In the meantime, UK scientists have found that including data from the top 2000 metres of the ocean on both temperature and salinity, rather than just sea surface temperature as Keenlyside et al did, improves decadal predictions.
"Although the first attempts at initialized decadal climate prediction show some encouraging signs, there are also conflicting results," Doug Smith of the UK Met Office told environmentalresearchweb. "For example, Smith et al (2007) predict that at least half of the years after 2009 will exceed the warmest year currently on record, whereas Keenlyside et al (2008) suggest that natural variability will completely offset anthropogenic global warming until about 2020."
According to Smith, these two studies differ in the data used for initialization: Smith et al use sea surface temperature and sub-surface ocean temperature and salinity, whereas Keenlyside et al only use sea surface temperature.
"Verifying decadal forecasts, and understanding the causes of these conflicting forecasts, is difficult because the observations, especially those beneath the ocean surface which are likely to be critical, are so sparse historically," said Smith.
Data from Argo floats, which can function to a depth of 2000 m, have only been available for the last decade or so. The project reached its goal of deploying 3000 probes in November 2007.
To investigate the importance of the initialization data, Smith and colleague Nick Dunstone turned to idealized experiments, "where we know what the truth is". They say that previous idealized experiments have shown that important aspects of climate, such as variations in the Atlantic meridional overturning circulation (AMOC), are potentially predictable.
"However, these studies assumed a near-perfect knowledge of the initial conditions," said Smith. "In our study we investigate the skill given more realistic observations, using the same procedure to initialize the model as is used to make actual forecasts."
Smith and Dunstone ran their model using initial conditions from a point 150 years into the model future. The pair found that ocean temperature and salinity in the upper 2000 m can successfully be used to initialize and predict the AMOC – and patterns of ocean heat content – on decadal timescales. "However, this is not the case in our model when only sea surface temperature data are used," said Smith. "Observations below 2000 m also provide some improvement, especially in the Southern Ocean. And including atmosphere data improves the skill slightly but only for the first year or so."
Smith stresses that decadal predictions are still in their infancy, and further research is needed to assess their reliability. "In our study we assumed a complete knowledge of ocean temperature and salinity," he said. "This is not achievable in reality, even with the Argo array. We therefore plan to repeat our experiments but only using model temperature and salinity data at locations where we have real observations."
Another important difference between the idealized experiments and real decadal forecasts arises from model errors, says Smith, which introduce biases into both the initialization and forecasts. The researchers plan to simulate this in future "sibling" experiments, as opposed to the "twin" experiments they have used so far. "Twin experiments use the same model as both the truth and for the predictions, thereby eliminating model error," said Smith. "Sibling experiments attempt to predict one model using a different one."
The researchers reported their work in GRL.