What We Publish

LongRunMIP: Motivation and Design for a Large Collection of Millennial-Length AOGCM Simulations

Maria Rugenstein, Jonah Bloch-Johnson, Ayako Abe-Ouchi, Timothy Andrews, Urs Beyerle, Long Cao, Tarun Chadha, Gokhan Danabasoglu, Jean-Louis Dufresne, Lei Duan, Marie-Alice Foujols, Thomas Frölicher, Olivier Geoffroy, Jonathan Gregory, Reto Knutti, Chao Li, Alice Marzocchi, Thorsten Mauritsen, Matthew Menary, Elisabeth Moyer, Larissa Nazarenko, David Paynter, David Saint-Martin, Gavin A. Schmidt, Akitomo Yamamoto, and Shuting Yang · Bulletin of the American Meteorological Society 100, 2551-2570 · 2019

Key finding. LongRunMIP assembles 50 millennial-length simulations from 15 complex climate models at 10 modeling centers, and in its example simulation, an abrupt quadrupling of CO2 in the model CESM104, the 150 years that standard intercomparison protocols simulate capture only 75 per cent of the eventual surface warming, with 88 per cent reached after 1,000 years and roughly another 4,000 years needed to close the last 0.5 watts per square meter of top-of-atmosphere imbalance.

Three stacked panels sharing a time axis that runs from 0 to 5,000 simulation years, for an abrupt quadrupling of CO2 in the model CESM104. The top panel shows CO2 held constant at 1,120 parts per million after a step up from 280. The middle panel shows global mean surface air temperature anomaly jumping to about 3.7 kelvin, reaching about 6 kelvin by year 1,000, then creeping to roughly 6.6 kelvin by year 5,000; an arrow labelled LongRunMIP gain spans everything past year 150. The bottom panel shows top-of-atmosphere radiative imbalance falling steeply from about 3 watts per square meter and approaching zero only near the end. Coloured bands mark the first 150 years simulated under standard protocols, the first 1,000 years, and the remainder.
One model's approach to equilibrium after CO2 is quadrupled, with the 150 years of a standard protocol shaded at the far left against the 5,000 years the model actually takes. Most of the warming appears early; the remainder arrives across a tail that no standard experiment simulates. Figure 1 from Rugenstein et al. (2019), Bulletin of the American Meteorological Society 100, 2551-2570. Reproduced under author reuse rights. Extracted from the published PDF and resized for web display.

What question did this research address?

Standard model intercomparison protocols run abrupt CO2 experiments for 150 years. Yet the questions that motivate them — what equilibrium climate sensitivity is, how the deep ocean warms, how radiative feedbacks change as the pattern of warming evolves — are questions about what a model does over millennia.

Millennial simulations existed before this work, but scattered across research groups, run at different forcing levels and lengths, and reported in papers that examined different aspects of the climate system. This paper set out to gather them into one standardized, public archive and to say what that archive can and cannot settle.

What did we find?

The archive is deliberately inclusive rather than protocol-driven. The minimum contribution is one simulation of any CO2 forcing scenario held constant for at least 1,000 years, from a CMIP5-class model with a dynamic atmosphere, ocean, and sea ice, plus a control run. Most contributions are step forcings to two, four, or eight times preindustrial CO2.

Because the simulations were run independently by their modeling groups rather than to a common protocol, the collection samples the CMIP5 range of models well but is uneven in length, forcing, and stored variables. The authors standardize units, sign conventions, and grids, and leave it to each user to decide how equilibrated a simulation must be for their question.

Surface temperature and top-of-atmosphere radiation reach a new steady state in most models within a few millennia. The deep ocean does not. Even several thousand years is insufficient for it to equilibrate, which is why the archive's own definition of equilibrium is left to the user rather than fixed.

That gap between the surface and the deep ocean is what makes short runs misleading. Radiative feedbacks depend on the evolving spatial pattern of warming, so a constant effective climate sensitivity estimated from a truncated simulation is an inadequate assumption rather than a good approximation.

The paper is explicit about what the models leave out. They carry the fast feedbacks — water vapor, lapse rate, clouds, and sea ice — but not ice sheets, carbon cycle changes, or orbital variation, so comparisons with paleoclimate proxies must account for those missing Earth-system feedbacks.

Why does it matter?

Equilibrium climate sensitivity is usually inferred by extrapolating from short simulations. Only fully equilibrated runs can test whether those extrapolation methods work, and this archive is the first collection large enough to make that test across many models.

The result also reframes what a 150-year experiment is. It is not a truncated view of the same answer but a measurement of a different quantity, because the feedbacks themselves change as warming spreads.

As infrastructure rather than a single result, the archive's value is that it is standardized and public, so questions about long-term equilibration can be asked without each group first spending millennia of computer time.

Citation

Maria Rugenstein, Jonah Bloch-Johnson, Ayako Abe-Ouchi, Timothy Andrews, Urs Beyerle, Long Cao, Tarun Chadha, Gokhan Danabasoglu, Jean-Louis Dufresne, Lei Duan, Marie-Alice Foujols, Thomas Frölicher, Olivier Geoffroy, Jonathan Gregory, Reto Knutti, Chao Li, Alice Marzocchi, Thorsten Mauritsen, Matthew Menary, Elisabeth Moyer, Larissa Nazarenko, David Paynter, David Saint-Martin, Gavin A. Schmidt, Akitomo Yamamoto, and Shuting Yang (2019). LongRunMIP: Motivation and Design for a Large Collection of Millennial-Length AOGCM Simulations. Bulletin of the American Meteorological Society 100, 2551-2570.

Related