Reduction of uncertainty in near-term climate forecast by combining observations and decadal predictions
Abstract. The implementation of adaptation policies requires seamless relevant information about near-term climate evolution, which remains highly uncertain due to the strong influence of internal variability. The recent development of approaches to improve near-term climate information by selecting members from large ensembles – based on their agreement with either observed or predicted sea surface temperature patterns – have shown promising results across timescales from weeks to decades. Here, we propose a new method to provide climate forecasts over Europe by combining information from both observations and decadal predictions through a two-stage member selection from ensembles of climate simulations. Several predictors are tested as observational metrics based on their influence on the European climate variability at annual to decadal timescale. A retrospective evaluation over Europe demonstrates the added value of this method in reducing the spread of uncertainty stemming from both internal climate variability and model uncertainty. This method can outperform both historical simulations and decadal prediction in 5- 10- and 15-year temperature forecasts of winter MED, as well as summer NEU and WCE. Significant skill improvements are visible for 10- and 15-year forecasts of winter Mediterranean surface temperature over land, when using the North Atlantic Oscillation or the Atlantic Multidecadal Variability as predictors in the first selection. The optimal predictor varies by region and should be evaluated on a case-by-case basis. This improved regional climate information supports more targeted adaptation strategies for the coming decades.