Climate Dynamics

, Volume 32, Issue 2, pp 381–395

Comparison of climate field reconstruction techniques: application to Europe

  • Nadja Riedwyl
  • Marcel Küttel
  • Jürg Luterbacher
  • Heinz Wanner
Article

DOI: 10.1007/s00382-008-0395-5

Cite this article as:
Riedwyl, N., Küttel, M., Luterbacher, J. et al. Clim Dyn (2009) 32: 381. doi:10.1007/s00382-008-0395-5

Abstract

This paper presents a comparison of principal component (PC) regression and regularized expectation maximization (RegEM) to reconstruct European summer and winter surface air temperature over the past millennium. Reconstruction is performed within a surrogate climate using the National Center for Atmospheric Research (NCAR) Climate System Model (CSM) 1.4 and the climate model ECHO-G 4, assuming different white and red noise scenarios to define the distortion of pseudoproxy series. We show how sensitivity tests lead to valuable “a priori” information that provides a basis for improving real world proxy reconstructions. Our results emphasize the need to carefully test and evaluate reconstruction techniques with respect to the temporal resolution and the spatial scale they are applied to. Furthermore, we demonstrate that uncertainties inherent to the predictand and predictor data have to be more rigorously taken into account. The comparison of the two statistical techniques, in the specific experimental setting presented here, indicates that more skilful results are achieved with RegEM as low frequency variability is better preserved. We further detect seasonal differences in reconstruction skill for the continental scale, as e.g. the target temperature average is more adequately reconstructed for summer than for winter. For the specific predictor network given in this paper, both techniques underestimate the target temperature variations to an increasing extent as more noise is added to the signal, albeit RegEM less than with PC regression. We conclude that climate field reconstruction techniques can be improved and need to be further optimized in future applications.

Keywords

PaleoclimateNCAR CSM 1.4ECHO-G 4Pseudoproxy dataRegEMPrincipal component regressionEuropean temperature reconstruction

Supplementary material

382_2008_395_MOESM1_ESM.eps (3.3 mb)
Supplementary Figure 3 for ECHO-G 4. European summer average temperature anomalies (30-year running mean) wrt 1900 to 1990 AD, for PC regression (top) and RegEM (bottom), using 30 pseudoproxies (see Fig. 1) with varying white noise added to the signal. The target (black line) is compared to the reconstruction results (colored lines). (EPS 3.25 mb)
382_2008_395_MOESM2_ESM.eps (3.3 mb)
Supplementary Figure 3 for NCAR CSM 1.4. As Figure 3, but with different TTLS parameter. (EPS 3.25 mb)
382_2008_395_MOESM3_ESM.eps (3.3 mb)
As supplementary Figure 3 for ECHO-G 4, but with different TTLS parameter. (EPS 3.27 mb)
382_2008_395_MOESM4_ESM.eps (3.3 mb)
Supplementary Figure 4 for ECHO-G 4. As supplementary Figure 3, but for winter. (EPS 3.26 mb)
382_2008_395_MOESM5_ESM.eps (3.3 mb)
Supplementary Figure 4 for NCAR CSM 1.4. As Figure 4, but with different TTLS parameter. (EPS 3.26 mb)
382_2008_395_MOESM6_ESM.eps (3.3 mb)
As supplementary Figure 4 for ECHO-G 4, but with different TTLS parameter. (EPS 3.26 mb)
382_2008_395_MOESM7_ESM.eps (3.2 mb)
Supplementary Figure 5 for ECHO-G 4. European summer average temperatures anomalies (30-year running mean) for PC regression (top) and RegEM (bottom). The white noise scenario SNR 1 (red line) is compared with two different red noise scenarios (orange and magenta lines); the target is shown in black. (EPS 3.15 mb)
382_2008_395_MOESM8_ESM.eps (3.2 mb)
Supplementary Figure 6 for ECHO-G 4. As supplementary Figure 5, but for winter. (EPS 3.16 mb)
382_2008_395_MOESM9_ESM.eps (5.7 mb)
Supplementary Figure 7 for ECHO-G 4. Spatial skill patterns of the European summer temperature reconstructions using PC regression (left) and RegEM (right) with white noise scenarios SNR ∞, SNR 1, and SNR 0.5. The skill is defined by the average of the bias (reconstructed values - target values) [shaded] and RE [contours] calculated for each gridpoint over the verification period from 1001 to 1899 AD. The scale refers to the bias, i.e. differences in temperature anomalies for winter and summer separately. Colors indicate reconstructed values that are about (greenish blue and green), higher (light green, yellow to red) or lower (light blue to violet) than the target values. (EPS 5.67 mb)
382_2008_395_MOESM10_ESM.eps (6.5 mb)
Supplementary Figure 8 for ECHO-G 4. As supplementary Figure 7, but for winter. Colors indicate reconstructed values that are about (light blue and greenish blue), higher (light green to green, yellow, red) or lower (dark blue to violet) than the target values. (EPS 6.45 mb)

Copyright information

© Springer-Verlag 2008

Authors and Affiliations

  • Nadja Riedwyl
    • 1
    • 2
  • Marcel Küttel
    • 1
    • 2
  • Jürg Luterbacher
    • 1
    • 2
  • Heinz Wanner
    • 1
    • 2
  1. 1.Oeschger Centre for Climate Change Research (OCCR) and National Centre of Competence in Research on Climate (NCCR)University of BernBernSwitzerland
  2. 2.Institute of Geography, Climatology and MeteorologyUniversity of BernBernSwitzerland