8.5 and SSP5-8.5 concentrations (Section 5.4; Ciais et al., 2013; Friedlingstein et al., 2014; Booth et al., 2017).

To address long-term scenario uncertainties, scenario storylines (or ‘narratives’) are often used (see Section 1.4.4 for a more general discussion on ‘storylines’, also covering ‘physical climate storylines’; Rounsevell and Metzger, 2010; O’Neill et al., 2014). Scenario storylines are descriptions of a future world, and the related large-scale socio-economic development pathways towards that world that are deemed plausible within the current state of knowledge and historical experience (Section 1.2.3; WGIII). Scenario storylines attempt to ‘stimulate, provoke, and communicate visions of what the future could hold for us’ (Rounsevell and Metzger, 2010) in settings where either limited knowledge or inherent unpredictability in social systems prevent a forecast or numerical prediction. Scenario storylines have been used in previous climate research, and they are the explicit or implicit starting point of any scenario exercise, including for the SRES scenarios (IPCC, 2000) and the SSPs (e.g., O’Neill et al., 2017a).

Recent technological or socio-economic trends might be informative for bounding near-term future trends, for example, if technological progress renders a mitigation technology cheaper than previously assumed. However, short-term emissions trends alone do not generally rule out an opposite trend in the future (van Vuuren et al., 2010). The ranking of individual RCP emissions scenarios from the IAMs with regard to emissions levels is different for different time horizons, for example, 2020 compared with longer-term emissions levels. For example, the strongest climate change mitigation scenario, RCP2.6, was in fact the second highest CO₂ emissions scenario (jointly with RCP4.5) before 2020 in the set of RCPs and the strong global emissions decline in RCP2.6 only followed after 2020. Implicitly, this scenario feature was cautioning against the assumption that short-term trends predicate particular long-term trajectories. This is also the case in relation to the COVID-19 related drop in 2020 emissions. Potential changes in underlying drivers of emissions, such as those potentially incentivized by COVID-19 recovery stimulus packages, are more significant for longer-term emissions than the short-term deviation from recent emissions trends (Cross-Chapter Box 6.1 on COVID-19).