| description abstract | new statistical optimization method is used to identify components of surface air temperature and precipitation on six continents that are predictable in multiple climate models on multiyear time scales. The components are identified from unforced ?control runs? of the Coupled Model Intercomparison Project phase 3 dataset. The leading predictable components can be calculated in independent control runs with statistically significant skill for 3?6 yr for surface air temperature and 1?3 yr for precipitation, depending on the continent, using a linear regression model with global sea surface temperature (SST) as a predictor. Typically, lag-correlation maps reveal that the leading predictable components of surface air temperature are related to two types of SST patterns: persistent patterns near the continent itself and an oscillatory ENSO-like pattern. The only exception is Europe, which has no significant ENSO relation. The leading predictable components of precipitation are significantly correlated with an ENSO-like SST pattern. No multiyear predictability of land precipitation could be verified in Europe. The squared multiple correlations of surface air temperature and precipitation for nonzero lags on each continent are less than 0.4 in the first year, implying that less than 40% of variations of the leading predictable component can be predicted from global SST. The predictable components describe the spatial structures that can be predicted on multiyear time scales in the absence of anthropogenic and natural forcing, and thus provide a scientific rationale for regional prediction on multiyear time scales. | |
