Limits of historical reconstructions using indirect evidence Hans von Storch, Eduardo Zorita, Julie Jones, Institute for Coastal Research, GKSS Research Centre, Germany Fidel González-Rouco University Complutense Madrid, Spain Hans von Storch, Eduardo Zorita, Julie Jones, Institute for Coastal Research, GKSS Research Centre, Germany Fidel González-Rouco University Complutense Madrid, Spain 9IMSC, Cape Town May 2004

Limits of historical reconstructions using indirect evidence Hans von Storch, Eduardo Zorita, Julie Jone Institute for Coastal Research, GKSS Research Centre, D Geesthacht, Germany Abstract We test how much variability at centennial timescales is lost when past climates are reconstructed from proxy data. We use data from an atmosphere-ocean climate model simulation of the last 1000 years, forced with estimates of natural and anthropogenic forcings, and the method of Mann et al. (1998) (1) to estimate Northern Hemisphere temperature from grid-point temperatures (with white noise added) in realistic locations. We find that the low frequency variability is greatly underestimated when the predictors contain 50% or more noise. The inclusion of more predictors from Africa and Asia results in only minor improvement. 9th International Meeting on Statistical Climatology May 2004 Cape Town, South Africa

ECHO-G simulations „Erik den Røde” ( ) and “Christoph Columbus” ( ) with estimated volcanic, GHG and solar forcing

vs Reconstruction from historical evidence, from Luterbacher et al. Late Maunder Minimum Model-based reconstuction

Christoph C Erik dR

Both, Erik den Røde and Christoph Columbus generate temperature variations considerably larger than standard reconstructions (Mann, Jones …). The simulated temperature variations are of a similar range as derived from NH summer dendro-data and from terrestrial boreholes.

For the present context, it does not really matter how „good“ the historical climate is reproduced by Erik den Røde and Christoph Columbus. The model data provide a laboratory to test MBH and other methods.

pseudo-proxies: grid point SAT plus white noise red: mimicking largest sample used in MBH blue: hypothetic additional data to obtain better coverage. Testing the MBH method

Mimicking MBH?

Estimated NH SAT from terrestrial deep soil temperature (TDST) in Erik den Røde.

Conclusion, 1 Erik den Røde, an effort to simulate the response to estimated volcanic, GHG and solar forcing, Low-frequency variability in Erik den Røde > Mann, Jones, & mainstream, but ~ Esper, boreholes, (some) instrumental data

Conclusion, 2 Erik den Røde-data used to test MBH and borehole approaches. Randomized grid-point SAT (i.e. white noise added) is used as pseudo proxy. MBH method, based on regression, gives significant underestimation of low-frequency NH mean SAT. Direct averaging of randomized local data results in considerably smaller errors in NH mean SAT.

Conclusion, 3 When using Erik den Røde-data and pseudo- proxy data, which share a correlation of about 0.5 with the grid point data, the best guess resembles the MBH estimate (incl. the ±2σ confidence band). The problem is common to ALL regression based methods, trained with temporal high-resolution data – except if the correlations are really high. A reliable reconstruction of centennial time scales requires either really high correlation, or process-based inverted data (e.g., borehole inversions), or local instrumental data.

Testing with HadCM3 simulation Erik HadCM3 Data provided by Simon Tett.