Methodologies will be developed (and tested on existing data) by partners 1, 4, 5 and 7 to allow a like-with-like comparison between palaeo and model data, despite the different characteristics of each data type (model output is often less reliable at small spatial scales, while some proxy data are representative of only single sites; palaeo reconstructions are not exact and have a quantifiable uncertainty range associated with them; and proxy data include noise, with varying spatio-temporal characteristics, due to processes unrelated to large-scale climate).
The forced model response will be evaluated by comparing simulated and proxy variability for a variety of diagnostics identified in workpackages 2 and 3 (i.e., European and hemispheric temperature, European and USA precipitation, climate indices and statistics of, e.g., atmospheric circulation patterns such as NAO and coupled climate modes such as ENSO). Comparison of variance, spectra, spatial patterns as defined by principal component analysis, etc., will be undertaken by partners 1, 2, 6 and 7, applying the methods developed earlier to take into account scale/error/noise characteristics of the data. In addition, signal detection and attribution techniques will be applied by partners 3 and 4 to attempt to detect, in the proxy data, the large-scale simulated responses to external forcing.
Comparison of simulated and reconstructed data will be a two-way process, with simulations also providing input into the critical assessment of the reconstructed climate data sets. Interpretation of the proxy data will be achieved by utilising the simulated patterns and timing of climate response to external forcing for the attribution of proxy variations to specific causes (partners 5, 6 and 7). Model output will be sampled (for specific locations, seasons and variables) and various statistical noise models added, to yield synthetic climate proxy records. Reconstruction and calibration techniques will then be tested to try to capture the (known) simulated climate for various regions and variables/indices, allowing an assessment of the suitability of various palaeodata networks (with given uncertainty/noise levels) for reconstructing past climate - especially given the existence of forced climate changes during the 20th century calibration period (partners 1 and 5).
Partner 4 will develop and apply innovative methods to merge the proxy and model data to obtain improved estimates of natural climate variability taking into account uncertainties in the palaeo data and the models. For this purpose, we will evaluate Bayesian analysis methods where prior (e.g., model-derived) beliefs can be updated in the light of new (proxy) data and statistical reconstruction techniques that, in previous studies, rely on the covariance structure of climate estimated using the relatively short, and possibly anthropogencially-contaminated, instrumental records of the 20th century.
These new estimates of natural variability, as well as the separate estimates from the palaeo and model data, will be used by partners 3 and 4 in a detection and attribution framework to interpret proxy data in the light of simulated climate change and to see if recent climate changes are unusual (for different variables and on different spatial and time scales, as dictated by the availability of different proxies) in the context of natural climate variability over the last 500 years. These results will be compared to earlier detection and attribution studies which mostly derived natural variability from long unforced control integrations of climate models.