Regression-based age estimation of a stratigraphic isotope sequence in Switzerland

• Main Author: Ghosh, Sucharita
• Language: English
• Published: 2018
• Subjects: Greenland; Greenland ice core; Greenland Ice core Project; Greenland ice cores; GRIP; ice core
• Online Access: http://doc.rero.ch/record/312724/files/334_2006_Article_59.pdf

Multi-proxy data such as pollen percentages, aquatic biota, and stable isotope ratios in lake sediments in conjunction with climate transfer functions can be used to reconstruct past climate. This has been the subject of some recent projects. Often, stable-isotope ratios of oxygen are used as an independent proxy for climate. Past climate so reconstructed can in turn be used to assess the response of past vegetation to climatic oscillations, for instance near the epoch boundaries. One important intermediate step is to establish the age of the stratigraphic sequence. Strong similarities between the δ 18O records from European lake sediments and the Greenland ice cores are of interest. The Greenland ice-core project (GRIP) provided δ 18O data that were dated using an ice-flow model. Although the physical laws behind the isotope series from ice and lake sediment are different, statistical methods can be used to match the two series. In this paper, a regression-based approach is suggested for series matching. The method is illustrated by analyzing a series of δ 18O records covering the Late-glacial interstadial (ca. 15,000-13,000 years b.p. [1950]) from Gerzensee, Switzerland. Regression methods for age-depth modelling have also been recommended by other authors. Such an approach leads to reproducible and statistically founded age estimates and can easily be updated to include new data and information as needed. In this paper, the modelling step is preceded by identifying comparable sub-sections in the two isotope series by empirically matching the local minima and maxima in the smoothed isotope values; regression models are then used locally for each sub-section. This accommodates for local differences in the parameters. Variations in the final age estimates caused by different choices of the smoothing (bandwidth) parameters used in the intermediate nonparametric smoothing step are also taken into account in this algorithm