Two new papers co-authored by Past to Future (P2F) researchers have set the experimental rulebook that climate modelling centres around the world will use as they run the next generation of global climate models, known as CMIP7.
Climate models are the tools scientists use to project how the Earth’s climate will change in the future. But before dozens of different research groups around the world can run their models and meaningfully compare the results, they first need to agree on exactly what each model should simulate—which time periods, which conditions, which questions to ask. This shared rulebook is called an experimental design, and getting it right is essential groundwork for everything that follows.
P2F researchers led the design of two such experiments, both of which are now published in the journal Geoscientific Model Development.
The first sets out how modelling centres should simulate the early Eocene, a period around 50 million years ago when the Earth was significantly warmer than today and had very high atmospheric carbon dioxide concentrations.
The early Eocene provides a window into a super-warm past climate state of our planet. Climate models that are used to predict the future can be tested under these extreme conditions, as we have geological data that tells us how warm this past world actually was.
Professor Dan Lunt, University of Bristol
Testing models against a period like this only works if the model set-up itself is accurate. That means feeding in the best possible representation of where the continents sat, how high the mountains stood, and what vegetation covered the Earth’s surface at the time. The new paper provides an updated set of these inputs, or “boundary conditions”, that allow modelling groups to configure their simulations as realistically as possible. As a result, confidence in the correct setup of the models improves, and any differences between the model simulation outputs and the geological data from past climates are not due to errors in these inputs. Within P2F, these improved boundary conditions will feed directly into more accurate simulations of past climate and, in turn, better simulations of the future.
The second paper defines how models should simulate a future with a sea-ice-free Arctic, using evidence from a past warm period roughly 127,000 years ago known as the Last Interglacial.
Abrupt-127k brings together the paleoclimate and sea-ice modelling communities to address a critical question: how well do our climate models represent a seasonally ice-free Arctic? The answer is central to our confidence in projections of future climate change and Arctic sea-ice loss.
Dr Louise Sime, British Antarctic Survey
More than ten international modelling groups are now planning to run the new abrupt-127k experiment using their state-of-the-art CMIP7 models, giving the next Intergovernmental Panel on Climate Change (IPCC) assessment (AR7) a unique out-of-sample test against a known warm Arctic climate state. “By examining how these models simulate a seasonally ice-free Arctic during the Last Interglacial,” Dr Sime adds, “this new international community will better assess the model representation of processes that will shape future Arctic sea-ice loss.”
Both experiments are now part of CMIP7’s official design, meaning modelling groups across the world will use them as they begin running simulations through 2026 and beyond. The first results are expected from late 2026 into 2027.
By using deep-past climate evidence to shape how today’s models are tested, this work reflects one of P2F’s central aims: making the past a genuine tool for building confidence in projections of the future.
Read the papers:
Lunt, D. J. et al.: DeepMIP-Eocene-p2: Experimental design for Phase 2 of the early Eocene component of the CMIP7/PMIP7 Deep-time Model Intercomparison Project (DeepMIP-Eocene), Geoscientific Model Development, 9 July 2026. https://doi.org/10.5194/gmd-19-6143-2026
Sime, L. C. et al.: A sea ice free Arctic: CMIP7 Assessment Fast Track abrupt-127k experimental protocol and motivation, Geoscientific Model Development, 7 July 2026. https://doi.org/10.5194/gmd-19-5881-2026