Because of the importance of the ScenarioMIP simulations across multiple research fields and to policy makers, the experimental design was developed collaboratively by researchers within the climate science, IAM, and impacts, adaptation, and vulnerability (IAV) communities. The idea for an activity within CMIP6 focused on scenarios was elaborated in discussions in 2013 among the IAM, IAV and climate modeling communities.

Key discussions occurred at the annual meeting of the integrated assessment and impacts communities in Snowmass, CO, in July 2013, and a meeting on CMIP6 at the Aspen Global Change Institute in Aspen, CO, in August 2013, Next Generation Climate Change Experiments Needed to Advance Knowledge and for Assessment of CMIP6 (Meehl et al., 2014).

The ScenarioMIP SSC together with other communities (see below) systematically investigated a number of issues that could substantially influence the experimental design, especially those that would affect the required number of model runs. First, the possibility was considered to identify a smaller subset of scenarios to be run by statistically sampling from among the large number of possible combinations of different SSPs, forcing targets, IAMs, and climate models. It was decided that this approach could currently not be carried out with a reasonable number of climate model simulations without sacrificing the representation of uncertainty for a given scenario. Second, the potential for pattern scaling or other statistical emulators of climate model output to meet some of the demand for scenario-based climate information was considered. A workshop held for this purpose concluded that pattern scaling has currently not yet been demonstrated to be able to reliably replace the need for climate model projections to generate information for impact studies (although it might play a role for some applications, e.g. Tebaldi and Arblaster, 2014; see workshop report at https://www2.image.ucar.edu/sites/default/files/event/PS2014WorkshopReport_0.pdf). Finally, the difference between scenarios (in terms of global average forcing or temperature change) that is required to produce significantly different climate outcomes was investigated. Initial studies indicate that scenario differences of at least 0.3 ∘C in global average surface temperature are likely necessary to generate statistically significant differences in local temperature over a substantial fraction of the surface, and substantially larger differences are required to produce similarly significant and extensive differences in precipitation outcomes (Tebaldi et al., 2015). Further work on this topic is desirable, especially to explore the sensitivity of additional impact-relevant variables, time and spatial scales of interest, and local forcings.

Informed by these conclusions, a process was organized by the SSC to develop a final protocol. This process included close interaction with the climate research, IAMs and IAV communities through presentations and discussions at a number of meetings in 2014 and 2015,

Session at the July 2014 Snowmass meeting on integrated assessment and impacts; joint meeting on proposed CMIP6 MIPs on scenarios, land use, and aerosols and chemistry, Aspen Global Change Institute, August 2014 (O'Neill et al., 2014a); WCRP-IPCC WG1 meeting in Bern, Switzerland, September 2014; WGCM18 meeting in October 2014; annual meeting of the Integrated Assessment Modeling Consortium, November 2014; IPCC expert meeting on scenarios, IIASA, Laxenburg, Austria, May 2015.

ScenarioMIP has three primary objectives: a.

Facilitate integrated research leading to a better understanding not only of the physical climate system consequences of these scenarios, but also of the climate impact on societies. The results of the ScenarioMIP experiments will provide new climate information for future scenarios that will facilitate integrated research across multiple communities including the (1) climate science, (2) integrated assessment modeling, and (3) impacts, adaptation, and vulnerability communities. This research will be key in informing mitigation and adaptation policy considerations, including processes that are part of the UN Framework Convention on Climate Change (UNFCCC) such as the 2015 Paris Climate Agreement.

Second, scenarios for integration can serve as a key means for producing better integrated scientific assessments, such as those connecting different working groups and the synthesis report of IPCC.

Third, the recent Paris Agreement adopted by parties to the UNFCCC (2015) has focused renewed attention on the goal of limiting warming to below 2 ∘C global mean temperature change relative to pre-industrial and encouraged countries to pursue efforts to limit warming to an even lower goal of 1.5 ∘C. Integrated scenarios can help inform dialogues and associated comparative climate changes to help address these political goals.

Finally, a common set of scenarios for integration reduces the need for individual research projects to develop their own scenario information to support scenario-based studies. The availability of common scenarios reduces possible redundancy in efforts and makes scenario-based research feasible for many groups that otherwise would not be able to carry it out.

Moss et al. (2010) introduced a parallel approach for developing new community scenarios, followed by an integration phase. One of the parallel tracks was the production of climate model projections based on the four RCPs as part of CMIP5 (Taylor et al., 2012). The other track developed alternative future societal development pathways (the SSPs) and emissions and land use scenarios based on them, generated with IAMs. The integration phase brings together the climate simulations and SSP-based societal futures to carry out integrated analysis.

The SSPs were developed over the last several years as a community effort and describe global developments leading to different challenges for mitigation and adaptation to climate change. A conceptual framework for the SSPs and how they could be used with RCP-based climate simulations to carry out integrated research was developed first (van Vuuren et al., 2012, 2014; O'Neill et al., 2014b; Kriegler et al., 2012, 2014a). The specific content of the SSPs was developed next (Riahi et al, 2016). These comprise five alternative narratives that describe the main characteristics of the pathways in qualitative terms (O'Neill et al., 2015) as well as quantitative descriptions for key elements including population (KC and Lutz, 2014), economic growth (Dellink et al., 2015), and urbanization (Jiang and O'Neill, 2015).

In short, the SSPs describe alternative evolutions of future society in the absence of climate change or climate policy. SSPs 1 and 5 envision relatively optimistic trends for human development, with substantial investments in education and health, rapid economic growth, and well-functioning institutions. However, SSP5 assumes an energy intensive, fossil-based economy, while in SSP1 there is an increasing shift toward sustainable practices. SSPs 3 and 4 envision more pessimistic development trends, with little investment in education or health, fast growing population, and increasing inequalities. In SSP3 countries prioritize regional security, whereas in SSP4 large inequalities within and across countries dominate, in both cases leading to societies that are highly vulnerable to climate change. SSP2 envisions a central pathway in which trends continue their historical patterns without substantial deviations.

IAM scenarios were then developed based on the SSPs by elaborating on their implications for energy systems (Bauer et al., 2016) and land use changes (Popp et al., 2016) and quantifying resulting greenhouse gas emissions and atmospheric concentrations (Riahi et al., 2016). These SSP-based IAM scenarios consist of a set of baseline scenarios, which provide a description of future developments in the absence of climate change impacts or new climate policies beyond those in place today, as well as mitigation scenarios which explore the implications of climate change mitigation policies applied to the baseline scenarios. Multiple IAMs were used for the quantification of the SSP scenarios, and a single “marker” scenario was selected as representative in each case. Scenarios in the ScenarioMIP design are selected from these marker scenarios.

Integrated analyses drawing on the qualitative and quantitative elements of the SSPs and climate change information from the CMIP5 simulations of the RCPs have already begun to appear (e.g., Alfieri et al., 2015; Arnell et al., 2014; Biewald et al., 2015; Dong et al., 2015; Hejazi et al., 2015) and climate model simulations with the RCPs will continue to be a key input to research on climate change and impacts for many years. ScenarioMIP is playing a key role by identifying an updated and expanded set of concentration pathways based on the SSPs to be run by climate models as part of CMIP6. These CMIP6 simulations will allow integrated analyses to be carried out using climate simulations based on the latest versions of climate models, for a larger set of concentration pathways based on the most recent versions of IAMs.

Figure 1 visualizes how SSPs can be combined with climate simulations from either CMIP5 or CMIP6, using the example of a forcing pathway stabilizing at 4.5 W m-2. In general, each SSP forcing pathway combination represents an integrated scenario of future climate and societal change which would be used to investigate issues such as the mitigation effort required to achieve that particular climate outcome, the possibilities for adaptation under that climate outcome and assumed societal conditions, and the remaining impacts on society or ecosystems. The full set of multiple SSPs and forcing outcomes forms a matrix of possible integrated scenarios (van Vuuren et al., 2012, 2014; Kriegler et al., 2012). Each row contains climate model simulations based on a forcing pathway (e.g., a 4.5 W m-2 pathway in Fig. 1), which can be used in combination with the societal conditions described by any of the SSPs, as long as it is feasible that SSP emissions could be made consistent with that forcing pathway (see Sect. 3.1.1 for a discussion of feasibility). We refer to these scenarios as SSPx–y, where x is the specific SSP and y represents the forcing pathway, defined by its long-term global average radiative forcing level.

Following practice established for the RCPs, the forcing level usually refers to the forcing achieved in 2100 but in some cases refers to an intended forcing stabilization level that is reached beyond 2100. Forcing is reported as global average radiative forcing, not effective radiative forcing (Myhre et al., 2013).

Currently, RCP simulations from CMIP5 are available to provide climate information for integrated scenarios combining SSP-based socioeconomic and energy–emissions–land use scenarios (as, e.g., SSP2-4.5) with the climate change projections from CMIP5 (e.g., the RCP4.5 simulations). CMIP5 RCPs were derived from earlier emissions and land use scenarios (van Vuuren et al., 2011b), and therefore the regional pattern of climate change resulting from an RCP climate simulation would not be identical with an SSPx–y simulation following a similar global forcing pathway. An enabling hypothesis of the parallel process is that differences in climate change projections would be small enough to still warrant integration of the two sets of information into mitigation, impacts and adaptation analysis. The ScenarioMIP design will include an updated and expanded set of forcing pathways directly derived from SSPs. Once they become available, climate model simulations based on these pathways will then be used to provide climate information for integrated studies.

As noted in Sect. 2.2, the highest priority objective for ScenarioMIP is to provide climate model simulations that can facilitate a wide range of integrated research on the climate impact on societies, including considerations of mitigation and adaptation. Thus, an overarching interdisciplinary science question addressed by ScenarioMIP simulations is

What are the mitigation efforts, climate outcomes, impacts, and adaptation options that would be associated with a range of radiative forcing pathways?

However in addition, ScenarioMIP simulations will be key to addressing two of the three CMIP6 science questions that have informed the overall CMIP6 design and the endorsement of proposed MIPs, related to the effects of external forcings on the Earth system and to the confounding effects of different sources of uncertainty on future anthropogenic climate change outcomes. Table 1 lists the two questions along with a number of sub-questions that ScenarioMIP experiments are intended to explore. In addition, studies addressing WCRP Grand Challenges (clouds, circulation and climate sensitivity, melting ice and global consequences, climate extremes, regional sea-level change and coastal impacts and water availability) will benefit from the availability of outcomes from future scenario simulations.

The scenario framework described in Sect. 2.3 raises specific questions that ScenarioMIP, in collaboration with other CMIP6-Endorsed MIPs (in particular, the Land Use MIP (LUMIP) and Aerosols and Chemistry MIP (AerChemMIP)) will also help address through coordinated experiments in which variants of ScenarioMIP scenarios will be run by other MIPs.

Are differences in regional forcing, or forcings not included in definition of targets (e.g., biophysical effects), a source of significant differences in climate outcomes across a matrix row?

The rows of the SSP forcing matrix shown in Fig. 1 are defined by forcing pathways that achieve the same level of global average radiative forcing in 2100. ScenarioMIP will carry out climate model simulations for one particular land use and concentration pathway that leads to this level of radiative forcing. However, in principle this forcing level can be achieved via pathways of emissions and land use that differ widely in terms of regional land use patterns, regional patterns of emissions of NTCFs, and mixes of global emissions of greenhouse gases (GHGs) and NTCFs. For example, the different SSPs making up a given row of the matrix will have different patterns of regional economic growth, energy system development, air quality policies, land use, and other characteristics that will lead to the same global average forcing outcome being achieved by different means in each case. Thus, an open scientific question is the degree to which climate outcomes can be expected to differ between land use and emissions pathways that achieve the same global average radiative forcing level in 2100 but have different patterns of regional forcing.

An assumption underlying the parallel process (Moss et al., 2010) and the SSP scenario framework is that these differences in climate outcomes are likely to be small relative to the overall uncertainty in applications of these simulations to integrated analyses (including impact assessments). This assumption is critical to be able to combine a ScenarioMIP climate simulation for a given SSP and forcing level with scenarios based on other SSPs achieving the same forcing level. Experiments carried out in other MIPs based on scenarios in the ScenarioMIP design will help test this assumption (see Sect. 3.3.3). If it turns out that climate outcomes are much more sensitive to local forcing differences than currently assumed, the ability to use ScenarioMIP simulations for each forcing level for all SSPs might not be possible for all studies. In that case, Earth system model (ESM) simulations specific to each combination of SSP and forcing pathway would be required.

In addition, the definition of global average forcing in 2100 includes the forcing effect of GHGs and NTCFs, but excludes the biophysical effects of land use change on climate (e.g., through albedo or changes to the hydrological cycle). Thus, it is also an open question whether alternative pathways that achieve the same level of global average radiative forcing as defined here, but differ in forcing due to the biophysical effects of land use change, would produce substantially different climate outcomes.

What are global and regional climate differences between scenarios with small differences in forcing levels?

The experimental design includes six out of eight 21st century scenarios that are within a maximum of 1.0 W m-2 of another scenario in terms of global average radiative forcing in 2100. Early in the design of the scenario framework, a criterion for selecting RCPs was that they be well separated in terms of radiative forcing (Moss et al., 2008). More recent work (Tebaldi et al., 2015) has refined this view, indicating that regional temperature outcomes that are statistically significantly different at a 5 % level for more than half the land surface area, and robustly so across the multi-model ensemble, require a separation of at least 0.3 ∘C in global average temperature. This difference in global temperature is roughly equivalent to about 0.75 W m-2 of global average forcing in an idealized 1 % yr-1 CO2 increase experiment, although the equivalent value is sensitive to the forcing pathway. For regional precipitation, a much wider separation is required to ensure that scenarios are statistically different. From a policy-making perspective the issue of scenario separation is also important, as policy interest often focuses on the differential impacts between climate change or forcing levels that are relatively close to each other. The ScenarioMIP design will allow for further analysis of these types of questions, providing simulations that will allow addressing region- and variable-specific sensitivities, dependence on geographic and temporal scale of variable differences, and the role of internal variability.

What are the effects of declines in forcing (overshoot scenarios)?

There is both scientific and policy interest in the climate outcomes associated with forcing pathways that exceed a given forcing level and later peak and decline back to that level (overshoot pathways). Such pathways may become increasingly a point of discussion if there is a persistent gap between moderate near-term emission reduction efforts and the ambition to limit climate forcing and global mean warming to very low levels. To this end, the lowest RCP (RCP2.6), and the low SSP scenarios, already exhibit a limited degree of concentration overshoot. One of the scenarios within the ScenarioMIP design describes a much stronger overshoot pathway with radiative forcing that peaks and declines within the 21st century and declines further thereafter, allowing for investigation of the effect of overshoot and declining forcing on the climate system and society. In particular, it allows investigating to what extent climate impacts are higher and what long-lasting and potentially irreversible changes in the climate system occur in an overshoot scenario.

Can pattern scaling, or other approaches to climate model emulation, be used to produce climate outcomes for forcing pathways not represented in the ScenarioMIP design?

Climate model emulators have the potential to provide a computationally efficient means of generating climate outcomes for arbitrary scenarios and, in so doing, facilitate the representation of uncertainty in applications to impact studies (Tebaldi and Arblaster, 2014). However, the state of development of such emulators is such that many situations remain where they are not suitable, their behavior deviating significantly from the more computationally complex, physically based models that they seek to emulate, or falling short of producing temporally coherent projections, or projections of multiple variables physically inter-related. A more systematic exploration and development of such techniques in order to realize their potential will be facilitated by the availability of ScenarioMIP simulations, according to a design that deliberately explores a large range of forcings (both with respect to a lower and upper end, recently found to be important in training emulators by Herger et al., 2015), non-traditional pathways like substantial overshoots and long-term extensions and, together with collaborating MIPs, the effects of regionally and time-varying forcers other than well-mixed, long-lived GHGs, in particular land use changes and NTCFs.

Can emergent constraints (i.e., statistical relationships between features of current and projected future climate that emerge from considering the multi-model ensemble as a whole) be used to recalibrate the ensemble and to reduce the uncertainty in the response to a given scenario of future forcing?

A longstanding open scientific question is the relation between present-day model performance and future projections. A method to relate observed aspects of the present-day mean climate or recent trends to the Earth system response in some quantity is the so-called Emergent Constraints method (Allen and Ingram, 2002; Bracegirdle and Stephenson, 2013; Hall and Qu, 2006). An emergent constraint refers to the use of observations to constrain a simulated future Earth system feedback. It is referred to as emergent because a relationship between such a feedback and an observable element of climate variability emerges from an ensemble of ESM projections, providing a constraint on the future feedback. If physically plausible relationships can be found between, for example, changes occurring on seasonal or interannual timescales and changes found in anthropogenically-forced climate change, then models that simulate correctly the seasonal or interannual responses might make projections more reliably. For example, Hall and Qu (2006) found that large inter-model variations in the seasonal cycle of the albedo between April and May in the 20th century are well correlated with similarly large inter-model variations in the snow–albedo feedback on climatological timescales. The observable variation in the seasonal cycle of the snow albedo is then a useful proxy for constraining the unobservable feedback strength to climate warming, as both are driven by the same physical mechanisms on different timescales. Other examples include constraints on climate–carbon feedbacks (Cox et al., 2013; Wenzel et al., 2014), the Austral jet stream position (Wenzel et al., 2016), cloud feedbacks and equilibrium climate sensitivity (Huber et al., 2011; Fasullo and Trenberth, 2012; Fasullo et al., 2015; Klein and Hall, 2015; Knutti et al., 2006; Sherwood et al., 2014), and relations of past and future sea ice or temperature trends (Boé et al., 2009; Knutti and Tomassini, 2008; Mahlstein and Knutti, 2012; Massonet et al., 2012). The ScenarioMIP design will allow for testing emergent constraint results under various forcing pathways. The results will be valuable for guiding the design of future ensembles, e.g., how many and which models are needed to maximize information at minimal computational cost.

The identification of the forcing pathways to be included in the ScenarioMIP design can be described in two parts: deciding on the forcing levels to include, and then on the specific SSP-based scenario that each forcing pathway should be based on. Additional decisions were then necessary on the number of ensemble members to request from each model for each scenario, and on long-term extensions beyond 2100.