First published in Nature 453 May 2008
http://www.nature.com/climate/2008/0805/full/453043a.html
Published online: 30 April 2008 | doi:10.1038/453043a
Climate change: Natural ups and downs
Richard Wood1
The effects of global warming over the coming decades will be modified by shorter-term climate variability. Finding ways to incorporate these variations will give us a better grip on what kind of climate change to expect.
Climate change is often viewed as a phenomenon that will develop in the coming century. But its effects are already being seen, and the Intergovernmental Panel on Climate Change recently projected that, even in the next 20 years, the global climate will warm by around 0.2 °C per decade for a range of plausible greenhouse-gas emission levels1. Many organizations charged with delivering water and energy resources or coastal management are starting to build that kind of warming into their planning for the coming decades. A confounding factor is that, on these timescales, and especially on the regional scales on which most planning decisions are made, warming will not be smooth; instead, it will be modulated by natural climate variations. In this issue, Keenlyside et al. (page 84)2 take a step towards reliably quantifying what those ups and downs are likely to be.
Their starting point is the ocean. On a timescale of decades, this is where most of the ‘memory’ of the climate system for previous states resides. Anomalously warm or cool patches of ocean can be quite persistent, sometimes exchanging heat with the atmosphere only over several years. In addition, large ocean-current systems can move phenomenal amounts of heat around the world, and are believed to vary from decade to decade3, 4.
To know and predict the state of the ocean requires an approach similar to weather forecasting: one sets up (initializes) a mathematical model of the climate system using observations of the current state, and runs it forwards in time for the desired forecast period. With a given climate model, enough observations to set the ball rolling and a large-enough computer to move it onwards, the exercise is conceptually straightforward.
But does it actually produce anything useful? We don’t expect to be able to predict the details of the weather at a particular time several years in the future: that kind of predictability runs out after a week or two. But even predicting, say, that summers are likely to be unusually wet during the coming decade would be useful to many decision-makers. Only recently, with the study from Keenlyside et al.2 and another from researchers at my own institution5, have climate modellers begun to explore whether such predictions are possible.
Keenlyside and colleagues’ model2 uses a very simple ocean initialization method in which they add heat to or remove it from the ocean surface until sea surface temperatures across the globe are close to observed values. They use their model to produce a set of retrospective ‘forecasts’ starting from earlier states, which they test against what actually happened. Their system produces refined temperature predictions a decade ahead for large parts of Europe and North America.
The enhanced accuracy of the model seems to stem from a greater ability to simulate natural variations in the meridional overturning circulation (MOC). This is a giant conveyor belt that brings warm water northwards into the North Atlantic, releases its heat to the atmosphere, and returns the cooled water to the south. There is evidence that the strength of this circulation can fluctuate naturally over periods of decades3; when it is strong, the climate in the North Atlantic region passes through a warm phase.
The authors use their model to predict that the MOC will weaken over the next decade, with a resultant cooling effect on climate around the North Atlantic. Such a cooling could temporarily offset the longer-term warming trend from increasing levels of greenhouse gases in the atmosphere. That emphasizes once again the need to consider climate variability and climate change together when making predictions over timescales of decades (Fig. 1).
Figure 1: Heat up?
Figure 1 : Heat up? Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, or to obtain a text description, please contact npg@nature.com
These three possible trends of winter temperature in northern Europe from 1996 to 2050 were simulated by a climate model using three different (but plausible) initial states6. The choice of initial state crucially affects how natural climate variations evolve on a timescale of decades. But as we zoom out to longer timescales, the warming trend from greenhouse gases begins to dominate, and the initial state becomes less important. Keenlyside and colleagues2 use observations of the sea surface temperature to set the initial state of their model. Their results indicate that, over the coming decade, natural climate variability may counteract the underlying warming trend in some regions around the North Atlantic. (Figure courtesy of A. Pardaens, Met Office Hadley Centre).
Full figure and legend (36 KB)
These results provide encouragement that such predictions may be possible, but substantial points require clarification. Chief among these is whether the authors’ initialization, which takes into account only sea surface temperatures, is in fact suitable to characterize the state of the MOC. The MOC extends to a depth of several kilometres and depends not just on temperature, but on the salinity of the ocean water. The answer will be some time in coming, as regular monitoring of the MOC has only just begun4.
In addition, there is the fact that, although it seems to improve predictions around the North Atlantic compared with models that have no initialization, the model’s accuracy is less over other regions such as the tropical North Atlantic and central Africa. This might be because of deficiencies in the climate model or in the initialization procedure. The two studies that have attempted decadal-scale modelling to date2, 5 differ in the regions for which their predictions are most accurate, so clearly there is much still to be understood.
Keenlyside et al.2 have so far used their model to predict decadal average temperature only. If it can be shown to accurately forecast other variables such as precipitation, as well as their variation during specific seasons, its range of applications would be greatly increased. But climate predictions for a decade ahead will always be to some extent uncertain. There are inherent limits to the predictability of climate variability, and unpredictable perturbations to the climate, such as volcanic eruptions, cannot be factored in; predictions must inevitably be probabilistic in nature.
Even so, the first attempts at decadal prediction2, 5 suggest that reasonably accurate forecasts of the combined effects of increasing greenhouse-gas concentrations and natural climate variations can be made. Climate scientists are gearing up to test and extend these ideas over the coming years, in the hope that we can then plan more confidently for the future.
Top of page
References
Author affiliation
1. Richard Wood is at the Met Office Hadley Centre, FitzRoy Road, Exeter, Devon EX1 3PB, UK. e-mail: richard.wood@metoffice.gov.uk