Correlation confidence limits for unevenly sampled data

Jason Roberts, Mark Curran, Samuel Poynter, Andrew Moy, Tas van Ommen, Tessa Vance, Carly Tozer, Felicity S. Graham, Duncan A. Young, Christopher Plummer, Joel Pedro, Donald Blankenship, Martin Siegert

Research output: Contribution to journalArticleResearchpeer-review

6 Citations (Scopus)

Abstract

Estimation of correlation with appropriate uncertainty limits for scientific data that are potentially serially correlated is a common problem made seriously challenging especially when data are sampled unevenly in space and/or time. Here we present a new, robust method for estimating correlation with uncertainty limits between autocorrelated series that does not require either resampling or interpolation. The technique employs the Gaussian kernel method with a bootstrapping resampling approach to derive the probability density function and resulting uncertainties. The method is validated using an example from radar geophysics. Autocorrelation and error bounds are estimated for an airborne radio-echo profile of ice sheet thickness. The computed limits are robust when withholding 10%, 20%, and 50% of data. As a further example, the method is applied to two time-series of methanesulphonic acid in Antarctic ice cores from different sites. We show how the method allows evaluation of the significance of correlation where the signal-to-noise ratio is low and reveals that the two ice cores exhibit a significant common signal.

Original languageEnglish
Pages (from-to)120-124
Number of pages5
JournalComputers and Geosciences
Volume104
DOIs
Publication statusPublished - 1 Jul 2017
Externally publishedYes

Keywords

  • Autocorrelation
  • Bootstrapping
  • Confidence limits
  • Gaussian kernel method
  • Unevenly sampled data

Cite this