To date, virtually all studies attempting to identify human effects on global climate have compared modeled and observed temperature fields. Such comparisons involved temperatures at the Earth's surface or in vertical profiles through the atmosphere. Here, we consider
whether human effects on climate can be more readily identified by considering the joint behavior of a number of different climate variables rather than the behavior of a single variable. Our multi-variable representation of the climate system (in both models
and data) comprises roughly 25-30 indices. The indices are large-scale (global or hemispheric) spatial averages, incorporating information on the atmospheric general circulation, temperature and moisture. Examples include time series of the NAO, ENSO, and the Pacific Decadal Oscillation, land-ocean and hemispheric temperature
contrasts, equator-to-pole temperature gradients, and high-latitude precipitation changes.
Index time series are computed in comparable ways frommodels and observations, using both high-pass and low-pass filtered data. For model data, we compute index time series not only from the spatially-complete fields, but also after mimicking observed coverage changes. The model-based multi-variable signal and noise estimates
are taken from perturbation experiments and a control integration performed recently with the ECHAM4/OPYC coupled model developed in Hamburg. The experiments involve time-varying changes in well-mixed greenhouse gases, tropospheric ozone and sulfate aerosol direct and indirect effects.
We first attempt to identify the model-predicted multi-variable fingerprints in the observations. We then compute EOFs of the normalized modeled and observed index time series, and apply an "optimal detection" strategy to enhance signal-to-noise ratios. Preliminary results indicate that this approach is useful for exploringclimate-change detection and attribution issues, and also offers a useful internal consistency check on the ``between-variable" covariance relationships in models and data