Spectra of paleoclimatic time series commonly show a continuous decrease of spectral amplitude with increasing frequency ("red-noise"). A first-order autoregressive process is often used as null hypothesis to assess whether or not the variability recorded in a time series is consistent with a stochastic origin of this type. Since most paleoclimatic time series are unevenly spaced, and the straightforward estimation of such red-noise background requires evenly spaced time series, some sort of interpolation is necessary. Unfortunately, interpolation results in a significant bias because it alters the estimated spectrum by enhancing the low-frequency components at the expense of high-frequency components. Hence, the estimated spectrum of an interpolated time series becomes too "red" compared to the true spectrum. To overcome this problem a Fortran 90 program for estimating red-noise spectra directly from unevenly spaced data has been developed.

The program is in the public domain and an accompanying paper with a program description appears in Computers & Geosciences, v.28, No. 3, April 2002, (an official journal of the International Association for Mathematical Geology). A preprint of the Computers & Geosciences paper can be downloaded from WDC Paleo.

Authors Michael Schulz of the Institut für Geowissenschaften, Universität Kiel and Manfred Mudelsee of the Institut für Meteorologie, Universität Leipzig would appreciate any comments from REDFIT users. Time series analysis programs RAMPFIT and TAUEST are also available from Dr. Mudelsee' s website.

Click here to download REDFIT now. The zip-archive includes Fortran 90 source code, binaries for Windows 95 (or above), and program documentation and example files. Please make sure to maintain the directory structure of the zip archive when unpacking this file.