Transfer functions for quantitative palaeoclimatic research in Norway

Participants: John Birks, Nalan Koç, Svein Olav Dahl, Sylvia M. Peglar, Endre Willasen

Transfer functions are empirically derived statistical models that link present-day biota or sedimentary variables to modern environment parameters (e.g. mean July air temperature). Assuming that these modelled relationships have not changed with time, transfer functions can then be used to transform fossil assemblages into quantitative estimates of the past environment. In the last 10 years considerable advances have been made in the mathematical theory of transfer functions by C.J.F ter Braak and associates in terms of transfer-function estimation, model validation, reconstruction prediction errors, diagnostics, validation, and evaluation, and optimal model selection There is now a rigorous unified theory and methodology involving the 'inverse' regression-type procedures of partial least squares (PLS) or weighted-averaging (WA)PLS depending on the type of data and their properties. Cross-validation and error estimation by Monte Carlo simulation are now standard for model selection and validation and reconstruction evaluation.

These methods have been shown to consistently out-perform earlier techniques such as modern analogues, Imbrie & Kipp factor-analysis regression, and response surfaces with a wide range of real data and with simulated data of known properties in rigorously designed comparative studies.

Proposed research -

(1) Extend the existing transfer-function data-sets in Norway for

(a) pollen and mean July and January temperatures and annual precipitation

(b) chironomids and mean July air and summer water temperature

(c) diatoms and lake pH and dissolved organic carbon, both related to regional climate

(d) marine diatoms, foraminifera, and dinoflagellates and surface-water ocean and fjord temperatures

(e) stomatal density and atmospheric CO₂ concentrations

(f) ¹³C and temperature and atmospheric CO₂ concentrations

(2) Develop public-domain user-friendly computer software and documentation to implement 'state-of-the-art' statistical methodology for transfer functions and palaeoclimatic reconstructions.

(3) Provide 'state-of-the-art' statistical advice about palaeoclimatic reconstructions from transfer functions within the NORPAST project.