Reconstruction of sea-surface temperatures from assemblages of planktonic foraminifera: multi-technique approach based on geographically constrained calibration data sets and its application to glacial Atlantic and Pacific Oceans

Abstract

We present a conceptual framework for a new approach to environmental calibration of planktonic foraminifer census counts. This approach is based on simultaneous application of a variety of transfer function techniques, which are trained on geographically constrained calibration data sets. It serves to minimise bias associated with the presence of cryptic species of planktonic foraminifera and provides an objective tool for assessing reliability of environmental estimates in fossil samples, allowing identification of adverse effects of no-analog faunas and technique-specific bias. We have compiled new calibration data sets for the North (N=862) and South (N=321) Atlantic and the Pacific Ocean (N=1111). We show evidence that these data sets offer adequate coverage of the Sea-Surface Temperature (SST) and faunal variation range and that they are not affected by the presence of pre-Holocene samples and/or calcite dissolution. We have applied four transfer function techniques, including Artificial Neural Networks, Revised Analog Method and SIMMAX (with and without distance weighting) on faunal counts in a Last Glacial Maximum (LGM) data set for the Atlantic Ocean (748 samples in 167 cores; based on the GLAMAP-2000 compilation) and a new data set for the Pacific Ocean (265 samples in 82 cores) and show that three of these techniques provide adequate degree of independence for the advantage of a multi-technique approach to be realised. The application of our new approach to the glacial Pacific lends support to the contraction and perhaps even a cooling of the Western Pacific Warm Pool and a substantial (>3 °C) cooling of the eastern equatorial Pacific and the eastern boundary currents. Our results do not provide conclusive evidence for LGM warming anywhere in the Pacific. The Atlantic reconstruction shows a number of robust patterns, including substantial cooling of eastern boundary currents with considerable advection of subpolar waters into the Benguela Current, a cooling of the equatorial Atlantic by ∼5 °C, and steep SST gradients in the mid-latitude North Atlantic. The transfer function techniques generally agree that subtropical gyre areas in both hemispheres did not change significantly since the LGM, although the ANN technique produced glacial SST in the southern gyre 1–2  °C warmer than today. We have revisited the issue of sea-ice occurrence in the Nordic Seas and using the distribution of subpolar species of planktonic foraminifera in glacial samples, we conclude that the Norwegian Sea must have been ice-free during the summer.

Introduction

All biological processes, from the growth of individual organisms to the dynamics of large ecological systems, are affected by the physical and chemical state of their environment. If we knew how the environment modifies the basic genetic design of organisms and how it controls their spatial and temporal distribution, we could use the fossil record of such organisms to reconstruct the state and variation of past environments. Whilst most fossils record and preserve environmentally valuable and extractible information, planktonic foraminifera are by far the most important signal carriers in palaeoceanography. The physical and chemical properties of foraminiferal shells provide a multitude of palaeoproxies, including passively recorded chemical and isotopic signals, signatures of environmentally modulated metabolic processes, and ecologically controlled aspects of taxonomic abundances and shell morphology (Henderson, 2002). Physical and chemical properties of foraminiferal shells follow different taphonomic pathways and proxies based on these properties rely on different sets of assumptions. This makes it possible to derive independent estimates of palaeoceanographic parameters from the same fossil assemblage, thus providing a unique opportunity to assess the robustness of such palaeoenvironmental reconstructions.

In an ideal world, one would wish to have a full mechanistic understanding of how a proxy works. Unfortunately, in most cases we lack the detailed knowledge of the chemical, physical and biological processes that operate to record and preserve an environmental signal in a proxy record. In the absence of a direct deterministic relationship we are forced to resort to the process of empirical calibration. Here, the relationship between a relatively easily measured parameter of a fossil and an environmental variable is derived by observing and describing such relationship in the present-day situation. This process is methodologically relatively simple but it involves a number of assumptions that have significant impact on the applicability and reliability of each empirically calibrated proxy (Hutson, 1977; Birks, 1995; ter Braak, 1995).

Due to the nature of the processes involved, chemical proxies are more likely to approach the desired mechanistic level of understanding, whilst proxies based on organism size, shape and abundance remain bound to the realms of empirical calibration. Despite this limitation, the possibility to complement and cross-validate the reliability of chemical proxies, especially those derived from samples with unfavourable taphonomic histories has provided the impetus to improve existing physical palaeoproxies (ter Braak and Juggins, 1993; Pflaumann et al., 1996; Waelbroeck et al., 1998; Malmgren et al., 2001) as well as development of new ones (e.g. Bollmann et al., 2002; Schmidt et al., 2004).

The most commonly used physical palaeoproxy involves mathematical analysis of census counts of microfossil assemblages: the so-called “transfer function” approach. Although this kind of species–environment calibration has a long history in the natural sciences its application to palaeoenvironmental reconstruction was brought to prominence by the CLIMAP group in their pivotal effort to reconstruct the sea-surface temperature field of the last glacial maximum ocean (CLIMAP Project Members, 1976, CLIMAP Project Members, 1981). The results of the CLIMAP reconstructions have been used extensively for forcing and validating ocean circulation models (e.g. Webb et al., 1997). These are the same models that are being used to evaluate the effects of man-made changes to our planet's surface and atmosphere. Many political decisions that are being taken today are based on such evaluations. It is therefore very important to understand properly how the validation data for these models are produced and, equally important, what is their reliability and precision.

Ever since the pioneering study by CLIMAP Project Members (1976), the Last Glacial Maximum (LGM), with its different, yet well defined, insolation and greenhouse gas forcing, has been the prime target for comparison of model outputs and proxy data. The international EPILOG initiative has been recently launched (Mix et al., 2001) with the view to revise the reconstruction of the glacial Earth surface. The MARGO effort is focused on providing a new LGM Sea-Surface Temperature (SST) reconstruction, using the enormous body of new, high-quality data produced in the last two decades. Transfer functions based on planktonic foraminifer counts have always played a central role in glacial palaeothermometry and the new calibration techniques developed in the last decade have contributed greatly to renewed interest in foraminifer SST reconstructions. In this study, we review the use of planktonic foraminifer counts for palaeotemperature reconstructions and present the rationale for a new approach based on the application of multiple techniques calibrated on geographically constrained calibration data sets. We then use this approach to develop a new LGM SST reconstruction for the Atlantic and Pacific Oceans and compare our results with previous studies and other palaeothermometers.