2.1. Observed and attributed changes in global indicators of the state of climate

In 2024, for the first time in instrumental records, estimates of global surface temperature exceeded 1.5 °C above 1850–1900 (Figure 2), a reference period often used to characterize pre-industrial conditions (WMO, 2026). This new temperature record in both land surface air temperature and sea surface temperature reflects internal climate variability on top of the long-lasting heating of the climate system resulting from human activities (WMO, 2026; Terhaar et al., 2025). Indeed, the internal variability in any single year is estimated to be ±0.25 °C (5–95% range) based on CMIP6 climate models, and ±0.17 °C based on observations (IPCC, 2021).

Advances in attribution methods provide a framework (Hope et al., 2021) to understand observed changes (Table 1), and to identify climatic impact-drivers: attribution of changes in atmospheric composition and resulting radiative forcing to human activities; attribution of changes in a range of variables describing the global and regional state of climate to human influence; attribution of weather and climate events to human-caused climate change (Perkins-Kirkpatrick et al., 2024); and attribution of observed impacts in various sectors, regions, natural and human systems to climatic drivers, themselves attributable to human influence. Here, we focus on indicators of global climate, and attribution outcomes.

Table 1.

Synthesis of the assessment of changes in large-scale indicators of the state of components of the climate system for observed changes (described in the left column) for observed changes (middle column) and their attribution to human influence up to the year 2019 (right column). The color coding indicates the assessed confidence/likelihood of the observed change, using the calibrated IPCC methodology to report confidence in findings, and the human contribution as a driver or main driver (more than 50% of observed change, as specified in the table box). Reproduced from the IPCC AR6 Synthesis Longer Report (IPCC, 2023), based on (IPCC, 2021).

Based on methods and datasets assessed in the Sixth Assessment Cycle of the Intergovernmental panel on Climate Change (IPCC, 2021) (Table 1), updated until the end of year 2024, the observed level of global warming has reached 1.24 (1.11–1.34) °C above 1850–1900 in 2015–2024, of which 1.22 (1.0–1.5) °C is attributable to human influence. Human-caused global warming continued at a record pace of 0.27 (0.2–0.4) °C per decade in 2015–2024, reaching 1.36 (1.1–1.7) °C by 2024 (Forster, Smith, Walsh, Lamb, R. Lamboll, Hall, et al., 2024; Forster, Smith, Walsh, Lamb, R. Lamboll, Cassou, et al., 2025). The increased global warming rate results from increased radiative forcing. It is primarily due to anthropogenic emissions of greenhouse gases, to which is added the loss of the cooling effects resulting from reduced emissions of aerosols, which track scenarios of strong pollution control. The probability of seeing an observed temperature of 1.52 °C in 2024 considering a human-induced warming equal to 1.36 °C is about one chance out of six; the same probability but conditional on the fact that 2024 followed an El Niño year and that the Atlantic multidecadal variability (AMV) was in a positive phase rises to one chance out of two. 2024 can therefore be treated as a “normal” year, i.e. very much expected at the actual human-caused global warming level when the internal modes of variability are taken into account and when assessed from a very large number of simulations from large ensembles.

Amongst greenhouse gases, the largest contribution to the observed warming effect arises from emissions of CO₂, followed by emissions of CH₄ with a direct effect and an indirect effect through its role as a precursor of tropospheric ozone formation (IPCC, 2021). Combustion and fugitive emissions for fossil fuels account from around 70% of greenhouse gas—caused warming, with other sources of greenhouse gas emissions arising from land-use, agriculture, cement production, waste, and F-gas emissions (Forster, Smith, Walsh, Lamb, R. Lamboll, Cassou, et al., 2025). Around 10% of global CO₂ emissions are driven by change in land-use and in particular deforestation, and 90% are driven by fossil fuel combustion and cement production (Friedlingstein et al., 2025).

In a fast changing climate, these annual updates of the key indicators of global climate change and of the global carbon budget represent a major effort in how the research community builds on the in-depth assessment of datasets and methodologies within IPCC reports to inform evidence-based decision-making with up-to-date and timely information. Annual update approaches have recently been implemented for wildfires (Kelley et al., 2025). A new methodological challenge will be to assess the contributions of carbon cycle feedbacks such as wetland methane emissions or wildfire CO₂ emissions to changes in atmospheric greenhouse gas concentrations. There is also a need for annual updates of climate forcing datasets for climate simulations (Naik et al., 2025). Moreover, annual updates of forced warming can also be used to update constrained projections (Ribes, Tessiot, et al., 2025).

Human-caused heating of the climate system drives widespread and rapid changes in each component of the climate system, atmosphere, ocean, biosphere and cryosphere (Table 1) and exacerbates the frequency and intensity of extreme events such as marine and terrestrial heatwaves, agricultural (soil moisture) drought, and extreme rainfall (WMO, 2026; IPCC, 2021), leading to negative impacts and losses and damages for ecosystems and human activities (IPCC, 2022a).

Global mean sea-level rise is also a direct consequence of the Earth’s energy imbalance and heating of the climate system, with an acceleration of its pace from 2 mm/year in the 1990s to more than 4 mm/year in the recent decade, with the mass loss from Greenland and Antarctica adding to glacier retreat and increased rates of ocean heat uptake and resulting thermal expansion (WMO, 2026) (Forster, Smith, Walsh, Lamb, R. Lamboll, Cassou, et al., 2025). Due to the slow adjustment and delayed response of glaciers, ice sheets and ocean heat at depth, past greenhouse gas emissions and the current level of global warming already imply multi-centennial committed sea level rise of more than 1 meter (Mengel et al., 2018; Nauels et al., 2019). Annual updates of committed future sea-level rise resulting from past emissions to date are not yet available. However, the magnitude and rate of future sea-level rise and the timing of when 1 m of sea level rise will be exceeded depend on both future emissions and ice sheet processes associated with deep uncertainty (IPCC, 2021).

The level of human-caused global warming is related to changes in global (Table 1) and regional climate characteristics, resulting in changes in characteristics (intensity, frequency, duration, timing, and spatial extent) of multiple climatic impact-drivers, which are “physical climate system conditions (e.g., means, events, and extremes) that affect an element of society or ecosystems” (Ruane et al., 2022).

This climatic impact-driver framework is based on essential climate variables and physical thresholds, accounting for instance for the combination of extreme heat and humidity which challenges the ability to regulate body temperature (heat health index), or combined hot, dry and windy weather conditions favorable to wildfires (forest fire danger index).

Compound, sequential or simultaneous climatic impact-drivers can affect systems in ways that extend beyond individual components. This is typically the case for compound flooding, with overlapping extreme sea level resulting from storm surge and fluvial flood resulting from extreme rainfall (Hermans et al., 2024). In Europe, there is the emergence of compound events such as the co-occurrence of marine and atmospheric heatwaves, drought and fire activity (Santos et al., 2024), and the pre-conditioning of compound heavy hourly rainfall events by heatwaves. An example of a sequential ecoclimatic event consists of a warm early season, leading to an early onset of vegetation growth, followed by late frost, damaging grapevines and orchards (Bastos et al., 2023). In France, winter wheat crop yields are more severely affected by successive events, such as winter and spring drought, followed by summer heat and drought, than by each individual driver (Shan et al., 2024). Simultaneous events can be for instance concurrent extreme events affecting large crop production regions.

Depending on system tolerance, climatic impact-drivers and their changes can be detrimental, beneficial, neutral, or a mixture of each across interacting system elements and regions. Current and future impacts and climate-related risks thus depend on the interplay between response measures affecting exposure, vulnerability, and hazards resulting from climatic impact-drivers, themselves related to global warming levels (Figure 1).

2.2. Amplified warming observed in Europe and mainland France

Warming is amplified over land, compared to the ocean. Europe is the continent outside the Arctic characterized by the largest rate of regional warming (Copernicus Climate Change Service (C3S) and World Meteorological Organization (WMO), 2025). The observed increase in regional surface air temperature averaged over the last decade has already exceeded 2 °C above 1850–1900 (Heino et al., 2023)—with a similar result in mainland France (Météo France, 2025b). The enhanced Europe warming rate results from feedbacks associated with the reduced snow cover duration, in the north-east parts and in mountains, as well as feedbacks associated with agricultural drought and surface heat, with a rapid increase in heatwave frequencies.

The European “heatwave hotspot” is increasing faster in observations than in climate model simulations, due to regional trends in atmospheric circulation (Vautard et al., 2023) and aerosol emissions (Schumacher et al., 2024). Attribution methodologies have been applied to the increase in surface air temperature in mainland France (Ribes, Boé, et al., 2022). During the last decade, the observed level of warming in mainland France has already reached the level expected to be reached by 2030, for a level of human-caused global warming of 1.5 °C (Météo France, 2025a).

2.3. Systematic comparisons between observed trends and climate simulations

Systematic comparisons have been performed between observed changes in the climate system, and the expected human-caused signals depicted in simulations performed with climate or Earth system models, decades ago (Hausfather, Drake, et al., 2020; IPCC, 2021; Stouffer and Manabe, 2017), including early projections of sea-level rise (Törnqvist et al., 2025).

A recent overview (Simpson et al., 2025) highlights the success of climate models to accurately represent historical increases in global mean temperature and global column water vapour, the global increase in extreme precipitation event intensity, the decline in wintertime cold extremes in the northern hemisphere, the increased recurrence of marine heat waves, and the increased amplitude of the seasonal cycle of sea-surface temperature in the northern hemisphere. Models have also successfully captured the expansion of the Hadley cell¹ edges, the weakening of the northern hemisphere jet stream and summer storm track, and the shift in the southern hemisphere mid-latitude jet in response to ozone depletion and its pause driven by ozone recovery.

Models have been successful in depicting the sign of the following changes, but could underestimate the magnitude of changes for the top-of-atmosphere radiative imbalance, for Arctic amplification and sea-ice retreat, for the weakening of the tropical overturning circulation, the increased contrasts of precipitation between tropical dry and wet regions, the increased precipitation variability; they could overestimate the magnitude of tropical troposphere warming.

Model-data discrepancies are identified for the pattern of tropical Pacific sea-surface temperature (SST) gradient, the strengthening of the wintertime North Atlantic jet, the increased in summer Greenland blocking, the exacerbated summer warming and drying in Europe (as described above), near-surface specific humidity above arid regions, and Southern Ocean SST and sea ice variations. The situation remains uncertain for the winter Eurasian temperature trend, Hadley circulation strength, southern hemisphere storm tracks and the poleward shift in the zonal mean jet streams.

Model-data discrepancies can arise from model caveats, including model representation of processes shaping climate feedbacks and/or model resolution, but also from issues related to observations, forcings used in simulations, and from internal (unforced) climate variability. Robust procedures are developed to identify and understand model-data agreement as well as sources of discrepancies. Such comparisons are embedded in diagnoses of the physical processes controlling the forced climate response to anthropogenic forcings, to advance the understanding of the physical climate response, the fitness for the purpose of models, and they inform the assessment of confidence in projections.

In the IPCC AR6 (Hausfather, Marvel, et al., 2022), the comparison between simulations and historical global temperature observations was used to apply different weights to combine projections from different climate models, further constrained by the assessment of climate sensitivity (Sherwood et al., 2020). Indeed, due to differences in observed and simulated patterns of warming, recent warming alone provides little information on climate sensitivity and future responses (Armour et al., 2024). Novel approaches are being explored to further constrain projections at global and regional scales (Y.-H. Kim et al., 2023; O’Reilly et al., 2024; Li et al., 2025), combining observations, simulations, machine learning (Nowack and Watson-Parris, 2025) and theoretical understanding (Byrne et al., 2024).

3.1. Socio-economic scenarios

Socio-economic scenarios are an ensemble of opportunity, incomplete in essence, distinct from predictions. Five Shared socio-economic pathways (SSP1, sustainability; SSP2, current trends; SSP3, regional rivalry; SSP4, inequality; SSP5, fossil-fueled development) have been designed to encompass a broad range of challenges for climate change adaptation and mitigation (O’Neill et al., 2017) (Figure 3). For climate-related impacts and risks, SSPs are used to evaluate exposure, vulnerability, and adaptation challenges.

Depending on mitigation levels, modelled emissions pathways from SSPs (Riahi et al., 2017) can be consistent with different global warming levels (Figure 3).

For instance, scenario SSP3 displays high challenges for adaptation, and high challenges for mitigation—making limiting global warming well below 2 °C unfeasible. As this scenario echoes some recent geopolitical events, it deserves specific attention regarding its implication for climate change: “SSP3 explores a narrative of resurgent nationalism, concerns about competitiveness and security, and regional conflicts which push countries to increasingly focus on domestic or, at most, regional issues. This trend is reinforced by the limited number of comparatively weak global institutions, with uneven coordination and cooperation for addressing environmental and other global concerns. Policies shift over time becomes increasingly oriented toward national and regional security issues, including barriers to trade, particularly in energy resources and agricultural markets. Countries focus on achieving energy and food security goals within their own regions at the expense of broader-based development, and in several regions move toward more authoritarian forms of government with highly regulated economies. Investments in education and technological development decline. Economic development is slow, consumption is material-intensive, and inequalities persist or worsen over time, especially in developing countries. (…) Growing resource intensity and fossil fuel dependency along with difficulty in achieving international cooperation and slow technological change imply high challenges to mitigation. The limited progress on human development, slow income growth, and lack of effective institutions, especially those that can act across regions, implies high challenges to adaptation for many groups in all regions” (O’Neill et al., 2017).

While the narrative of SSP can echo some aspects of the current world context, it does not imply that model pathways based on SSP3 are closer to current trends or should be considered of higher relevance, because SSP storylines are scenarios for the medium and long term, and because the implementation of the SSP3 storyline in models may not reflect current trends, such as international trade, technological shifts towards renewables, policy momentum in many nations, climate finance and more cooperation than in the SSP3 worst case.

Different mitigation strategies can be consistent with different warming levels by 2100. SSP-based scenarios are labelled SSPx-y, where “x” refers to the shared socio-economic pathway, describing socio-economic trends, while “y” refers to the magnitude of resulting radiative forcing resulting from that scenario, in W⋅m⁻² by 2100 (Figure 1). The response of the climate system was assessed (IPCC, 2021) for five illustrative pathways based on SSPs and which encompass the whole range of future anthropic forcings available from the scientific literature. These scenarios combine socio-economic assumptions, and different levels of mitigation, land use, and atmospheric pollution controls for emissions of aerosols and precursors of ozone formation. The comparison between scenarios and emission inventories requires to accounting for different accounting conventions for land-based carbon fluxes (Gidden et al., 2023).

High and very high emissions scenarios (SSP3-7.0 et SSP5-8.5) have respectively a doubling of CO₂ emissions by 2100 and 2050, respectively (Figure 1). Such high emissions scenarios are less plausible owing to technological progress and mitigation policies already implemented. Indeed, more than 24 countries, including France, have passed peak CO₂ emissions and started sustained emissions reductions (Lamb et al., 2022). On the global scale, CO₂ emissions continue to increase, albeit at a smaller rate than one decade ago. Avoided emissions arise from mitigation policies (Stechemesser et al., 2024), such as reduced deforestation, low-carbon electricity production, gains in energy efficiency and technological innovation policies allowing for reduced costs and increased capacities for PV electricity production and batteries (IPCC, 2022b).

The intermediate emissions scenario (SSP2-4.5) is characterized by relatively stable global CO₂ emissions until mid-century, followed by a slow decline (Figure 1). This is the scenario closest to the extrapolation of current mitigation policies and nationally determined contributions pledged within the Paris Agreement (UNEP, 2024).

Low of very low emissions scenarios are marked by sharp declines in global CO₂ emissions reaching net zero CO₂ emissions by 2050 and 2070, respectively, followed by various levels of net negative CO₂ emissions (Figure 1). Both are also characterized by a rapid reduction in methane emissions.

This core set of 5 scenarios, with differences in air pollution control and mitigation stringency, cover a broader range of greenhouse gas and air pollutant futures than assessed in earlier IPCC WGI reports.

3.2. Constrained global surface temperature projections and implications for France

The uncertainty on future climate projections depends on the socio-economic uncertainties affecting future emissions (scenario uncertainty), but also on the relationship between emissions and concentrations (carbon feedback uncertainty, affecting the natural response of the land and ocean carbon sinks), and on the uncertainty of the climate response, which is related to the understanding of the physical response of the climate system to changes in atmospheric composition resulting from climate feedbacks and climate sensitivity as well as the structural uncertainties in the climate model. To that is added the irreducible and intrinsic uncertainties related to external natural forcings (solar and volcanic activity), and internal climate variability.

One first approach to explore the uncertainty of the future climate response is based on climate model results from the Sixth phase of the Climate Model Intercomparaison Project CMIP6 (Eyring, Bony, et al., 2016), and the range of simulated responses to a given atmospheric concentration or emissions scenario, which includes both large ensembles to span the range of simulated internal variability, and multi-model ensembles, to span the uncertainty of climate feedbacks and climate sensitivity (IPCC, 2021).

Earth system models which include the explicit modelling of some processes of the carbon cycle interplay with climate can be driven by emissions, or by prescribed atmospheric concentrations, based on a central estimate of the carbon cycle feedbacks. For a few Earth system models for which both emission-driven and concentration-driven simulations were available for the same scenario, the differences were small, and do not affect the assessment of global surface temperature projections by more than 0.1 °C by 2100, with a 0.1 °C larger spread (ibid.). However, this is only true for high emission scenarios, but not for overshoot scenarios, due to non-linearities and higher uncertainties of the simulated carbon cycle in response to a change from positive to negative emissions (Asaadi et al., 2024).

When comparing scenario-based, multi-model projections in CMIP5 to CMIP6 (Durack et al., 2025), about half of the increase in simulated warming in CMIP6 compared to CMIP5 arises from a higher radiative forcing in nominally-corresponding scenarios: the radiative forcing is larger in SSP5-8.5 than in the earlier RCP8.5 concentration pathway. The plausibility of such high emissions scenarios is challenged due to developments in climate policies and advances in clean technologies. However, the high atmospheric greenhouse gas concentration levels corresponding to these high emissions scenarios cannot be ruled out, due to uncertain carbon cycle feedbacks, which, in lower emissions trajectories, could result in concentrations above the central levels used to drive climate model projections. The other half of the increased simulated warming arises from a larger prevalence of high climate sensitivity within CMIP6 compared to CMIP5 climate models (IPCC, 2021).

Methodological advances (Sherwood et al., 2020) now allow to combine new evidence of the Earth’s climate sensitivity, based on the understanding and quantification of the Earth’s energy imbalance, the instrumental record of global surface temperature change, paleoclimate reconstructions, and climate feedbacks, in a holistic view (Jeevanjee et al., 2025). In particular, feedback processes are now understood to become more amplifying on multi-decadal time scales, depending on the pattern of warming (Armour et al., 2024). This has to be accounted for when using insights from historical records or past cold periods (Cooper et al., 2024). Accounting for the pattern effect provides a framework to combine lines of evidence based on paleoclimate and recent trends, and supports an updated assessment of the best estimate of equilibrium climate sensitivity of 3 °C, with a very likely range of 2 to 5 °C, and a likely range of 2.5 °C to 4 °C which is narrower than previously assessed. Clouds remain the main cause of uncertainty in climate feedback, but advances in the understanding of cloud processes have led to a narrowing of the associated uncertainty range by about 50% in the AR6 compared to the AR5 (IPCC, 2021). Uncertainties resulting from cloud processes are further evidenced by recent studies (Goessling et al., 2025).

On average, CMIP6 models have higher mean values and wider spreads than these assessed best estimates and ranges, resulting from larger cloud feedback, and some of the high-sensitivity and low-sensitivity CMIP6 models are less consistent with the observed historical warming trend and paleoclimate data (Lunt et al., 2024). The models with very high responses provide insights into low-likelihood, high-impact futures which cannot be excluded based on the current state of knowledge (IPCC, 2021).

In order to combine all available lines of evidence, and not just raw climate model results, the IPCC AR6, for the first time, explicitly combined new projections for the selected SSP-RCP scenarios with observational constraints, based on past simulated warming, as well as the updated assessment of equilibrium climate sensitivity and transient climate response. Moreover, climate predictions initialized from the observed climate state have also been used for the period 2019–2028 (ibid.). More recently, methodologies have been developed to allow for annual updates of forced warming and constrained projections (Ribes, Tessiot, et al., 2025), and remain to be implemented on a routine basis.

The outcome of these approaches is reflected in Figure 1, showing the best estimate and likely range of the assessed climate response to the five illustrative SSP-RCP scenarios. They show that the effect of emissions reductions on the slowing down of the rates of global warming would be detectable from natural climate variability within around 20 years. As a result, a level of 1.5 °C of global warming is expected to be reached (averaged over 20 years) in the coming decade across all scenarios, with the crossing time expected in the early 2030s. On other time horizons, future warming depends on the scenario. For instance, a 2 °C level of global warming is crossed under intermediate to high emissions scenarios, and is unlikely to be crossed for low and very low emissions scenarios.

The IPCC assessment does not provide guidance nor a likelihood assessment of the plausibility of climate future scenarios, but highlights that recent and future emissions trends expected from current policies are closest to an intermediate scenario. Future warming implied by current policies has also been updated annually (UNEP, 2024). However, current policies might be reversed, and a large uncertainty range is attached to the best estimate of the climate response to a given scenario: under SSP2-4.5, the very likely range (90% interval) of assessed global warming by 2081–2100 spans from 2.1 to 3.5 °C of global warming, for a best estimate of 2.7 °C.

These global constrained projections can also be used to develop consistent regional warming levels. For instance, the French reference trajectory for adaptation is based on the best estimate of the climate response to this intermediate emissions scenario, assuming a level of global warming of 2 °C by 2050 and 3 °C by 2100, and then applying similar constrained projections methodologies to translate a time horizon and level of global warming to the corresponding level of warming in mainland France (Corre et al., 2025; Ribes, Boé, et al., 2022). The reference trajectory also assumes climate stabilization at 3 °C by 2100, while net zero CO₂ emissions are not expected to be reached by that time horizon from intermediate or current policies scenarios (UNEP, 2024). Such differences have implications for long-term sea-level projections consistent with different future climate scenarios.

Going back to the global scale, the assessment of the Earth system response has also been used to develop simplified climate models, or emulators (Nicholls et al., 2022). These emulators have been used to classify more than 1200 modelled emissions pathways into 8 categories (Kikstra et al., 2022) as a function of the resulting level of global warming, and to distinguish the trajectories allowing to limit global warming to 1.5 °C (with 50% probability), without overshoot (as in SSP1-1.9) or with a limited (<0.1 °C) or high (0.1 °C to 0.3 °C) overshoot, the trajectories allowing to limit global warming well below 2 °C (with 67% probability, as in SSP1-2.6 or 50% probability), below 2.5 °C or 3 °C (as in SSP2-4.5) or 4 °C with >50% probability (as in SSP3-7.0), or to exceed 4 °C by 2100 with >50% probability (as in SSP5-8.5) (Figure 1). These approaches allow to describe what are the characteristics of mitigation pathways consistent with specific warming levels, for instance in terms of phasing out fossil fuels consistent with limiting global warming to different levels (Achakulwisut et al., 2023).

For consistency, emulators consistent with constrained global surface temperature projections have been used for future projections of ocean heat content, glaciers and ice sheet contributions to sea-level rise, based on ice sheet processes associated with at least medium confidence (Figure 3, colored lines in the sea-level projection panel). As a result, global mean sea-level projections are consistent with global surface temperature projections, also allowing to relate global warming levels to future sea-level rise over the time scales of centuries to millennia. The consistency between insights from modelling and from paleoclimate evidence from past warm periods and high stands of global mean sea level provide confidence in the outcome (IPCC, 2021).

Recent studies have identified deficiencies in state-of-the-art simple climate models for the replication of physical mechanisms linking ocean heat and carbon uptake. Future work is thus needed to improve the geophysical consistency between models used in the scenario generation process and assessment (Sanderson et al., 2023; Séférian et al., 2024).

For many climatic impact-drivers, their characteristics and spatial patterns directly depend on the level of global warming, independently of the timing when this level of warming is reached, and of the emissions and land use pathway leading to that level of warming. As a result, the level of global warming is used as a dimension of integration between mitigation pathways and climate-related risks. The interactions between resulting hazards, exposure and vulnerabilities of human activities and societies, species and ecosystems, lead to climate-related risks. The assessment of regional and sectoral risks is conducted within five broad risk typologies, assessed as a function of global warming levels, and, where relevant, socio-economic scenarios (Figure 3). Mitigation and maladaptive responses can indeed also drive negative effects. For instance, land-based mitigation can exacerbate risks to land rights, water and food security, ecosystems and biodiversity (IPCC, 2023; IPCC, 2019).

Risks and projected adverse impacts and related losses and damages escalate with every increment of global warming (Figure 3). For many climate-related risks, they are assessed in the IPCC AR6 to be higher for any future global warming level than assessed in the IPCC AR5. While future changes in slow components in the climate system are unavoidable, due to their adjustment times, they can be limited by limiting future warming and thus by deep, rapid and sustained reductions in greenhouse gas emissions (IPCC, 2023).

Current mitigation efforts are insufficient to respect the Paris Agreement long-term temperature goal, limiting global warming well below 2 °C with the aspiration to limit global warming to 1.5 °C. As a result, the risk assessment also encompasses the possibility of exceeding a level of global warming, and then returning below that level of global warming through assuming the ability to reach and maintain net negative CO₂ emissions (so-called “overshoot” scenarios) (Figure 1).

The plausibility of overshoot scenarios depends on the plausibility of upscaling carbon dioxide removal, with uncertainties on technologies, political and public support, business models, sustainability and equity considerations, and uncertainties associated with permanence and feedbacks (ibid.). Any mitigation delay and associated emissions leads to a higher level of peak warming, and every increment of additional peak warming leads to increased impacts with no expectation of immediate reversibility after overshoot (S.-K. Kim et al., 2022), such as continuous warming of the Southern Ocean, changes in the deep ocean, ocean and land ecosystems and biodiversity, and leads to an additional long-term sea-level commitment. A few studies indicate distinct regional climate characteristics following overshoot compared to stabilization scenarios. Long-term reversibility of climatic impact-drivers after overshoot furthermore has limited relevance for adaptation decision-making spanning the next 50 years. Finally, uncertainties resulting from uncertain climate and carbon cycle feedbacks triggered by peak warming may challenge the ability to stabilize future warming, calling for sustainable carbon dioxide removal capacity to hedge against high warming outcomes (Schleussner et al., 2024).

3.3. Deep uncertainty, storylines and low-likelihood, high-impact eventualities

The current rise in atmospheric CO₂ level, mostly resulting from the combustion of fossil fuels, is unprecedented in up to 16 million years (CenCO2PIP et al., 2023). The resulting climate heating is driving the emergence of novel climate conditions, unprecedented in terms of pace and magnitude over centuries to millennia (IPCC, 2021; Osman et al., 2021), and, combined with human pressures on ecosystems and biodiversity (IPBES, 2019), the degradation of ecosystems and loss of biodiversity and nature’s contributions to people (IPCC, 2022a). As described above, future climate scenarios are associated with various types of uncertainties, both in terms of socio-economic drivers and changes, in terms of Earth system responses, and in terms of complex risks, and systems transformations, with the potential for surprises. IPCC assessment reports use risk frameworks to explore uncertainties, and have developed a calibrated language to report confidence in future projections, quantitatively (ranges of outcomes), and qualitatively.

Information on uncertainty is critical to inform risk management, and robust decision-making under uncertainty. The latter framework embraces the ideas that science-based analysis should “seek to facilitate human creativity, deliberation and judgement in solving complex problems rather than aspire to proscribe the best decisions”, and “science can help decision-makers manage deep uncertainty, not just reduce it” (Lempert et al., 2024).

Regarding future climate scenarios, the IPCC AR6 has explicitly expanded its guidance for the degrees of certainty to encompass deep uncertainty, with the following definition: “A situation of deep uncertainty exists when experts or stakeholders do not know or cannot agree on: (1) appropriate conceptual models that describe relationships among key driving forces in a system; (2) the probability distributions used to represent uncertainty about key variables and parameters; and/or (3) how to weigh and value desirable alternative outcomes”.

Two framing concepts related to deep uncertainty have been developed in climate science literature, and are thus reflected in IPCC (2021), physical climate storylines, and low-likelihood, high-impact outcomes.

Physical climate storylines provide self-consistent narratives, without a quantified probability of occurrence, of possible unfolding of a physical trajectory of the climate system. In IPCC (ibid.), this concept was used for the exploration of low-likelihood, high impact events, or cross-scale interactions for informing adaptation; to put historical events in the context of a changing climate and as an alternative approach to event attribution studies; and within climate services, as providing climate information integrated with socio-economic information. An overview of recent developments highlights the variety of approaches associated with this concept, and the causal chain that they explore, the type of evidence used, and how they incorporate value judgements (Baldissera Pacchetti et al., 2024). These storylines play a key role in the links between climate science and decision-making (Sillmann et al., 2024), but have not yet been explicitly incorporated within the French national adaptation strategy.

Low-likelihood, high impact outcomes are defined as “Outcomes/events whose probability of occurrence is low or not well known (as in the context of deep uncertainty) but whose potential impacts on society and ecosystems could be high. To better inform risk assessment and decision-making, such low-likelihood outcomes are considered if they are associated with very large consequences and may therefore constitute material risks, even though those consequences do not necessarily represent the most likely outcome” (IPCC, 2021).

Compound and current extremes contribute to the increasing probability of low-likelihood, high-impact outcomes, and will be more frequent with increasing global warming levels, which also increase the likelihood of events unprecedented in the observational record of a given region, giving a misleading perception of “surprise” as assessed in Box 11.2, in IPCC (ibid.). For instance, the occurrence of extreme heat reaching 50 °C in Paris is estimated to be implausible under the current climate, but is possible, as a very rare event, in a 2 °C warmer world (Yiou et al., 2024).

Low-likelihood, high-warming storylines describes future global warming exceeding the assessed very likely range, for a given emissions scenarios, for instance based on models with large climate sensitivities (not shown in Figure 1 for scenarios and projections). This has for instance implications for the timing when specific levels of global warming are reached and is described in Box TS.3 in IPCC (2021).

Another example of a low-likelihood, high-impact storyline refers to the possibility, consistent with paleoclimate evidence, of experience several large volcanic eruptions, which would alter the multi-decadal climate trajectory compared to scenario-based projections, as assessed in Cross-Chapter Box 4.1 in IPCC (ibid.). There is a lack of inclusion of such storylines in risk management and adaptation strategies worldwide, which only focus on human drivers of future climate change; such eventualities would have major implications for e.g. food security.

Low-likelihood, high-impact storylines have also been used to describe what would be the consequences of abrupt changes in components of the climate system. There is medium confidence that the projected decline in the Atlantic Meridional Overturning Circulation will not lead to an abrupt collapse before 2100, but such a collapse might be triggered by unexpected meltwater inflow from the Greenland ice sheet. The storyline approach allows to describe, if such a collapse were to occur, the state of knowledge on its consequences for regional changes in hydroclimate, as described in Box TS.3 in IPCC (ibid.).

Likely range sea-level rise projections (Figure 4) do not include ice-sheet processes characterized by low confidence and deep uncertainty. Higher magnitudes of sea-level rise could be driven by earlier-than-projected disintegration of marine ice shelves, the onset of uncertain ice sheet instability processes in Antarctic ice sheet sectors, or faster-than-projected melt and ice flow from the Greenland ice sheet. In the IPCC AR6, a low-likelihood, high-impact storyline was developed for the very high CO₂ emissions scenario, and this worse case outcome was explicitly communicated together with sea-level projections (Figure 3, dashed line) (Kopp et al., 2023). Further research has focused on high-end sea-level rise as a function of the scenarios and time frame (Stammer, van de Wal, et al., 2019), for the purpose of informing practitioners and for stress tests in risk-intolerant contexts (van de Wal et al., 2022). Such high-end storylines, which are relevant for coastal adaptation pathways and their limits, have been incorporated in national sea-level information, for instance in the USA (Kopp et al., 2023) and the UK (Palmer et al., 2024), but are not included in the current approach to sea-level information in the French context, which is raised as a concern for potential maladaptation (Le Cozannet and Cazenave, 2024).

Other areas of deep uncertainty are associated with the Antarctic sea-ice, and with carbon cycle feedbacks. No specific storyline was developed within the IPCC AR6, but carbon cycle feedback processes where explicitly mentioned as a qualitative source of uncertainty for the ability to return following a period of overshoot: “Adverse impacts that occur during this period of overshoot and cause additional warming via feedback mechanisms, such as increased wildfires, mass mortality of trees, drying of peatlands, and permafrost thawing, weakening natural land carbon sinks and increasing releases of greenhouse gases would make the return more challenging”. Recently, a storyline approach combining narratives and scenarios was developed to explore future CO₂ and methane emissions resulting from permafrost thaw (Schuur et al., 2022), showing that a sole focus on country-level emissions without accounting for Arctic carbon feedbacks is not likely to be enough to limit future warming.

Concerns about the loss of efficiency of natural carbon sinks have recently been enhanced by increased tree mortality, ecosystem degradation and large-scale wildfires (Friedlingstein et al., 2025), as well as the recent intensification of wetland methane feedback (Zhang et al., 2023), processes which are not fully incorporated into Earth system models. The ocean and land carbon responses to both CO₂ and climate change are a major source of uncertainty for understanding the full Earth system long-term response to zero CO₂ emissions, and could result in additional amounts of warming, challenging the plausibility of overshoot scenarios (Palazzo Corner et al., 2023), and the ability to limit future global warming without a carbon removal capacity suitable for buffering potential carbon cycle feedbacks (Schleussner et al., 2024). It has thus been argued that carbon-climate interactions and feedbacks are central to how the Earth system will evolve in the future, and should be central to future projections. This calls for carbon emissions and land-activity-driven simulations to be central to future climate scenarios, instead of concentration-driven simulations (Sanderson et al., 2023).

This brief overview highlights the importance of physical climate storylines to high-impact eventualities and tail risks, in addition to the best estimates and most likely climate futures and in complement of probabilistic estimates of occurrences of events and to inform robust decision-making (Lempert et al., 2024). Such storylines are important for instance for climate proofing of risk-averse infrastructure, for climate stress tests of critical infrastructure, and for cascading risk management strategies. Storyline information is not (yet) included in French adaptation strategies, possibly for fear of over-adaptation (over-investments in adaptation), a point explicitly mentioned in the National Adaptation strategy (PNACC3). However, there is a growing interest in such information to inform long-term strategic planning for energy, health and telecommunication infrastructure, calling for credible and reliable information on potential high-impact scenarios and storylines, as developed for the UK (Palmer et al., 2024; Arnell et al., 2025).