As Peter Cox, Met Office Chair in Climate System Dynamics at the University of Exeter, UK, put it, the process was like "arranging the marriage of data on the past and models of the future".

The past can provide many clues about the way Earth’s climate may react in future to the changes imposed by man. "Predictions of future climate change are still very uncertain," said Cox, citing the IPCC temperature rise range for 2100 of 1.1 to 6.4 °C. "We need to narrow that range to be useful for policymakers. We can do shorter timescales or use palaeodata to constrain the models." Cox likened this to constraining the "wild modeller" by a "data cage".

There are three uncertainties in climate prediction, according to Cox. The largest is how human emissions of carbon dioxide will change – unfortunately data from the past can’t help much with this one. Then there is how sensitive the climate is to carbon dioixde, and the reverse problem of how sensitive carbon dioxide (or the carbon cycle) is to climate. As climate changes, land- and ocean-based carbon sinks are likely to become less efficient, and factors such as El Niño and volcanoes can also affect the fraction of carbon dioxide emissions that remain in the air.

"When we’ve been doing climate models we’ve been in denial about the second part – that carbon dioxide is sensitive to climate," said Cox.

Palaeodata can help with these two last uncertainties, by putting limits on estimates of climate sensitivity and climate-carbon cycle feedbacks. According to Corinne Le Quéré of the British Antarctic Survey, models of how much the efficiency of carbon sinks will decrease in the future currently vary by a factor of 10, but using data from the past should help to narrow this range.

"Palaeoclimate shows what the climate system is capable of," said Jess Adkins, associate professor of geochemistry and global environmental science at California Institute of Technology, US. "Even though the exact current state didn’t happen in the past, palaeo helps us learn how climate works – what processes matter and which don’t."

But the climate modelling and palaeoclimate communities generally have different mindsets and skills. Modellers are used to using data straight away, whereas palaeoclimate scientists have to interpret their measurements first in a process that can take up to 10  years.

"Almost always the first thing we do is dig a hole," said Adkins, representing the palaeo community. "The first thing you know is the depth. Converting depth into time is a serious part of the science and is quite difficult. Radioactive decay is one of our most accurate ways of doing it. We’ll measure pretty much anything if we think it will help us.”

There was general agreement that both communities would benefit from an input of scientists fluent in both types of skills.

"We need a new kind of scientist, who is used to interpreting data and comparing models, not just either one thing or the other," said Gavin Schmidt of NASA. "They won’t necessarily have the same type of skills as geochemists or modellers."

The symposium could well have seeded some useful collaborations that will start to provide research results over the next few years. Which is no bad thing, because, as Schmidt said: "We are heading into unknown territory with only the past and physics to guide us. What we have seen in the past so far is trivial compared to what we are anticipating for the future."