"In our model, internal variability partially offsets the global warming signal for the first few years after 2005, but the climate continues to warm thereafter," Doug Smith of the Met Office told environmentalresearchweb. "The new Met Office Decadal Climate Prediction System (DePreSys) potentially provides very valuable guidance to help planners and businesses adapt to climate change."

The natural internal variability of the climate system arises from factors such as El Niño, fluctuations in the thermohaline circulation, and changes in ocean heat content. The new model is able to take these parameters into account by including data about the present state of the ocean and atmosphere, something that's been difficult to do in the past because of a scarcity of data for the ocean.

"Including initial condition information, in order to predict internal variability, substantially improves the skill of temperature predictions throughout a decade," said Smith.

The researchers used the system to make a ten-year forecast from June 2005. The model correctly predicted the relative cooling in the Southern Ocean and tropical Pacific over the last two years. It also predicts that the year 2014 will be 0.30 +/- 0.21 °C warmer than 2004, and that at least half the years after 2009 will be hotter than 1998, the warmest year currently on record.

What's more, it's easier to check the accuracy of short-term predictions. "Centennary-scale predictions of anthropogenically forced climate change cannot be verified, because there is no past analogue against which to assess the models," explained Smith. "Predictions of decadal climate change and variability, on the other hand, can be assessed in tests for past cases. Such tests indicate that the model predictions agree very well with observations."

So what's next? "Having only looked at temperature so far, we plan to assess the skill of other climate-impact related variables, particularly precipitation, river flow, drought, sea level, crop growth and storms," said Smith. "We are also developing our system to include uncertainties arising from model errors in addition to those coming from imperfect initial conditions."

The researchers reported their work in Science.