2.1.1 Climate emulators in IPCC reports

Climate emulators have been used by the IPCC from its very start. For instance, the First Assessment Report explains that “simpler models, which simulate the behaviour of [general circulation models (GCMs)], are also used to make predictions of the evolution with time of global temperature from a number of emission scenarios. These so-called box-diffusion models contain highly simplified physics but give similar results to GCMs when globally averaged.” (IPCC, 1992). Emulators, because of their computational simplicity, can be used much more widely than complex GCMs or Earth system models (ESMs).

Because of the limited ability in the 1990s to perform long-term coupled atmosphere–ocean runs with a broad coverage of different GHGs and aerosols and an interactive carbon cycle, the early assessment reports relied heavily on simple climate models, including the WGI reports. A technical overview report about their strength and limitations was published by the IPCC in 1997 (Houghton et al., 1997). With an increasing availability of Earth system models of intermediate complexity (EMICs), coupled atmosphere–ocean general circulation models (AOGCMs), and ultimately the fully fledged ESMs, the focus shifted in the physical WGI reports towards the use of progressively more complex models. However, in the AR6 WGI report, climate emulators were used to fill in gaps from experiments of interest that are not run by ESMs (e.g. SPM Figs. 2c and 4b, IPCC, 2021b) and also to bridge the gap between expert assessment of the climate system and some of the unconstrained projections resulting from ESMs (Hausfather et al., 2022; Lee et al., 2021; Forster et al., 2021). Multiple lines of evidence in support of the assessment of climate sensitivity and other climate characteristics led to IPCC WGI AR6 adopting a new approach, which also involved calibrating climate emulators to translate the assessment of key climate characteristics into the global mean temperature projections. Additionally, the increased focus on translating insights from WGI to other stakeholders and scientific communities included stronger cross-WG collaboration and triggered a concerted effort for climate emulator calibration on the basis of a wide range of WGI assessment results.

Figure 2Summary statistics (a: emissions, b: atmospheric concentrations, c: global mean surface temperature) over time across all scenarios in the AR6DB that received a temperature classification and across scenarios in AR6 temperature categories C1 and C3. Panel (a) shows emissions as modelled by IAMs (“Native”), after harmonisation (“Harmonized”), and after infilling missing reported emissions (“Infilled”). Panel (b) and panel (c) show climate outcomes per climate model, using the median value of each variable from the climate emulator probabilistic distributions.

Download

In the WGIII report, there are two key reasons for using climate emulators to assess the temperature outcomes of long-term climate mitigation scenarios. The first reason is time and resources. With a large number of scenarios available from a wide variety of studies, it would take too much computing time to rapidly simulate all scenarios with one ESM, let alone with a wider set of models such as those that participate in international initiatives like the Coupled Model Intercomparison Project (CMIP). For instance, a quick turnaround was required between WGIII's literature cut-off date (11 October 2021), by which scenarios had to be confirmed as published, and WGIII's deadline for Final Government Draft submission by authors (1 November 2021). It is computationally not feasible for modern ESMs to run all scenarios in this timespan. Typically, an IPCC report undergoes multiple expert and government reviews. This means that the climate assessment is repeated multiple times over the course of an IPCC report drafting cycle, which for WGIII AR6 took 3 years from the first lead author meeting to the approval of the SPM. The second reason mirrors the reasoning in WGI, i.e. using climate emulators to combine multiple lines of evidence to represent the overall best estimate and uncertainty range (Lee et al., 2021). In the WGIII context, a single ESM, or even a set of them, is unlikely to match the best estimate as well as physical climate uncertainty of the assessed temperature response to anthropogenic emissions with a good representation of uncertainty as assessed by WGI and might not even reproduce historically observed global mean temperatures well (Smith and Forster, 2021).

2.1.2 Long-term mitigation pathway assessments in previous IPCC WGIII reports

This exercise sits within a tradition of large-scale assessments and previous IPCC WGIII reports, though the practice of grouping mitigation scenarios based on climate emulator outcomes is more recent. Using two models, the First Assessment Report (FAR) WGIII (IPCC, 1990) evaluated three mitigation scenarios (SA90) and two reference scenarios, calculating their atmospheric CO₂ and CO₂-equivalent concentrations, but did not directly assess global temperature outcomes related to these scenarios. The 1992 supplement to the FAR (IPCC, 1992) evaluated six alternative emissions scenarios (IS92 a–f) and provided global warming estimates using the best estimate of climate sensitivity available at that time. In a 1994 follow-up report, the radiative forcing characteristics of the IS92 pathways were assessed in much more detail (IPCC, 1994). The Second Assessment Report (SAR; IPCC, 1996) assessed a wider range of socio-economic scenarios and used a more extensive set of simple climate models (Houghton et al., 1997) but did not use these to assess the temperature implications of the mitigation scenario literature. In a similar fashion, WGIII of the Third Assessment Report (TAR; IPCC, 2001) also did not perform its own temperature assessment or grouping of mitigation scenarios by climate categories but used CO₂ concentrations as stabilisation levels for the assessment of the mitigation pathways (e.g. SPM.1 and Table 2.6 in IPCC, 2001).

The WGIII Fourth Assessment Report (AR4; IPCC, 2007) contained the first IPCC temperature assessment of emissions scenarios from the available literature: 177 scenarios were assessed, covering a mix of CO₂-only and multi-gas studies. Scenario characteristics were compared by grouping them into six categories based on climate targets as reported in each of the original peer-reviewed articles assessed by the IPCC. Where data were unavailable, scenario characteristics for either CO₂ concentrations or radiative forcing within each category (15th and 85th percentiles) were derived using the relationship between CO₂ concentrations and radiative forcing and the relationship between CO₂ concentrations and equilibrium temperature. Only six scenarios fell into the lowest warming category, which was associated with 2.5–3.0 W m⁻² radiative forcing and CO₂ concentrations of 350–400 ppm in 2100, with a rough estimate of 2.0–2.4 ^∘C global mean surface temperature increase above pre-industrial levels (here referring to the era before the industrial revolution of the late 18th and 19th centuries, while in the rest of the paper “pre-industrial” refers to the period from 1850 to 1900) at equilibrium using a climate sensitivity of 3 ^∘C per doubling of CO₂ concentrations. The highest category covered the 6.0–7.5 W m⁻² range of forcing and featured only five scenarios. The report was clear about the limitation of this approach, writing in Sect. 3.3.5 that “it should be noted that the classification is subject to uncertainty and should thus be used with care” (IPCC, 2007).

In the Fifth Assessment Report (AR5) WGIII report (IPCC, 2014), a larger database of 915 scenarios was available for the assessment of mitigation pathways. These scenarios differed in their design (e.g. ever-growing emissions, climate stabilisation, or peak-and-decline scenarios) as well as in how many gases were included. Despite the methodological difficulties in comparing multiple types of scenarios, AR5 still grouped scenarios into different climate categories to enable comparison of their key characteristics (IPCC, 2014). With the scenario literature at that time often using 2100 radiative forcing targets to design scenarios, including the representative concentration pathways (RCPs), CO₂-equivalent concentrations in 2100 were chosen as a classification indicator (CO₂-equivalent concentrations represent the concentration of CO₂ that would cause the same radiative forcing as a given mixture of CO₂ and other forcing components). The calculation of CO₂-equivalent concentrations in 2100 from emissions was standardised. All scenarios with at least information on total Kyoto gas emissions were assessed using the climate emulator Model for the Assessment of Greenhouse Gas Induced Climate Change (MAGICC) version 6.3 (Meinshausen et al., 2011b, a). This model version drew on a probabilistic ensemble where concentration and radiative forcing outcomes were constrained by observations and physical climate parameter uncertainties assessed in AR4 (Meinshausen et al., 2009; Schaeffer et al., 2015), with model updates to better reflect the climate sensitivity distribution as assessed in AR5 WGI (Rogelj et al., 2012). To group scenarios, the median CO₂-equivalent concentration of total radiative forcing of this probabilistic ensemble was used. For emissions harmonisation, to avoid artefacts in the temperature projections resulting from differences in model-reported and historical emissions, emissions were set to historical observation values in 2010, with the difference from model-reported values linearly declining to zero in 2050 (Krey et al., 2014). At minimum, CO₂ from the energy and industrial processes (E&IP) sector (also known as CO₂ from the use of fossil fuels and industry or CO₂-FFI, as used in AR6) and CH₄ and N₂O from the E&IP and land use sectors from each individual scenario needed to be available. For emissions infilling of other species, a set of heuristics was applied to fill in any missing F-gases, carbonaceous aerosols, and/or nitrate emissions (Krey et al., 2014). Another set of practical heuristics was developed to classify scenarios that did not report all necessary GHGs and other emissions or did not report emissions until the end of the 21st century. The classification of such scenarios into groups was based on Kyoto gas forcing only (given a lack of total forcing) in 2100, cumulative CO₂ emissions from 2011 to 2100, and cumulative CO₂ emissions from 2011 to 2050, in order of preference; 114 scenarios were classified into the lowest category of 2.3–2.9 W m⁻² in 2100, with associated 2100 median temperatures ranging from 1.5 to 1.7 ^∘C above 1850–1900 levels.

The Special Report on Global Warming of 1.5 ^∘C (IPCC, 2018) – abbreviated as SR1.5 – featured an extensive climate assessment of emissions scenarios with the most advanced methods so far. After the introduction of temperature targets in international climate policy in the Cancún Agreement of 2010 (UNFCCC, 2010) and the subsequent adoption of the Paris Agreement with its specific long-term temperature goal as stated in Article 2 of the agreement (UNFCCC, 2015), SR1.5 was the first IPCC report where scenarios were categorised based directly on their projected global mean temperature outcomes. This temperature categorisation followed the practice established by the Emissions Gap Reports series of the UN Environment Programme (Hare et al., 2010; Rogelj and Shukla, 2012; Rogelj et al., 2011). SR1.5 only assessed scenarios with information until 2100 for at minimum CO₂ from E&IP and (total) CH₄, N₂O, and sulfur emissions. The SR1.5 approach used the same harmonisation method as AR5, but because an absolute offset harmonisation method would have turned some non-CO₂ emissions pathways negative, SR1.5 rather used a multiplicative (“ratio”) method (Forster et al., 2018). For the infilling of emissions species not reported, including F-gases and black carbon (BC), values from the low forcing scenario RCP2.6 (van Vuuren et al., 2011; Meinshausen et al., 2011a) were used, in line with the focus of the report on 1.5 and 2 ^∘C consistent scenarios. A total of 368 scenarios (out of 529 submitted scenarios) were grouped into six temperature categories, five of which were to indicate different categories of below 2 ^∘C scenarios (Forster et al., 2018; Rogelj et al., 2018: Tables 2.4 and 2.SM.12). Using a MAGICC6 set-up similar to that used in AR5 (Meinshausen et al., 2011b, a; IPCC, 2014), the temperature exceedance probabilities at peak temperature and in 2100 were used to define these categories. In addition, the climate emulator Finite Amplitude Impulse Response (FaIR) version 1.3 (Smith et al., 2018) was used to run all scenarios for a sensitivity analysis. FaIRv1.3 and MAGICC6 produced substantially different temperature and forcing levels for the same emissions scenarios, with FaIRv1.3 typically projecting less warming and MAGICC6 more, mostly due to effective radiative forcing from non-CO₂ components. MAGICC6 was used for the main classification because it was more established in the literature, provided direct comparability with AR5 in the absence of a more recent IPCC WGI assessment, and had been tested against CMIP5 models (Forster et al., 2018, 2SM-3).

AR6, for the first time in IPCC WGIII assessments, used a fully integrated temperature-based classification of mitigation scenarios, with the climate emulators used in WGIII being fully consistent with WGI of the same assessment cycle following an extensive calibration and testing exercise of emulators, building on the recent literature (Nicholls et al., 2021) to assess their suitability for reproducing assessed climate ranges (Forster et al., 2021). The use of climate emulators in WGIII was motivated by several considerations. Firstly, the main physical reason for using a radiative forcing-based measure over temperature in earlier reports, namely an uncertain climate sensitivity (Krey et al., 2014, page 1312 of AR5 WGIII), has been ameliorated by much more robust constraints on both equilibrium climate sensitivity (Sherwood et al., 2020) and the transient climate response (Forster et al., 2021). This allows a more robust estimate of the temperature response from a given emissions pathway. Secondly, there was considerable ambiguity in earlier assessments of which forcing agents were included in the radiative forcing classification as sometimes total anthropogenic forcing estimates (or subsets thereof) were used and sometimes only GHGs were included. Thirdly, the “CO₂-equivalent concentration” classification in earlier reports created some confusion for readers in the context of the more widely used but rather different concept of CO₂-equivalent emissions. Finally, and most importantly, the Paris Agreement long-term global temperature goal makes a global temperature classification of emissions scenarios directly relevant to informing policy decisions.

2.2.1 Reproducibility, openness, and transparency

During the preparation of AR6, accessibility and reproducibility of scientific results were identified as key aspects to be addressed in the production of the report. This relies on the transparency and reusability of the products and tools underpinning the production of these scientific results (Pirani et al., 2022).

The long-term global emissions pathways literature largely relies on IAMs, an increasing number of which are becoming accessible via open-source codes and training material for potential users (Skea et al., 2021). In the WGIII report, increased attention has gone into documenting the core assumptions and characteristics of IAMs in order to facilitate their interpretation and reproducibility. These pathways have been published in peer-reviewed articles, and none of them is created by the IPCC itself. What is however done for the IPCC assessment report is the consistent comparative analysis of the temperature outcomes of the different scenarios based on their emissions.

Until now, the climate-assessment process utilised by the IPCC has been described in the report but has never been discussed in detail or been made openly available to the community as a software tool. Making the climate-assessment process open-source will not only facilitate the reproducibility of the report's scientific findings, but also facilitate future analyses of new data applying a methodology consistent with the AR6 WGIII report.

Making the climate-assessment process open-source can be seen as a continuation and extension of previous efforts such as in AR5 and SR1.5, where the scenario data and climate-assessment information were made accessible in a format following community standards (Huppmann et al., 2018b, a; IIASA, 2014). In addition, increased transparency was provided by releasing the calculations to get from the scenario data to the presented figures and tables in SR1.5 (Huppmann et al., 2018c). Moreover, a growing number of studies have analysed emissions pathways and their temperature outcomes including climate-policy target quantification (Höhne et al., 2021; Meinshausen et al., 2022a) and grey-literature mitigation scenario assessment (Brecha et al., 2022).

2.2.2 The inclusion of multiple climate emulators

The two emulators used in SR1.5 exhibited substantial differences in the near-term warming, and it was unclear how many of these differences were structural and how many were from different calibrations (Forster et al., 2018). Since then, emulator diversity and the understanding of differences between emulators have improved. Structural uncertainties have been probed by comparing idealised simulations of a range of emulators with different physical characteristics, all run with the same best-estimate climate sensitivity (Nicholls et al., 2020). Emulators were able to simulate global mean surface temperatures of more complex models within a root-mean-square error of 0.2 ^∘C over a range of experiments across a range of scenarios. As the ESMs themselves have structural differences, the emulator with the best fit to a given ESM varied. Because it is not known which ESM best captures reality, these results present an inherent structural uncertainty. This structural uncertainty is therefore best explored by using a diverse range of emulators to assess the climate response across scenarios. Diversity comes from both how emulators capture the emissions to the radiative forcing relationship across the considered emissions and from how the transient surface temperature response to a given forcing is represented. To allow for a multi-model assessment, four emulators were calibrated to the same set of WGI AR6 physical responses (Forster et al., 2021). The calibration approach varied amongst the emulators (Smith et al., 2021a). Nevertheless, they produced a similar best estimate and range of responses to the assessment they were trying to match. Newly developed techniques (Nicholls et al., 2021) were applied to evaluate the probabilistic distributions of each emulator. Based on these techniques, WGI concluded that FaIRv1.6.2 and MAGICCv7.5.3 were generally able to match the best estimates of multiple climate indicators, including the change in global mean surface temperature to within 5 % and to match the very likely ranges to within 10 % (Forster et al., 2021). AR6 WGIII, including Chapter 3, used MAGICC to characterise the median estimates of global warming projections. The difference between FaIRv1.6.2 and MAGICCv7.5.3 is greatly reduced and much better understood (Nicholls et al., 2022a) compared to the largely unexplained differences that existed at the time of SR1.5.

2.2.3 Increased detail for non-CO₂ greenhouse gases and aerosols

CO₂ is the dominant driver of long-term global climate change, but non-CO₂ GHG emissions and aerosols play a significant role on different timescales, and reducing warming from non-CO₂-related emissions is important for meeting climate targets. IPCC WGI (IPCC, 2021a) found that historical CO₂-induced warming was 0.8 ^∘C (1850–1900 to 2010–2019), while methane-induced warming was 0.5 ^∘C and sulfate aerosol-induced cooling 0.5 ^∘C, with additional changes from other emissions components and sources. Therefore, while cumulative CO₂ is the strongest determinant of temperature outcomes, particularly because of its long-lived nature and high emissions, non-CO₂ emissions pathways including short-lived climate forcers (SLCFs) are important when analysing temperature projections under different scenarios (Damon Matthews et al., 2021; Samset et al., 2020; Rogelj et al., 2015, 2018; Allen et al., 2009).

Historically, IAMs have predominantly focussed on modelling CO₂ emissions, with other major GHG emissions like methane receiving less attention. Other emissions including minor GHGs, aerosols, and aerosol precursors are covered by fewer models. Some gases that are represented in climate emulators are not modelled for any long-term global scenario IAM considered in AR6, though these particular emissions have a relatively small projected impact on climate change. To maximise the richness and diversity of scenarios available in a given assessment (Guivarch et al., 2022), a process of infilling scenarios with missing emissions data is performed. There is, however, no unique way of infilling scenarios with missing data.

Previous assessments (Sect. 2.1) already undertook a process of infilling, but due to limited available peer-reviewed literature and tools, these methods were rather simple and did not include emissions-species-specific methods or scenario-specific infilled pathways. As an example, in IPCC SR1.5, missing data were taken from SSP1 to 2.6 on the basis that the assessment was focused on 1.5 and 2 ^∘C scenarios rather than the full range including baseline scenarios. This means that there can be an inconsistency between infilled and original IAM emissions in terms of the implicit underlying socio-economic drivers or compound emissions. Particularly in the short term, SLCFs can have a significant effect on temperature. With new literature and tools available (Lamboll et al., 2020), the AR6 WGIII scenario workflow adopted a more systematic approach to infilling that captures more detail in non-CO₂ emissions of scenarios (IPCC, 2022a).

3.4.1 MAGICC

MAGICC (Model for Assessment of Greenhouse gas Induced Climate Change) v7.5.3 is an emissions-driven Earth system model emulator. Its atmosphere is represented as four interconnected boxes (Northern Hemisphere and Southern Hemisphere ocean, Northern Hemisphere and Southern Hemisphere land). The ocean boxes are coupled to a 50-layer upwelling–diffusion–entrainment ocean model. A full description of MAGICC can be found in Meinshausen et al. (2011b), with updates as described in Meinshausen et al. (2020) and Nicholls et al. (2022a, 2021). MAGICCv7.5.3 was calibrated using the Monte Carlo Markov chain technique described in Meinshausen et al. (2009), with an updated step to re-weight the derived posterior to improve the match with the WGI-assessed ranges. The probabilistic distribution used in the climate assessment uses 600 ensemble members, balancing computational costs with ensemble size. As also described in the documentation of the climate-assessment workflow, the MAGICCv7.5.3 binary and probabilistic distributions are packaged separately from the climate-assessment workflow and can be accessed at https://magicc.org/download/magicc7 (last access: 15 December 2022) for use with the climate-assessment workflow.

3.4.2 FaIR

FaIRv1.6.2 is a fully open-source emissions-driven atmospheric model emulator with a state-dependent carbon cycle coupled to a two-ocean layer climate response module (Smith et al., 2018; Millar et al., 2017). The calibration for AR6 was performed using a 1-million-member prior ensemble. Parameters for the carbon cycle and climate response are derived from distributions based on CMIP6 models (Leach et al., 2021; Smith et al., 2018) and assessments made in AR6 WGI (Forster et al., 2021). This prior ensemble is simultaneously constrained on historical temperature (1850–2019), ocean heat content change (1971–2018), near-present-day (2014) CO₂ concentration, and airborne fraction of CO₂ in idealised 1 % per year CO₂ increase experiments at the time of doubled CO₂, the latter of which is assessed by Chapter 5 of WGI (Canadell et al., 2021). Post-constraint checks are performed to ensure that ECS, TCR, and future warming lie close to the AR6-assessed ranges. The constrained ensemble used for probabilistic assessment contains 2237 ensemble members. The calibrated, constrained ensemble for running FaIR is packaged separately from the climate-assessment package and is available as a JSON file from https://doi.org/10.5281/zenodo.5513021 (C. Smith, 2021).

3.4.3 CICERO-SCM

The CICERO simple climate model (CICERO-SCM, Skeie et al., 2017) is also an emissions-driven climate model emulator. The emulator consists of a carbon cycle model (Joos et al., 1996), simplified expressions relating emissions of components to forcing, either directly or via concentrations (Etminan et al., 2016; Skeie et al., 2017), and an energy balance/upwelling diffusion model (Schlesinger et al., 1992; Schlesinger and Jiang, 1990). The ensemble was based on a previously calibrated 30 400-member ensemble (Skeie et al., 2018). A 600-member subset of this ensemble was chosen to best fit the assessment made in WGI (Smith et al., 2021a), with a technique also described in Nicholls et al. (2021). For AR6 the ensemble was calibrated to the current temperature change from 1850–1900 to 1995–2014, with additional cut-offs for unrealistically low aerosol forcing or ECS values. The constrained ensemble for the climate assessment contains 600 members and is provided in a JSON file that is available with the climate-assessment workflow code (Kikstra et al., 2022a). CICERO-SCM has also recently been ported to Python, facilitating use on multiple computer operating systems.

3.5.1 Scenario classification used in AR6

The extensive climate-assessment process provides increased confidence compared to previous assessments in the relationship between probabilistic temperature outcomes and the original modelled scenario. Therefore, the AR6 assessment used, like in SR1.5, a temperature-based set of classification rules, which are shown in Table 2. These categorisation criteria and their associated likelihoods are always associated with limits to global warming, looking at the simulated peak warming in the 21st century and the global mean surface temperature in 2100. For the categories that limit the global median temperature increase to less than 2 ^∘C above 1850–1900 levels (C1–C4), the categorisation rules follow the same scheme as in SR1.5. Beyond these, AR6 WGIII includes categories relevant for higher emissions scenarios that cover the 2–2.5 ^∘C (C5), 2.5–3 ^∘C (C6), 3–4 ^∘C (C7), and 4 ^∘C and higher (C8) global warming ranges, looking at modelled pathways until 2100. As already noted in SR1.5, temperature-based categorisation is affected by uncertainty in future warming, uncertainty in past warming, and the reference period against which temperature levels are compared (e.g. whether “pre-industrial”, which has a variety of interpretations, or specifically 1850–1900 is taken as a reference period; Chen et al., 2021), but the relative difference between warming levels and thus between temperature categories is more robust (IPCC, 2018).

3.5.2 Overshoot degree years

The categories C1 (“limit warming to 1.5 ^∘C (>50 %) with no or limited overshoot”) and C2 (“return warming to 1.5 ^∘C (>50 %) after a high overshoot”) are separated based on their level of overshoot of 1.5 ^∘C. This separation in the classification used in the IPCC report is purely based on the probability of overshoot (IPCC, 2022a), regardless of its magnitude or duration. In practice, however, the separation based on probability also corresponds to the peak temperature of overshoot. Here, we characterise this difference in overshoot for scenarios in more detail.

The extent and duration of the overshoot and the rate of change in overshoot temperatures are important for climate impacts (Hoegh-Guldberg et al., 2018). Temperature levels may be largely independent of the path dependence of CO₂ emissions and removals (Tokarska et al., 2019) under limited overshoot with limited permafrost feedbacks (Gasser et al., 2018), but many climate impacts are not (Seneviratne et al., 2018; Hoegh-Guldberg et al., 2018), including sea level rise and species extinction (IPCC, 2022b). For some impacts, the peak temperature during overshoot may be the most important factor, whereas in others it is rather the integral of overshoot (i.e. the magnitude of the overshoot combined with the duration of overshoot), such as sea level rise in 2300 (Mengel et al., 2018).

To further analyse the characteristics of scenario categories beyond the analysis in AR6, we use the concept of overshoot degree years (ODYs), which is similar to what was shown as “overshoot severity” in Table 2.SM.12 in SR1.5 (Forster et al., 2018) and was included in the metadata of the SR1.5 scenario database (Rogelj et al., 2018; Huppmann et al., 2018b) as “exceedance severity”. Inspired by Geden and Löschel (2017) and recent scenario studies investigating temperature overshoot (Drouet et al., 2021; Riahi et al., 2021; Johansson, 2021; Tachiiri et al., 2019), we add an analysis of the overshoot severity of all assessed pathways of the AR6 WGIII report as the cumulative years above a certain global warming level, multiplied by the projected average annual climatic ^∘C overshoot in each year.

In this study, we look at ODY_1.5 (^∘C ⋅ year) as the cumulative overshoot degree years above 1.5 ^∘C relative to 1850–1900 from the start of each scenario until 2100 or the year specified otherwise: ${\sum }_t\max (\mathrm{0},T_t-T_{\mathit{\theta }})$, where T is the annual mean climatic global warming above 1850–1900, t is the year, and T_θ is the overshoot threshold temperature. The indicator could allow one to define limits for overshoot targets and thus be related to net-negative emissions in scenarios that return to below 1.5 ^∘C. Additionally, it could be useful in studies that investigate the irreversibility of certain climate change impacts and could be an indicator of the resilience of a system. For instance, in the case that some human system or ecosystem is unable to adapt permanently but would be able to withstand up to 10 ODY_1.5, either through limited resilience or by using temporary adaptation measures, this would indicate when, under a certain scenario, the system may collapse. The AR6 Working Group II (WGII) report on Impacts, Adaptation and Vulnerability (IPCC, 2022b) states with medium confidence that shorter durations and lower levels of overshoot are projected to come with less severe impacts. ODY is not an indicator that can be used for all purposes, as for some questions the rate of temperature change or the level of peak warming reached in a given scenario may be more relevant. Still, at the very least an indicator like this acknowledges that not only the magnitude of overshoot, but also the timescales, are important when assessing overshoot risks (Ritchie et al., 2021) and bridges the gap with stylised overshoot scenarios (Huntingford et al., 2017). Analysing IAM scenarios in this way could be a useful link to the broader tipping point literature (Lenton et al., 2019) and potentially inform climate change policy, impact, and adaptation studies.

3.5.3 Alternative policy-relevant scenario classifications

There are multiple possible indicators that can be chosen to classify and group scenarios (see the discussion above and e.g. Table 3.4 in AR4 WGIII; IPCC, 2007). AR4 discussed this mainly as a matter of stabilisation of greenhouse gas concentrations using a specific indicator as a proxy along the chain from mitigation costs through emissions to impacts. In response to the introduction of temperature goals in international policy decisions and the spearheading of a temperature-aligned approach in science-policy reports by the UN Environment Programme (Hare et al., 2010; Rogelj and Shukla, 2012; Rogelj et al., 2011), SR1.5 and AR6 WGIII based their classifications on global warming levels. Global warming levels were used as one of the integrating dimensions in the AR6 WGI report (Chen et al., 2021) and in the AR6 WGII report as well as across WGs. However, it is also possible to append such a classification with a mix of indicators, for instance to reflect a global climate agreement like the Paris Agreement. For example, the IPCC WGIII AR6 report also reports a sub-category, C1a, of C1 scenarios (IPCC, 2022d). The additional criterion for this sub-category is that net-zero GHG emissions are attained, generally in the second half of this century, which can be interpreted to reflect Article 4.1 of the Paris Agreement (Fuglestvedt et al., 2018; Rogelj et al., 2021). Related examples of such mixed classifications exist in the literature. For example, one recent paper proposes a specific interpretation of the Paris Agreement (Schleussner et al., 2022), proposing that pathways can be seen as “Paris-compatible” if they (a) “[do] not ever have a greater than 66 % probability to overshoot 1.5 ^∘C”, (b) “[are] very likely (90 % chance or more) … not ever exceeding 2 ^∘C”, and (c) achieve net-zero greenhouse gas emissions using global warming potentials with a 100-year time horizon (GWP100).

5.1.1 Moving beyond a binary quality-vetting process

New methods could be devised to advance the methods used to vet scenarios that are considered. Vetting scenarios for instance for their historical alignment and variable coverage is important to allow for a certain level of confidence that the modelled climate outcomes are internally consistent with the full modelled scenario given the methods of the climate-assessment workflow (see the Supplement for more detail). In the current AR6 process, a scenario was either found fit for purpose or not considered in the analysis of global temperature outcomes. Future assessments could attempt to move beyond such a binary procedure and for example look at assigning relative weights to scenarios based on how well they match recent trends and to increase the diversity of the evidence base, with the global scenarios with a climate assessment in the AR6DB being dominated by only a handful of modelling frameworks (Supplement Table S2). In the report, it could lead to more information being available for partial assessments of scenarios. For the climate assessment, knowing which emissions trajectories are more in line with past trends could be used as information to determine how to infill a trajectory when it is missing. Moreover, new methods and evidence are required to assess the performance of emissions-driven climate emulators with higher confidence. Most of the CMIP exercises run concentration-driven experiments instead of the emissions-driven runs that would most directly inform emulator calibration and improvement. This research gap is particularly wide for understanding the climate consequences of scenarios with net-negative CO₂ and GHG emissions.

5.1.2 Towards improving understanding of the role of aerosols in climate mitigation pathways

The role of aerosol and aerosol precursor emissions in warming projections of scenarios remains uncertain. This is in part due to large climate uncertainties that remain in the various aerosol–climate interactions and in emissions inventories and in part because of a lack of a broadly representative set of scenarios for regional aerosol emissions. There is also still a relatively modest focus of the IAM community on modelling alternative effects of aerosol and precursor emission processes, with aerosols generally not being part of scenario protocols in multi-model IAM studies.

5.1.3 Connecting to regional climate impact studies and IPCC WGII

The advancements in integration of insights and assessments from different scenario research communities across climate mitigation and physical climate sciences in AR6 fell short of being fully reflected in the assessment of climate change impacts in WGII. However, the methods described in this paper could be one way to allow for such further integration. A closer connection between scenarios and the assessment of physical climate science on the one hand and impacts, vulnerability, and adaptation studies on the other hand could provide an extremely impactful contribution to the next IPCC assessment cycle. For instance, the current climate-assessment workflow from emissions to global temperature change could be extended to enable the inclusion of regional emissions details and effects on regional climate such as from local aerosol forcing. This could for instance come in the form of emulators to provide regionally downscaled mean and extreme temperature projections using tools such as MESMER (Beusch et al., 2020, 2022; Quilcaille et al., 2022) and ClimateBench (Watson-Parris et al., 2022) or other modelling approaches that utilise regional emissions data available in the AR6DB to enable differentiation between for instance regional aerosol emissions pathways (Fig. 10). A natural next step is to move one step further down the cause–effect chain from regional climate change to regional climate impacts. Using such a chain of emulators (Beusch et al., 2022) could enable probabilistic assessments of various types of impacts both at different global warming levels and under scenarios not considered by Earth system models, supplementing the evidence base used for adaptation and impact assessments made in IPCC WGII. Even without regional impacts, relevant global metrics can be obtained from this kind of workflow such as global sea level rise. In turn, the scenario development and IAM community could draw lessons from such studies too, for instance by exploring parts of the impacts, vulnerability, and adaptation space that are found to be understudied.

Figure 10An overview of the current climate-assessment package (v0.1.1) and its workflow as applied for the IPCC AR6 mitigation scenarios climate, in blue. In orange are a few possible future extensions of this community climate-assessment workflow.

Download

5.4.1 Improvements for harmonisation

This paper has analysed the changes in temperature estimates as the result of different methods using an ad hoc set-up. This set-up could serve as an inspiration for a future diagnostic tool and the development of benchmarks. Future work could consider extending or adjusting the decision tree currently available in aneris. For instance, to facilitate earlier convergence times, for CO₂ emissions in scenarios that reach and sustain net-zero CO₂ emissions, the decision tree could incorporate the convergence year dependent on the scenario design. A significant limitation of the harmonisation part of the workflow comes from the uncertainty in historical emissions and how such uncertainties and corrections are projected into the future. Harmonisation now collapses this uncertainty, sometimes updating emissions estimates that are out of date but other times forcing sets of estimates predicated on different measurements to agree with each other. In some cases, the trends of harmonised data can be markedly different to the trends in the original pathways – for instance, if historical emissions of an F-gas were overestimated but are projected to fall over time, the return to the original value can cause a net-positive gradient. Going forward, it would be worth investigating the impact of historical emissions choices and uncertainty on results.

5.4.2 Improvements for infilling

In a similar fashion, infilling performance can also be improved in a few different ways. One way would be to improve upon the infiller database, for instance by simply having a wider variety of modelled scenarios including especially aerosols and individual fluorinated gases, allowing for more differentiated infilled pathways. For some species, however, such as aerosols and ozone precursors, more research is needed to confidently select the most reasonable pathways or to infill a trajectory when it is missing. Another more advanced way would be to consider assigning weights to emissions trajectories in the scenario database. Lastly, and perhaps most influentially, future workflows could consider developing an automated infilling method decision tree for each emissions species. In AR6, two different methods and infiller databases are used, but always with the same lead gas, CO₂ from energy and industrial processes. For example, it may be preferable to let black carbon act as a lead component for infilling an organic carbon time series, when available.

5.4.3 The order of emissions processing steps

Another particular choice that could be evaluated in future work is the order of emissions processing. In AR6, following SR15, scenario vetting is done first, harmonisation second, and infilling (based on a harmonised set of emissions trajectories) last. Such a strategy ensures that the pathways that are infilled always start from a reasonable point and are influenced less by differences in historical emissions databases. Moreover, in this way two pathways that are identical except for when they were last harmonised should have the same infilled emissions. However, it would also be possible to do infilling before harmonisation, which would derive inter-species statistics used for infilling more directly from the modelled processes in the IAMs. This can only be guaranteed if they are infilled after harmonisation to the latest values. Lastly, by reducing the range of projections when using the QRW method, the risk of out-of-sample infilling is reduced.

5.4.4 Potential for further development of a community tool

The climate-assessment workflow is available as an installable open-source Python package with an MIT licence (Kikstra et al., 2022a). The code utilises functions of existing scientific software packages including “pyam” (Huppmann et al., 2021) and has been parallelised to enable one to do runs of many scenarios. It could be used as a community tool for scenario assessment that enables both easier access to well-calibrated climate emulators and the possibility of assessing a wider range of scenarios due to the possibility of infilling emissions trajectories. Such access to a climate-assessment tool can facilitate the development of socio-economic scenarios, for instance when new models only have the ability to model a limited number of emissions species. Results have already been used to allow for calculating the non-CO₂ contribution to warming, which is used to estimate the remaining carbon budget (Lamboll and Rogelj, 2022; Lamboll et al., 2022a).

There are many ways that the climate-assessment workflow could be extended and applied in future work. Some were already listed in Sect. 5.1.3 and visualised in Fig. 10. Here, we highlight additionally the possibility of connecting more climate emulators to this workflow as well as newer versions of already connected emulators through the “openscm” interface (Nicholls et al., 2022b). Firstly, to enable a robust assessment of climate mitigation pathways, a multi-emulator set-up is crucial for understanding both differences between the multiple models out there, including those that participated in RCMIP (Nicholls et al., 2021), and connecting to a common interface can enable easier intercomparisons. Secondly, having a wider set of simple climate models available and connected to this workflow could allow wider applications as the models differ in the detail and methods with which processes are modelled and thus also differ in which variables can be projected alongside scenarios.