Variability of Multisite Decadal Running Means Arising from Year-To-Year Fluctuations

Decadal mean variables are frequently used to characterize decadal climate variabilities. Decadal means are often calculated using yearly data, which can represent variability at time scales from annual to centennial. Residuals from interannual fluctuations may contribute to the variability in decadal time series. Such variability is more difficult to be predicted at the long range. Removing it from the decadal variability means that the remaining variability is more likely to arise from slowly varying multidecadal or longer time scale external forcing and internal climate dynamics, which are more likely to be predicted. Here, a new approach is proposed to understand the uncertainty, potential predictability, and drivers of decadal mean variables. The covariance matrix of multivariate decadal running means is decomposed into unpredictable fast decadal variability and the potentially predictable slow decadal variability. EOF analysis is then applied to the decomposed matrices to find the dominant modes, which may be related to the drivers of the two types of variabilities in the multivariate decadal means. The methodology has been applied to 140-yr datasets of North Pacific sea surface temperature and the Northern Hemisphere 1000-hPa geopotential height. For sea surface temperature, the Pacific decadal oscillation is the major driver of the fast decadal variability, while the radiative forcing and the Atlantic multidecadal oscillation are major drivers of the slow decadal variability. For the 1000-hPa geopotential height, fast decadal variability is associated with the northern annular mode, the east Atlantic mode, and the Pacific decadal oscillation. Slow decadal variability is associated with the northern annular mode and the Atlantic multidecadal oscillation.