Reduced global warming from CMIP6 projections when weighting models by performance and independence


Lukas Brunner, Angeline G. Pendergrass, Flavio Lehner, Anna L. Merrifield, Ruth Lorenz and Reto Knutti

Earth System Dynamics
Work Package 2



Improving our projections of future climate change is key to preparing for the impacts it may bring. This study investigates whether giving climate models different levels of weight based on their similarity and ability to simulate the past will improve their ensemble climate projections. The team found a 17% increase in skill in the weighted ensemble over the historical period, and slightly lower projections of average future warming than when using an unweighted ensemble (3.7°C against 4.1°C by 2100 in a high-climate change scenario). The uncertainty range of the weighted ensemble was also narrower. This represents a useful approach to potentially improve how we use our future climate projections, allowing policymakers to enhance their strategies to mitigate and adapt to the impacts of future climate change.

Projections of how the climate may change in the future are crucial to developing successful mitigation and adaptation policies to limit and manage the risks of climate change. Our latest generation of climate models produce an ensemble of projections, with several ensembles produced in a recent international exercise called CMIP6. Projections from these ensembles are often represented as an average result with a range of uncertainty. This range is often based on the spread of all the models. The constituents of the ensemble are often given equal importance in the final projection, but is this the best way to produce our climate projections? This study uses the Climate model Weighting by Independence and Performance (ClimWIP) method to assign individual models more or less importance (weight) in the final projection, based on how well they simulate the past and how independent they are of other models. This is an important opportunity to improve how our climate projections are used, allowing policymakers to make better-informed decisions on which actions to take in future.

The team found that using the ClimWIP method revealed those CMIP6 models which suggest the most warming to be less probable. This brings down the warming projections for the weighted ensemble. The weighted ensemble projects mean average warming in a high greenhouse gas emission scenario (SSP5-8.5) by 2081-2100, relative to 1995-2014, of 3.7°C, compared to 4.1°C without weighting. The likely (66% chance) range of the projections is 3.1°C to 4.6°C, a 13% decrease in the spread of the unweighted projections. In a future scenario with much lower greenhouse gas emissions (SSP1-2.6) the warming by the end of the century is projected to be 1°C, 0.1°C lower than the unweighted ensemble. The range in this case is 0.7°C to 1.4°C, a 24% decrease. The weighted ensemble had around a 17% increase in skill at simulating the past than the unweighted ensemble when its continuous ranked probability skill score was assessed in a perfect model test.

This study used 129 simulations from 33 CMIP6 models. These simulate climatic changes between 1850 and 2100, with this study focussing on the period after 1980. Part of the study investigates the skill of the weighting method by using CMIP5 models to create pseudo-observations. Each pseudo-observation is used to inform the weighting in the historical period (1995-2014) and the weighted CMIP6 distribution is subsequently evaluated against it in the future. For the main results, the same approach was repeated using real observations from the ERA5 and MERRA2 datasets, produced by ECMWF and NASA, respectively. The other part of the weighting system used the ‘relatedness’ of the models. Climate models can share concepts, code, even whole components between them, and may be derived from each other to focus on a particular question. This means that certain models may not be truly independent and treating them as such may amount to double counting in ensemble predictions. Models were placed on a family tree and closely related models were given less weight, avoiding giving their shared components undue influence. It is important to note that this method does assume that a model with good present-day simulations will also have good future projections.

Policy relevance
The results presented here provide a promising way to improve our projections of how the climate may change in the future, using our climate models most effectively. The projections presented in this paper also highlight the urgency of action to meet the Paris Agreement targets. Improved future climate predictions will help policymakers formulate effective strategies to adapt to and mitigate the effects of future climate change.


The sixth Coupled Model Intercomparison Project (CMIP6) constitutes the latest update on expected future climate change based on a new generation of climate models. To extract reliable estimates of future warming and related uncertainties from these models, the spread in their projections is often translated into probabilistic estimates such as the mean and likely range. Here, we use a model weighting approach, which accounts for the models’ historical performance based on several diagnostics as well as model interdependence within the CMIP6 ensemble, to calculate constrained distributions of global mean temperature change. We investigate the skill of our approach in a perfect model test, where we use previous-generation CMIP5 models as pseudo-observations in the historical period. The performance of the distribution weighted in the abovementioned manner with respect to matching the pseudo-observations in the future is then evaluated, and we find a mean increase in skill of about 17 % compared with the unweighted distribution. In addition, we show that our independence metric correctly clusters models known to be similar based on a CMIP6 “family tree”, which enables the application of a weighting based on the degree of inter-model dependence. We then apply the weighting approach, based on two observational estimates (the fifth generation of the European Centre for Medium-Range Weather Forecasts Retrospective Analysis – ERA5, and the Modern-Era Retrospective analysis for Research and Applications, version 2 – MERRA-2), to constrain CMIP6 projections under weak (SSP1-2.6) and strong (SSP5-8.5) climate change scenarios (SSP refers to the Shared Socioeconomic Pathways). Our results show a reduction in the projected mean warming for both scenarios because some CMIP6 models with high future warming receive systematically lower performance weights. The mean of end-of-century warming (2081–2100 relative to 1995–2014) for SSP5-8.5 with weighting is 3.7 C, compared with 4.1 C without weighting; the likely (66%) uncertainty range is 3.1 to 4.6 C, which equates to a 13 % decrease in spread. For SSP1-2.6, the weighted end-of-century warming is 1 C (0.7 to 1.4 C), which results in a reduction of −0.1C in the mean and −24 % in the likely range compared with the unweighted case.