New and Improved Climate Change Models

In a paper published in Science, a group of British researchers have managed to greatly improve the predictive power of climate change models by doing some "hindcasts", that is using actual data from the past to adjust and improve the climate models. As adjusted, the models show that we face a very warm decade ahead after a brief cooling period. Quirin Schiermeir provides a very good analysis of the article:

British researchers have substantially improved the performance of a global climate model by adding observations about the actual state of the ocean and atmosphere. The results are of seminal importance for those trying to produce reliable short-term 'climate forecasts' on global and regional scales, experts say.

The team, led by Doug Smith of the UK Met Office's Hadley Centre in Exeter, has developed a climate prediction system that is capable of including natural variability in the climate system — such as that arising from anomalies in ocean circulation or ocean heat content — into modelling carried out by a global climate model. Although most climate models are run with preset boundary conditions, such as the strengths of solar radiation and the level of greenhouse gas, the initial state of the atmosphere and ocean are left undefined, because setting them up with the right initial conditions is time-consuming and costly. "In theory, any climate model could be initialized in this way," says Smith. "It's just a very large piece of work."

To assess the increase in accuracy the new approach offers, the team looked at various ten-year periods from 1982 to 2001, using the new system to predict a decade's worth of global mean temperatures on the basis of the initial conditions and the known changes in greenhouse gas. In these 'hindcasts', the new Decadal Climate Prediction System (DePreSys) reproduced what had actually happened up to 50% more accurately than an otherwise similar model which did not assimilate data on the state of the atmosphere and oceans at the beginning of its runs. The team reports its results in this week's Science1.

When called on to forecast the coming decade, DePreSys produced a 10-year temperature curve distinctly different from those predicted by conventional models. The curve is based on 20 different model runs starting on 20 different days in 2005, each with its own set of initial conditions.

If the model is correct, in the next few years natural variability — mainly in factors affecting the heat content of the ocean — will offset some of the climate warming resulting from humanity's greenhouse emissions. But global warming will be taking only a brief breather: half the years from 2009 to 2014 will be warmer than 1998, which is currently the warmest year on record.

This prediction is still rather tentative, though. While setting up the initial conditions helps, the various inherent difficulties in the model's attempts to capture all the processes going on mean that the system's forecasts will be far from perfect. The predicted temperatures carry healthy error bars, but Smith points out that the flattening it foresaw in the first few years seems so far to be accurate.

The work is getting strong support from other climate modellers. "This is an outstanding study, and conceptually a big leap forward," says Jochem Marotzke, a director at the Max Planck Institute for Meteorology in Hamburg, who oversees the World Climate Research Programme's research on decadal climate predictions. As real-time observations, in particular those of the oceans, are becoming increasingly easily available, quantifying and predicting internal variability in the climate system is now within reach, he adds.

"This study is not primarily important for the brave temperature prediction it makes," he says. "The key thing is that we now have a convincing concept for combining observations and models. It may not be the last word, but it does prove that concrete decadal predictions are possible."

Read it all here. The actual Science article can be found here (subscription required).