Just when everyone was getting sick of explaining that climate models are producing projections not predictions per se, it seems that some of them are indeed producing predictions. There’s a paper (pdf) in Science from a team at the Hadley Center that shows how using real initial conditions improves the accuracy of ten year climate forecasts. They do a bit for hindcasting first, looking at historical data and comparing model runs with real initial conditions with run-of-the-mill runs. Then they do some prediction. This prediction is being treated as saying that we’re at the end of a little plateau, and that at the end of this decade things will warm up further, giving a run of years in the early 2010s where the chances for new global records are good. Quirin Schiermeier wrote a story on this for news@nature, reporting that the modelling community seems pretty impressed. Here’s a bunch more coverage (88 pieces at the time of googling), and for those with a subscription to Science here’s the estimable Dick Kerr, who had longer to write the story than the rest of us…