Climate models predict significant changes to the Earth’s climate in the coming century, but the predictions range greatly, signifying high levels of uncertainty.
Uncertainty can be problematic as climate predictions form the basis for policy changes. With high uncertainty, there is an increased probability that the models may over-estimate or under-estimate the speed and scale of climate change. If the models over-estimate it, we may invest huge amounts of money trying to avert a problem which isn’t as serious as the models suggest. Alternatively, if the models under-estimate the change, we will end up doing too little, too late in the mistaken belief that the changes in weather patterns will be small and gradual. To cope with this problem, we need to quantify the uncertainty in these predictions.
There are two complementary approaches to assess and reduce uncertainty:
- Improve the parameterisations while narrowing the range of uncertainty in the parameters. This continuous process requires:
i) Improving the models by using the latest supercomputers as they become available.
ii) Gathering more and more (mainly satellite) data on a wide range of atmospheric variables (such as wind speed, cloud cover, temperature…).
- Carry out large numbers of model runs in which the parameters are varied within their current range of uncertainty. Reject those which fail to model past climate successfully and use the remainder to study future climate.
The second approach is the one taken by climateprediction.net. Our intention is to run hundreds of thousands of state-of-the-art climate models with slightly different physics in order to represent the whole range of uncertainties in all the parameterizations. By running the model thousands of times, we hope to find out how sensitive individual models are to small disturbances, as well as to changes in carbon dioxide and sulphates in the atmosphere. Whilst these tweaks are slight enough not make the approximations any less realistic, they allow us to explore how climate may change in the next century under a wide range of different scenarios.
This technique, known as ensemble forecasting, requires an enormous amount of computing power. Even with the incredible speed of today’s supercomputers, climate models have to include the effects of small-scale physical processes (such as clouds) through simplifications (parameterizations). As mentioned previously, there is a large range of uncertainty in the precise values of many of the parameters used, meaning we do not know precisely what value is most realistic. Considering a range can occasionally be an order of magnitude, any single forecast represents only one of many possible ways the climate could develop.
The only practical solution is to appeal to distributed computing, which combines the power of thousands of ordinary computers, each of which tackles one small but key part of the global problem. By using your computers, we will be able to improve our understanding of, and confidence in, climate change predictions more than would ever be possible using the supercomputers currently available to scientists. The climateprediction.net experiment should help to “improve methods to quantify uncertainties of climate projections and scenarios, including long-term ensemble simulations using complex models”, identified by the Intergovernmental Panel on Climate Change (IPCC) in 2001 as a high priority.
The aim of these experiments is to give decision-makers a better scientific basis for addressing one of the biggest potential global problems of the 21st century: climate change. By looking at the uncertainties associated with each model, we can selectively use the best models to make predictions for the future, thereby giving policy makers an accurate picture to date of climate predictions.