Reassessing transfer-function performance in sea-level reconstruction based on benthic salt-marsh foraminifera from the Atlantic coast of NE North America

Abstract: The need to increase the number and distribution of sea-level records spanning the last few hundred years has led to particular interest in developing high-precision, geologically based sea-level reconstructions that capture decimetre and multi-decadal scale changes. Transfer functions for tide level are statistical tools that quantify the vertical relationship between inter-tidal microfossils and elevation within the tidal frame and their use in sea-level reconstruction is growing in popularity. Whilst a range of sampling strategies, dataset qualities and underlying statistical models have been used in transfer-function development, all variants share the common requirement of accurately extracting precise species–elevation relationships from surface data, and reliably applying these to fossil assemblages to infer past conditions. We present surface foraminiferal data from six transects sampled at three sites spanning a large latitudinal range extending from Newfoundland (Canada) to North Carolina (USA). These data demonstrate that significant spatial differences exist within the high-marsh foraminiferal assemblages commonly used to reconstruct past relative sea-level (RSL). We standardise these data to account for inter-site differences in tidal range using several variants of the standardised water level index (SWLI) of Horton et al. (1999b) and show that the best performance is achieved by using the highest occurrence of foraminifera as the upper tidal datum level. The standardised surface foraminiferal data are used to develop a suite of foraminiferal transfer functions for tide level which are then applied to fossil assemblages from two sediment cores to reconstruct palaeomarsh-surface elevation. We highlight the manner in which species–elevation relationships are extracted and modified during transfer-function development, and the impacts that choices in dataset composition and transfer-function type have on the resulting reconstructions. Our results graphically illustrate the importance of these choices and the impacts of temporal and spatial variability in foraminiferal distributions, none of which are adequately represented or discernible from the standard summary statistics of performance that commonly accompany transfer-function reconstructions. We conclude that a more explicit treatment of the transfer-function development process is required to support the growing body of precise RSL reconstructions that are now appearing in the literature. To that end, we make the following four recommendations for a framework to assist in transfer-function development that will allow better record inter-comparison and minimise the potential for producing precise but ultimately inaccurate reconstructions: 1) sampling for surface data should focus on capturing complete species response curves rather than simply compiling modern analogues for fossil material; 2) The similarity between the surface assemblages used to produce transfer functions and the fossil assemblages to which they are applied, should be better quantified using a range of statistical approaches, and must accompany any reconstruction; 3) where modern analogues are lacking in local surface assemblages, data from additional sites should be added, with selection based on fulfilling the requirements of the previous two recommendations. The manner in which individual species optima are combined to produce a composite optimum, and the extent to which this provides a reasonable representation of the surface data should also be documented; 4) the application of WA PLS component 2 (and higher) cannot be justified solely on the grounds of increased performance measures (e.g. r2 jack and RMSEP), and if applied, the pattern of optima updates should be used to highlight species that may distort reconstructions. [Copyright &y& Elsevier]