Optimal climate change signal detection using space-time empirical orthogonal functions

Space-time optimal filtering is used to estimate the response of the Earth's surface temperature to both natural and anthropogenic climate forcings over the past century. In this study, a variety of site/record-length configurations is used to represent the climate change signals, natural variability and observational data streams. The hypothesized space-time patterns of the response to the climate forcings are generated from an energy balance climate model (EBCM) and four coupled ocean/atmosphere general circulation models (GCMs). The natural variability statistics are estimated from 1000-year control runs from four coupled GCMs and then such natural variability is represented precisely in terms of the space-time empirical orthogonal functions (EOFs), obtained directly from the lagged space-time covariance matrix on the approximate time interval. Optimal filters are constructed using the hypothesized space-time patterns along with the EOF-represented natural variability. The constructed filters are then applied to the observational surface temperature data. With a proposed orthogonalization process, signals ‘not-of-interest’ are optimally removed from the data stream in the estimation process.; Comparisons show that the responses to the anthropogenic climate forcings generated from the EBCM and four GCMs are very similar to each other. A robust greenhouse-gas signal is detected in the observational surface temperature data, but with a weak amplitude only about 60–70% of that expected from the hypothesized signal patterns. An anthropogenic aerosol signal is very weak and not statistically significant. The estimated amplitude of aerosol indicates that four GCMs in this study might overpredict the response to greenhouse gas-plus-aerosol forcing by a factor of two to five. A volcanic signal is also robust but with about 65% expected amplitude. A solar-cycle signal is significant at a level of 90% and somewhat stronger than the amplitude of the EBCM-predicted response to the 11-year component of the solar luminosity variations.; A simplified version of NCAR Community Climate Model (CCM2) is used to generate natural variability and an artificial solar signal for the optimal filtering scheme. A theoretical examination of whether detection performance could be enhanced by the addition of vertically distributed spatial information rather than information only on the surface is presented.