Abstract

© 2017 Elsevier Ltd The discovery of IP 25 as a qualitative biomarker proxy for Arctic sea ice and subsequent introduction of the so-called PIP 25 index for semi-quantitative descriptions of sea ice conditions has significantly advanced our understanding of long-term paleo Arctic sea ice conditions over the past decade. We investigated the potential for classification tree (CT) models to provide a further approach to paleo Arctic sea ice reconstruction through analysis of a suite of highly branched isoprenoid (HBI) biomarkers in ca. 200 surface sediments from the Barents Sea. Four CT models constructed using different HBI assemblages revealed IP 25 and an HBI triene as the most appropriate classifiers of sea ice conditions, achieving a > 90% cross-validated classification rate. Additionally, lower model performance for locations in the Marginal Ice Zone (MIZ) highlighted difficulties in characterisation of this climatically-sensitive region. CT model classification and semi-quantitative PIP 25 -derived estimates of spring sea ice concentration (SpSIC) for four downcore records from the region were consistent, although agreement between proxy and satellite/observational records was weaker for a core from the west Svalbard margin, likely due to the highly variable sea ice conditions. The automatic selection of appropriate biomarkers for description of sea ice conditions, quantitative model assessment, and insensitivity to the c-factor used in the calculation of the PIP 25 index are key attributes of the CT approach, and we provide an initial comparative assessment between these potentially complementary methods. The CT model should be capable of generating longer-term temporal shifts in sea ice conditions for the climatically sensitive Barents Sea.

DOI

10.1016/j.gca.2017.11.001

Publication Date

2017-11-01

Publication Title

Geochimica et Cosmochimica Acta

Volume

222

First Page

406

Last Page

420

ISSN

0016-7037

Embargo Period

2018-11-10

Organisational Unit

School of Geography, Earth and Environmental Sciences

Share

COinS