2006 Philadelphia Annual Meeting (22–25 October 2006)

Paper No. 1
Presentation Time: 8:05 AM

RECONSTRUCTING SEA-LEVEL CHANGE FROM SALT-MARSH FORAMINIFERA: STATISTICS AND ‘COMMON SENSE'


GEHRELS, W. Roland, School of Geography, Univ. of Plymouth, Plymouth, PL4 8AA, United Kingdom, wrgehrels@plymouth.ac.uk

Multivariate statistics (‘transfer functions') have become a popular tool in sea-level studies, particularly in those that use salt-marsh foraminifera to reconstruct former sea levels. Transfer functions are based on the uniformitarian principle that the present-day altitudinal relationship of foraminifera to sea level can be used to reconstruct the former sea level from the fossil foraminifera. The main advantages of the transfer-function approach are that (1) sea levels can be reconstructed in an objective and repeatable way and (2) altitudinal errors can be quantified. As with any quick, user-friendly computer-based method, the danger of using transfer functions is that they may be applied in situations in which they are not appropriate.

This paper reviews some of the advantages and disadvantages of the transfer-function approach, with particular attention to ecological and statistical reasons why a transfer function may not work properly. These include: (1) The modern distribution of microfauna is not adequately documented. Inaccurate transfer function results can be produced due to large seasonal variability of foraminiferal occurrences or if assemblages are included in the modern data set that are known to exist well beyond the sampled range. Low species diversity exacerbates this problem. (2) A fossil assemblage is not represented in the modern training set (‘the modern analogue problem'). This will often require a larger (regional) dataset that includes the missing taxa, or it involves disregarding some fossil taxa. In either case, the r2 value is not a realistic performance measure of the transfer function as considerable uncertainties about the ‘true' sea-level relationship of the foraminifera may be introduced. (3) Available software limits the user to either unimodal or linear regression models for the entire foraminiferal dataset. Some salt-marsh foraminifera are linearly distributed along the height gradient (e.g. Jadammina macrescens), others unimodally (e.g. Tiphotrocha comprimata). By using a single model some species distributions will be inadequately described. There are not always clear-cut solutions to these problems, but by selecting a sensible sampling strategy for both the modern and the fossil environments the effectiveness of transfer functions can be maximized.