Trends and Correlation Estimation in Climate Sciences: Effects of Timescale Errors

Manfred.Mudelsee [ at ]


Trend describes time-dependence in the first moment of a stochastic process, and correlation measures the linear relation between two random variables. Accurately estimating the trend and correlation, including uncertainties, from climate time series data in the uni- and bivariate domain, respectively, allows first-order insights into the geophysical process that generated the data. Timescale errors, ubiquitious in paleoclimatology, where archives are sampled for proxy measurements and dated, poses a problem to the estimation. Statistical science and the various applied research fields, including geophysics, have almost completely ignored this problem due to its theoretical almost-intractability. However, computational adaptations or replacements of traditional error formulas have become technically feasible. This contribution gives a short overview of such an adaptation package, bootstrap resampling combined with parametric timescale simulation. We study linear regression, parametric change-point models and nonparametric smoothing for trend estimation. We introduce pairwise-moving block bootstrap resampling for correlation estimation. Both methods share robustness against autocorrelation and non-Gaussian distributional shape. We shortly touch computing-intensive calibration of bootstrap confidence intervals and consider options to parallelize the related computer code. Following examples serve not only to illustrate the methods but tell own climate stories: (1) the search for climate drivers of the Agulhas Current on recent timescales, (2) the comparison of three stalagmite-based proxy series of regional, western German climate over the later part of the Holocene, and (3) trends and transitions in benthic oxygen isotope time series from the Cenozoic. Financial support by Deutsche Forschungsgemeinschaft (FOR 668, FOR 1070, MU 1595/4-1) and the European Commission (MC ITN 238512, MC ITN 289447) is acknowledged.

Item Type
Conference (Invited talk)
Primary Division
Primary Topic
Publication Status
Event Details
AGU Fall Meeting, 03 Dec 2012 - 07 Dec 2012, San Francisco, USA.
Eprint ID
Cite as
Mudelsee, M. , Bermejo, M. , Bickert, T. , Chirila, D. , Fohlmeister, J. , Köhler, P. , Lohmann, G. , Olafsdottir, K. and Scholz, D. (2012): Trends and Correlation Estimation in Climate Sciences: Effects of Timescale Errors , AGU Fall Meeting, San Francisco, USA, 3 December 2012 - 7 December 2012 .

Add to AnyAdd to TwitterAdd to FacebookAdd to LinkedinAdd to PinterestAdd to Email

Research Platforms


Edit Item Edit Item