Show from pseudo-proxy studies that previous reconstruction methods underestimate the low-frequency variability of NH temperature. Formulate a new method that preserve the low-frequency variability. Test new method with pseudo-proxies. Apply the new method to compilations of real proxies. Ensemble pseudo-proxy method: The reconstruction methods are tested on many temperature fields generated from a single climate model experiment. Pseudo-proxies generated by adding noise to local temperatures. A reconstruction method that does not underestimate low-frequency variability Bo Christiansen, Danish Meteorological Institute
Surrogate worlds based on a single AOGCM experiment with the usual forcings (GLIMPSE, Stendel et al. 2006). Retain spatial and temporal correlation structures as well as annual variability and trend. Possible to increase amplitude,A, of variability. A=2 A=3 OriginalA=1 New surface temperature fields Surrogate worlds represent the world as it could have been with the same forcing but different phases of the noise
Training Direct field Indirect field Direct global Indirect global CCA RegEM Target 57 proxies, white-noise proxies A=2 A=3 Pseudo-proxy experiments Christiansen et al., J. Clim. 2009
Training 57 proxies White-noise proxies Pseudo-proxy experiments: Reconstructing a trend A= 1, 2, 3 A=2 Direct field Indirect field Direct global Indirect global CCA RegEM Relative Trend A=1A=3 Christiansen et al., J. Clim. 2009, see also Comment and Reply 2010
How to avoid variance loss
The new method: LOC 1.Select the proxies that explain a significant part of the local temperature in the calibration period. 2.Reconstruct local temperatures by 1-dim. indirect regression, i.e., use the local temperature as the independent variable. 3.Make the spatial average.
Pseudo-proxy experiments Realistic noise 22 proxies Training
Pseudo-proxy experiments 57 proxies Realistic noise Training
Mean temperature in the period for the target (light shading) and the reconstructions (dark shading) based on surrogates with an prescribed warm period. Calibration period the last 100 years Statistics from 100 pseudo-proxy experiments 57 pseudo-proxies, realistic noise Relative trend in the last 100 years. Calibration period the first 100 years.
Real Proxies Real world reconstructions based on 14 decadally smoothed proxies from the NH hemisphere (Hegerl 2007). The proxies cover the´period and the calibration period is Observed temperatures are from HadCRUT2v. The average temperature north of 30 N is reconstructed. Training
35 proxies covering the last millennium compiled by Fredrik C. Ljungqvist, Stockholm University
Real Proxies Real world reconstructions based on 35 proxies. The proxies cover the´period and the calibration period is Observed temperatures are from HadCRUT2v. The average temperature on the NH is reconstructed. 20 proxies pass the F-test at p=0.01. Training
Conclusions Based on pseudo-proxy experiments we find that previous reconstruction methods of the hemispheric-sale temperature underestimate low-frequency variability. We have formulated a new method, LOC. This method first reconstruct local temperatures where proxies are present, followed by spatial averaging. The local reconstruction is based on indirect regression. Pseudo-proxy experiments show that this method preserve low.- frequency variability. The price is an exaggeration of the high-frequency variability. Applied to compilations of real proxies LOC gives reconstructions of the NH temperature with strong variability. Little Ice Age is 1.5 K colder than the present period and the Medieval Warm Period as warm as the present period. Christiansen 2010, submitted J. Clim., terrestrial/staff/boc/loc_rec.pdf