MIPs in CMIP6 as a whole asked for many experiments that jointly placed a big computational demand on climate modeling centers. The requested experiments were designed to address the MIP-specific scientific questions. The three MIPs discussed here contributed to that demand, and the diversity of research interests across the modeling centers meant that some experiments received more attention than others. Setting priorities with tiers was useful to the extent that it highlighted the priority of experiments from the MIP's perspective. In so doing, the tiers guided the participating modelers to focus on some experiments in order to have a larger model ensemble where the MIPs wanted contributions the most. However, in retrospect, some of the Tier 2 experiments may have been more useful than Tier 1. An example here is piClim-histaer (Tier 2) from RFMIP, which quantified the spread in magnitude and timing of historical aerosol forcing in CMIP6 models. It was informationally rich, and was a contributing factor in deriving the aerosol ERF time series for IPCC-AR6 WG1.

Experiments following already known strategies with standard output requests are quicker to set up. These have the advantage that no additional personnel are needed to implement newly requested diagnostic output, e.g., for RFMIP-IRF. On the contrary, the time commitment is longer for an experiment design that needs the implementation of a new parameterization, e.g., for RFMIP-SpAer, which requires dedicated human resources at the modeling center to carry out the work including coding, testing, and performing the experiments and associated scientific exploitation .

A greater number of experiments performed creates more data for statistical analyses and for addressing a variety of research questions, but it is taxing in light of restricted resources. In preparation for the next phase of AerChemMIP and RFMIP, the type and number of experiments in the experimental protocol will therefore be revised based on a refined set of research questions and the desire to reduce the computational burden for modeling centers as much as possible. In this process, AerChemMIP activities will be closely coordinated with other community MIPs with common or similar interests.

In preparation for the second phase of AerChemMIP and RFMIP, we reviewed the status of the experiments and their usage in peer-reviewed publications, as summarized in Table . A total of 67 models performed CMIP6 historical experiments (published via the Earth System Grid Federation (ESGF) in June 2023) that were used in as many as 15 100 publications (as listed by Google Scholar in June 2023). Available model output to assess differences in forcing and response was, however, limited; e.g., output for the mid-visible aerosol optical depth is available only for 45 of the 67 models providing historical experiments. Most of the historical experiments (40) are performed with NTCF emission-driven models. The ESMs with prescribed aerosols (19) in the historical experiments mostly (13) used the MACv2-SP parameterization . MACv2-SP was developed in the framework of RFMIP and is, due to the unexpected relatively broad implementation in ESMs, now included in the work of the CMIP climate forcings task team, although the targeted exploitation of MACv2-SP in RFMIP-SpAer was, with one publication , small compared to the usage of other experiments of RFMIP and AerChemMIP so far.

In total, RFMIP and AerChemMIP received output from 103 experiments, leading to 214 publications to date. We separate the RFMIP and AerChemMIP experiments here into three classes, namely experiments with full coupling between the atmosphere and ocean (hist-X), with prescribed sea surface temperatures and sea ice at preindustrial level (piClim-X), and with prescribed transient changes in sea surface temperatures and sea ice from a historical experiment (histSST-X). Intercomparing these classes, piClim-X experiments were performed the most, with a total of 50 contributing models, followed by hist-X, with 36 models. However, hist-X is used 3 times more often in scientific publications (146) compared to piClim-X (52). The higher computational demand of hist-X, therefore, seems justified by the much larger scientific output compared to the experiments without a coupled ocean (histSST-X and piClim-X), as measured by the number of published articles.

There are inevitable trade-offs in the final experimental choices that can be categorized along the three axes of (1) model complexity, addressing the number of process interactions represented in ESMs and the fidelity of the processes simulations; (2) model resolution, referring to the grid spacing; and (3) simulation length, covering the length and number of members in ensemble simulations. These axes, schematically depicted in Fig. , span a triangle in the complexity–resolution–length space. The volume of the tetrahedron between the origin and the marked triangle indicates the computational need for the experiments. The computational need scales non-linearly. Doubling the simulation length or number of simulations doubles the required computational resources that are needed along these axes, but this is not true for the model resolution and complexity. Increasing the model resolution by a factor of 2, for instance, requires computational resources that are an order of magnitude larger. To account for the non-linearity in the computational need, the volume of the tetrahedron would be calculated on scaled values; i.e., an experiment with twice as fine a resolution would be marked 4 times further away from the origin on the resolution axis. The maximum volume of the tetrahedron is limited by the computation capacity abyss, i.e., the available computing capacity at the modeling center.

Although computing power continues to grow, trade-offs along the three axes of the experimental design and prioritization will continue to be necessary. This is, for instance, the case in light of the computational cost of interactive chemistry against the resolution and the number of simulations. All model experiments, irrespective of whether the models have interactive chemistry, compete for priority at modeling centers due to limited computing resources. Experiments with complex ESMs are necessary to understand interactions of chemical species in concert with climate change, for example, the carbon cycle or atmospheric composition–climate interactions. To that end, ESM experiments are performed that have interactive aerosol and chemistry schemes in addition to the fully coupled atmosphere–ocean–land system, making these models complex and resource-heavy. For instance, an ESM could simulate changes in vegetation cover due to increased greenhouse gases that in turn have an impact on dust–aerosol emissions, in addition to potential changes in soil moisture and winds. In less complex models, the vegetation cover is, for instance, prescribed such that the number of interactive physical processes is smaller. The computational demand of complex ESMs for simulating many processes limits the attribution of computing resources along the other two axes of experimental design, namely performing a large number of experiments, which allows the impact of model–internal variability to be reduced, and choosing a fine enough spatial resolution, which explicitly resolves more physical processes on the model grid. For some research questions, the complexity of ESMs can be reduced to a certain degree. For instance, concentrations of well-mixed greenhouse gases can be prescribed instead of being simulated from emissions if one is interested in computing the forcing and response to a given change in the atmospheric composition (Fig. ). It makes creating large ensembles of ESM experiments possible that are needed to split, for instance, the imbalance in the radiation budget at the top of the atmosphere into a mean radiative forcing and contributions from internal variability. Similarly, a separation of the response in temperature or air quality into a forced signal and a contribution from internal variability is possible. The required ensemble size and length of experiments for sufficiently reducing the influence of model–internal variability on the global mean radiative forcing (e.g., ), climate responses (e.g., ), and impacts on air quality (e.g., ) depend on the magnitude of the forced signal against the magnitude of the internal variability.

The necessary number of simulated years for separating the signal from internal variability depends on the scientific question. The signal-to-variability ratio is, for instance, sufficiently good for the response of the global mean of precipitation and the ERF in the global mean for most climate forcers in the current experiments. Specifically, the suggestion from for performing 30 years of model experiments with the same boundary data proved useful to diagnose global ERF in most time slice experiments, except for land use changes . We learned that the exact precision of ERF depends on the model's internal variability, inducing year-to-year perturbations in the radiation budget . Longer simulations of 45 years are needed to diagnose the forcing of some longer-lived trace gases due to the timescale for gas transport through the stratosphere via the Brewer–Dobson circulation . For regional radiative effects, the 30- and 45-year-long simulations are not sufficiently long to obtain a statistical significance for all anthropogenic perturbations in all regions. In UKESM1, the anthropogenic aerosol radiative effects are, for instance, statistically significant at the 95 % level over about 50 % of the globe, but the effects are only statistically significant for 10 % of the globe for land use and non-methane ozone precursors . Similarly, regional aerosol forcing is not statistically significant over all world regions in models contributing to RFMIP . For simulated climate responses, the ensemble sizes and simulation lengths were not sufficient for addressing all research questions of interest in the three MIPs, especially for regional responses. Quantifying the regional response of climate to forcing requires larger ensembles of simulations, which the Regional Aerosol MIP (RAMIP; ) is currently addressing through requesting larger ensembles of experiments with regional perturbations of aerosols than available from AerChemMIP.

Complex models simulating many processes and their interactions are desirable and needed for specific research questions but pose challenges in reducing model-based uncertainty in the assessment of the climate response to various forcings. Model diversity in terms of, for instance, the combination of parameterizations, intricacy and fidelity of represented processes, choice of coupling of model components, choice of the dynamical core, and resolution is desirable. Model simulations ideally converge to similar solutions for a given question, e.g., how the Earth's temperature responds to anthropogenic perturbations. The diversity in model results should therefore reduce over time to gain confidence in our conclusions drawn from simulated responses to imposed perturbations.

There are two challenges to reducing model-based uncertainty that can be emphasized in the context of MIPs. One challenge concerns the diversity in the level of complexity included in the ESMs, which is, for instance, due to choices made for the interacting processes and the representation of chemistry and aerosols, as well as the specification of the spatial resolution by the modeling centers. As an example, this diversity is clearly evident in the complexity of aerosol processes, with some CMIP6 models simulating the evolution of different aerosol species and their interactions e.g.,, while other models prescribe spatial distributions of aerosol optical properties e.g.,. Such differences in model capabilities have implications for understanding the reasons for differences in their results e.g.,.

The second challenge comes from the consideration of model diversity in the level of complexity inherent in the process of designing a MIP protocol, since, for instance, a few models can simulate processes that most others cannot. Again, MIPs already have a specific class of models in mind. For AerChemMIP, emission-driven models were targeted, whereas RFMIP also included contributions from models with less complex representations of aerosols; e.g., those using prescribed aerosol optical properties. Hence, RFMIP received more output from model experiments than, for instance, AerChemMIP. RFMIP and AerChemMIP were endorsed by CMIP6 and had different structural organizations, while PDRMIP started earlier and was in comparison more self-organized and flexible in the MIP life cycle. PDRMIP, therefore, comprises an ensemble of models of different complexity. Specifically, some of the models in PDRMIP performed experiments with prescribed emissions, whereas others used concentrations resulting in an ensemble of experiments partially driven by emissions and partially driven by concentrations of climate forcers. Yet, MIP experimental protocols do not prescribe the level of process complexity and the resolution of ESMs. This freedom is well justified, since ESMs might otherwise not be able to participate in a MIP if they cannot fulfill stricter requirements. A wider participation of ESMs in MIPs ensures a sufficiently large multi-model ensemble needed to robustly quantify forcings and climate responses considering structural model differences. A full exploration of the role of climate–composition feedbacks with focus on biogeochemical processes, however, remains a challenge due to this difficulty.

New opportunities arise from machine learning approaches. These can, for instance, contribute to improving or speeding up process representations in ESMs, as well as designing smart tools for post-processing and evaluating ESM output. We see primarily four areas in which machine learning could help in advancing the research in our community. These are (1) to include faster and more precise representations of processes in models, e.g., for replacing or modifying physical parameterizations that are thought to not work sufficiently well in all conditions in which they are needed; (2) to develop novel ways to gain a better understanding of physical and chemical interactions, e.g., through data mining employing machine learning techniques; (3) to fill observational gaps, e.g., in satellite products to allow the creation of spatially complete data to more easily validate model results against observational information; and (4) to mimic climate responses to radiative forcing, e.g., to prioritize experiments for the design of new MIP protocols.

Proof of the concept of applying machine learning in our research field exists. One example is using deep learning for the design of new parameterizations (e.g., ). Atmospheric chemistry parameterizations can, for instance, be replaced by fast representations based on machine learning . The causes of multi-model diversity highlighted in previous studies can also be elucidated using machine learning. There is an increase in the availability of globally gridded fused model–observation data products (e.g., ) that can be used as benchmarks in the model evaluation of the atmospheric composition. Novel aspects of such benchmarks include providing data relevant to health impacts e.g., and using machine learning techniques for global mapping of atmospheric composition e.g.,. used deep learning to quantify the sensitivity of surface O3 biases to different input variables in a CMIP6 model (UKESM1), thereby providing a new understanding of biases and enabling future projections of bias-corrected surface O3. Similarly, such approaches have been used to improve our understanding of model diversity in other aspects of atmospheric composition, e.g., surface PM . Including necessary variables for such algorithms in the model output of future MIPs can enable a multi-model intercomparison of different contributions to model biases and provide bias-corrected data for future projections of changes that can be tailored toward impact studies, e.g., concerning future air quality and human health.

Emulators (e.g., ), a class of models that mimics the behavior of an ESM, can help to prioritize new ESM experiments. Emulators are trained on output from existing experiments with ESMs, of which there are now many, e.g., from the CMIP6-endorsed MIPs and several CMIP phases. Unlike ESMs, emulators perform fast calculations instead of the numerical integration of non-linear physical and chemical equations over time on a three-dimensional grid. Both techniques allow spatially resolved predictions of temperature and other variables, but emulators can do it at massively reduced computational costs compared to ESMs . Once established, emulators can be used to explore the climate response to radiative forcing, e.g., to inform experimental designs of future emission scenarios in CMIP6 . In terms of physically based emulators of the climate system (i.e., simple climate models), RFMIP and AerChemMIP experiments were invaluable in the determination of aerosol ERF, ozone ERF, and the factors influencing methane chemical lifetime. Some of these relationships were developed in the lead-up to IPCC-AR6 WG1 and used directly in the report (e.g., ).

Training emulators requires a broad range of ESM experiments, such that they interpolate rather than extrapolate into unseen climate conditions. This training data could be made up of CMIP experiments, an ensemble of idealized experiments , or perturbed parameter ensembles, where several ESM experiments with systematically different settings in parameterizations are performed to study sources of model–internal uncertainties (e.g., ). Emulators have been used for some time , and modern techniques also utilize machine learning to allow validation against observations . Emulators can incorporate model spreads similar to the output from classical MIPs with ESMs. A review of emulation techniques that are routed in statistical mechanics highlights the potential to further improve emulators for use in climate sciences using machine learning . The difficulty of accounting for non-parametric biases of CMIP models in emulators, however, remains . Nevertheless, emulators have already been proven useful to sample parametric differences and to study climate change e.g.,.

Much finer spatial resolutions with horizontal grid spacings of a few kilometers hold the potential to overcome some of the long-standing challenges concerning the representation of clouds, precipitation, and circulation in global climate simulations. Representing clouds and circulation correctly in models with resolutions of several tens to hundreds of kilometers is an outstanding challenge e.g.,. The high spatial resolution naturally improves the representation of clouds and precipitation, at least in part, due to better-resolved orographic effects on atmospheric dynamics and the explicit simulation of convective cloud systems, along with the mesoscale circulation , although not all model biases are eliminated . These processes are tightly connected to atmospheric composition changes and associated effects on the atmosphere, including feedback mechanisms. Furthermore, the coupling of atmospheric processes with the land improves in kilometer-scale experiments. It can reduce biases in the simulated temperature and precipitation , which can help to better understand regional climate change that involves land-mediated feedbacks. Moreover, better-resolved ocean dynamics hold the potential for surprises in understanding climate responses, e.g., with respect to future projections of temperatures and rare high-impact events .

Kilometer-scale experiments, therefore, allow new insights into processes in the Earth system, following the perturbation–response paradigm, and can leverage the experiences made with regional kilometer-scale climate experiments for different world regions (e.g., ). Kilometer-scale experiments with horizontal grid spacings finer than 10 km are presently only possible for climate studies on limited area domains or globally for restricted time periods of a few weeks to single years . Global kilometer-scale simulations for climate change assessments would require a step change in the collaboration between climate science and high-performance computing . Given the role of resolution in maintaining concentrated emissions, non-linearities in chemistry, and fronts in the atmospheric transport of pollutants, more kilometer-scale climate change experiments might prove valuable to advance the understanding of climate and air quality interactions. Such experiments within limited area domains would also help to alleviate the computational cost of both high process complexity and high spatial resolution.

Global coupled atmosphere–ocean simulations with a resolution of a few kilometers can be done, and progress has been made in incorporating some representation of atmospheric composition, such as the carbon cycle . For some questions on atmospheric composition and the associated air quality and climate response, kilometer-scale experiments are already used, e.g., for a better understanding of aerosol–cloud interactions , which is one of the key uncertainties in ERF from ESMs e.g.,. Another question that can be better addressed with kilometer-scale experiments is the resolution dependence of radiative forcing and feedbacks, especially for those that involve clouds that are a key uncertainty in our understanding of climate change with ESMs . Another question is to what extent more resolved meteorological processes aid in improving the representation of atmospheric composition and air quality, e.g., concerning health impacts in urban areas.

The community of the three MIPs will not be able to mainly rely on global kilometer-scale model experiments in CMIP7, since several fully coupled kilometer-scale ESMs with interactive aerosols and chemistry fast enough to perform multi-decadal simulations are unlikely to be ready in time for CMIP7. In light of this restriction, we see two main routes forward for immediately using spatially refined information in our next MIPs. The first possible way is to use the output from global kilometer-scale experiments that are run for other purposes to drive offline models for aerosols and chemistry or atmospheric radiative transfer calculations. This approach is suitable to answer some but not all research questions in our community; this concerns, for instance, the response of dust emission fluxes to changes in winds and moisture e.g., but not the implication of such dust emission changes for air quality and climate responses. For the latter, regional one- or two-way dynamical downscaling experiments could be used.

We perceive dynamical downscaling as the second main avenue for our near-future work to obtain regionally refined spatial information. Regional climate modeling is already well developed and organized via CORDEX (Coordinated Regional Climate Downscaling Experiment), with a focus on providing regional climate change information. Regional climate models with the capability to perform experiments with coupled aerosols and chemistry exist, for instance, in Europe and the USA (e.g., ) but have not been used in our past MIPs. For CMIP7, UKESM2 and CESM (Community Earth System Model) also aim to have regional model configurations. At least two different regional composition–climate models therefore could exist and be used in future MIPs. The regional models will nevertheless need output from global ESM experiments with coupled aerosols and chemistry as boundary data for performing the regional experiments. As such a need for experiments with classical global ESMs is retained, at least for CMIP7, although moving towards global kilometer-scale modeling with a sufficient coupling of physical processes to aerosols and chemistry to address the community's research interests will be a goal to aspire to.

The concept of radiative forcing is central to the perturbation–response paradigm for understanding climate change e.g.,, as radiative forcing is eventually balanced by a temperature response mediated by feedback processes. Definitions of radiative forcing have evolved over time to allow an increasingly wide range of different climate forcers or contexts for perturbations to be considered interchangeably , as schematically depicted in Fig. . Early definitions of radiative forcing focused on changes in the net radiation at the tropopause, ideally after the stratosphere had adjusted to a new radiative equilibrium in the presence of the forcing agent (stratospherically adjusted radiative forcing; ). These definitions have been generalized in the concept of ERF (ERF; see ).

Quantifying IRF for ESMs is desirable, even if the IRF of the forcing agent is constrained by other methods. There is little fundamental uncertainty for the IRF of CO2 changes, as indicated by errors of the order of a fraction of a percent from the most accurate line-by-line radiative transfer models . However, due to the high computational demand, ESMs do not compute the radiative transfer with a line-by-line model. Instead, they rely on parameterizations, speeding up the computation at the expense of accuracy. Consequently, a model spread in IRF occurs despite so little fundamental uncertainty. For instance, a spread in CO2 IRF has persisted across CMIP phases and accounts for a majority of the model spread in the CO2 ERF . Quantifying the model's IRF, e.g., with double calls of the radiative transfer calculations (Sect. ), is particularly relevant in light of the model–state dependence of IRF when referring to ESM differences in atmospheric conditions that affect the radiative transfer. Both CMIP6 models and theoretical arguments suggest that CO2 IRF is correlated with temperature; i.e., CO2 IRF increases as the surface warms and the stratosphere cools. This feedback-like effect on radiative forcing is thought to account for a ∼ 10 % increase in CO2 IRF for present day versus preindustrial and ∼ 30 % for quadrupled CO2 . It requires clarity in the experimental design and reporting of resulting ERF estimates to disentangle the contributions of forcing from feedbacks in future experiments.

There are several methods to quantify ERF across ESMs that can be further improved and standardized (Fig. ). As mentioned earlier (Sect. ), ERF is often computed from model experiments using prescribed sea surface temperatures and sea ice (fixed SST method; ) and has been adopted in RFMIP (Fig. ). RFMIP requested 30-year-long experiments for ERF calculations , following recommendations based on CMIP5 output . That experiment length proved to be sufficient for ERF estimates of most climate forcers in RFMIP, e.g., for ERF of anthropogenic aerosols, although more simulated decades further improve the precision of the ERF calculation . Different from RFMIP, AerChemMIP found that a spin-up time associated with long-lived trace gases, e.g., halocarbons, is necessary before calculating the ERF. This meant that the approach of 30-year-long time slice experiments was not entirely appropriate for the AerChemMIP experiments for all individual climate forcers (Sect. ). The longer spin-up period should be accounted for in future requests for new experiments for ERF calculations of such climate forcers.

The fixed SST method has two advantages compared to the traditional approach, based on coupled atmosphere–ocean experiments . First, the impact of internal variability on ERF estimates is reduced through sufficiently long experiments with prescribed sea surface conditions that are computationally less expensive. Second, the use of fully coupled experiments to estimate ERF relies on a linear relationship between ERF and temperature, which is now known not to be true in general and can lead to ERF estimates that differ from the results of fixed SST experiments .

One weakness of fixed SST experiments in estimating ERF is the adjustment of land surface temperatures. A change in land surface temperatures affects the energy budget, leading to biased estimates of ERF of the order of 10 % . Such unwanted influences on the ERF estimate can be post-corrected . If the capability of fixed land surface temperatures was facilitated in more ESMs, biases in ERF arising from surface temperature adjustments would be virtually eliminated in the future. If adopting the fixed sea and land surface temperature method (Fig. ) in a MIP becomes feasible, the change in the radiation budget would then be equal to the change in the energy budget of the system, which overcomes the limitations of other methods for estimating ERF. Prescribing sea and land surface temperature is different from the experiments carried out for CMIP6 and RFMIP. The requested experiments used prescribed sea surface temperatures and sea ice, following the experimental design of the Atmospheric Model Intercomparison Project (AMIP; ), but the land surface temperatures were freely evolving. Prescribing the sea ice, sea, and land surface temperatures has not been done in a MIP to date.

The radiative forcing of anthropogenic aerosols depends on the optical properties and the effects on clouds. Improved diagnostics and observational constraints in the output analysis for aerosol burden and optical properties would be useful for better understanding the model diversity in the associated radiative forcing and the climate response. As discussed in Sect. 3, the significant diversity across ESMs in the simulated distributions of aerosol burden, optical properties, radiative effects, and the resulting climate responses, including temperature and precipitation, limits building confidence in model projections of climate change. Analysis of relevant and correlated model diagnostics, together with observational constraints, can shed light on the source of diversity in the full cause-and-effect chain and inform improvements in the treatment of aerosols in models. For example, underscores the diversity in aerosol absorption, as the dominant cause of model diversity in historical precipitation changes in CMIP6. RFMIP experiments point to overestimated aerosol absorption from anthropogenic black carbon and a relatively small share of natural aerosol absorption, which leads to direct radiative effects of anthropogenic aerosols in some CMIP6 models which are implausible in light of other lines of evidence . That multi-model assessment was not as broad as it could have been, due to the limited availability of requested output for aerosol properties and diagnostic calls to the radiative transfer scheme for aerosol effects in the CMIP6 models. Wider availability of relevant output from the next phases of RFMIP and AerChemMIP would allow deeper exploration of the intermodel differences in the radiative forcing of anthropogenic aerosols.

There is an opportunity to increase synergies with impact assessments of climate change through improved model diagnostics. A common tropopause diagnostic, included for the first time in CMIP6 models due to AerChemMIP, was available for the calculation of tropospheric O3 burden in a consistent manner. However, the tropopause height was found to vary across the models, and as O3 is found in large concentrations in the stratosphere and upper troposphere, the tropopause definition contributed to the model spread in the calculated tropospheric O3 burden. Through TriMIP, it was identified that a tropopause defined by the O3 mixing ratio results in a smaller model spread in tropospheric O3, which is relevant for air quality assessments.

Moreover, not all ESMs currently output diagnostics for PM, and those models that calculate PM use different formulas and combinations of species. Such differences make any intercomparison of PM between models and observations difficult. To circumvent this issue, AerChemMIP tested estimating PM from model output, following , but associated uncertainties are hard to quantify. Future MIPs could standardize calculations for PM2.5 and PM10 across experiments, e.g., consistent with air quality assessments, following the standards of the World Health Organization. It would allow the use of PM measurements for air quality monitoring as an independent validation dataset and could create a bridge to health impact studies.

Another opportunity to connect more with impact-oriented research can arise from ESM experiments for additional future socioeconomic and mitigation-based pathways, such that uncertainty in emission developments, including mitigation and associated impacts of atmospheric composition changes, can be systematically explored. In addition to new phases of AerChemMIP and RFMIP, examples are a MIP on future methane removal in support of potential climate solutions or on fire emission developments, possibly accounting for the new capability to represent fire feedbacks and leveraging on experiences made in the Fire Model Intercomparison Project (FireMIP; ). If stronger interactions with communities concerned with climate change impacts would be pursued, e.g., with the Vulnerability, Impacts, Adaptation, and Climate Services (VIACS; ) and the Inter-Sectoral Impact Model Intercomparison Project ISIMIP; community, the usage of output from MIPs for societally relevant problems could be enhanced. Such engagement could lead to a better-integrated understanding of links between climate change, extremes, air quality, and the impacts in different sectors, e.g., health, energy, and economics, for climate change preparedness.