ALBERT

Description: Precipitation including its wet and dry extremes is changing as a consequence of a warming climate, and is modulated by different modes of climate variability. Assessing precipitation changes in the next few decades requires therefore an understanding of its responses to both external forcing and internally generated climate variability. We assess a large ensemble of historically forced simulations from CMIP6 in their fidelity to capture observed precipitation changes, using several skill metrics widely used in the field of climate prediction. We find that these simulations skillfully represent observed mean and extreme precipitation changes over large parts of Europe, Asia, northeastern North America, parts of South America and western Australia, whereas a lack of skill is apparent in western North America and parts of Africa. In particular in regions with moderate skill and where observed precipitation exhibits decadal-scale variations, the use of very large ensembles beyond 100 members can be beneficial to increase the accuracy of simulated precipitation and identify a predictable signal. We then apply a constraint that aligns the phasing of internal variability between the models and the observations at a given time. We demonstrate how this constraint enhances the accuracy of near-term projections for the next 20 years. In particular, the constraint improves near-term projections of drought in areas that have recently been affected by severe and prolonged drought events, such as the southwestern US and Southeast Australia. This analysis highlights regions where observed precipitation changes are explained primarily by forcing, or additionally modulated by decadal climate variability.