Detecting climate signals using space-time EOFs
Academic Article
Overview
Research
Identity
Additional Document Info
Other
View All
Overview
abstract
Estimates of the amplitudes of the forced responses of the surface temperature field over the last century are provided by a signal processing scheme utilizing space-time empirical orthogonal functions for several combinations of station sites and record intervals taken from the last century. These century-long signal fingerprints come mainly from energy balance model calculations, which are shown to be very close to smoothed ensemble average runs from a coupled ocean-atmosphere model (Hadley Centre Model). The space-time lagged covariance matrices of natural variability come from 100-yr control runs from several well-known coupled ocean-atmosphere models as well as a 10 000-yr run from the stochastic energy balance climate model (EBCM). Evidence is found for robust, but weaker than expected signals from the greenhouse [amplitude 65% of that expected for a rather insensitive model (EBCM: T2xCO2 2.3C)], volcanic (also about 65% expected amplitude), and even the 11-yr component of the solar signal (a most probable value of about 2.0 times that expected). In the analysis the anthropogenic aerosol signal is weak and the null hypothesis for this signal can only be rejected in a few sampling configurations involving the last 50 yr of the record. During the last 50 yr the full strength value (1.0) also lies within the 90% confidence interval. Some amplitude estimation results based upon the (temporally smoothed) Hadley fingerprints are included and the results are indistinguishable from those based on the EBCM. In addition, a geometrical derivation of the multiple regression formula from the filter point of view is provided, which shows how the signals "not of interest" are removed from the data stream in the estimation process. The criteria for truncating the EOF sequence are somewhat different from earlier analyses in that the amount of the signal variance accounted for at a given level of truncation is explicitly taken into account.
