The study reinforces the notion that certain poorly understood systems such as ice sheets or vegetation are integral to accurately predicting future temperatures. It also paints an ever-bleaker outlook for our planet at a critical time when world leaders are gathering for a United Nations conference in Copenhagen to discuss practicable ways of mitigating climate change.

"If we want to build an agreement that is going to last for many, many centuries – so for our grandchildren's grandchildren's grandchildren's grandchildren – then we need to be taking in these issues," lead author Dan Lunt of the University of Bristol told

A fiendishly complicated system

Modelling climate changes across the whole Earth system is a difficult task that requires fluid-dynamics equations to be solved all over a rotating sphere at small increments in time. The models must also account for relatively small-scale phenomena such as clouds, and interactions between, for example, the atmosphere, ocean and biosphere. One of the most challenging aspects of climate modelling is to factor in the processes that evolve over thousands, even millions, of years.

Yet with even with the best supercomputers certain systems have proved too complex to model accurately, or evolve too slowly to reach equilibrium in a simulation's duration. As a result no one is sure what the true
response of these systems might be to mounting CO2 emissions.

Lunt's group – which also includes members from the University of Leeds, Northumbria University, the British Antarctic Survey, NASA and the US Geological Survey – has tackled this problem by trying to unravel the elements of an ancient climate retrospectively. They studied a period in the Earth's "mid-Pliocene" period three million years ago for which they have long-term data on temperature and some of the more troublesome systems, such as ice sheets and vegetation.

Warming underestimated

The researchers discovered that the model gave the correct retrospective temperature predictions only when ice-sheet and vegetation data for the mid-Pliocene period were included. Surprisingly, when they ran the model with more modern ice-sheet and vegetation data, the retrospective predictions underestimated the mid-Pliocene period's global warming by 30–50%.

Although this result highlights how integral some slow systems are for accurate long-term predictions, Lunt is quick to point out that his group cannot say how these factors could affect short-term predictions. "We're not saying that in a decade the temperature will be 30–50% more than old predictions would be," he says. "What we are saying is that the predictions of the climate's equilibrium state are likely to be underestimates by that much."

Reto Knutti, a climate scientist at ETH Zurich, thinks it is an "important" study, but agrees it does not tell when the extra warming would come into effect. "[It] confirms in a more quantitative way what people have been speculating: that the sensitivity of temperature to CO2 could be significantly larger if slow feedbacks are included," he says. "What is missing at this point is an estimate of timescales, i.e., whether these slow feedbacks become important after a few centuries or after thousands of years."

"This is certainly very exciting science," says Gabriele Hegerl, a climate scientist at the University of Edinburgh. However, she also believes that it might have little relevance to present discussions on mitigation, because short-term climate change is likely to be governed more by our CO2 emissions. "These estimates are helpful, but they can only be guides," she adds.

This research is published in Nature Geoscience.