Introduction and Background

Multivariate statistical analysis of microfossil census data from marine and terrestrial deposits has proved to be a powerful tool for paleoclimatic and paleoceanographic investigations. Despite the development of a number of proxy indicators of sea surface temperature (SST) (oxygen isotopes, Mg/Ca ratios, alkenones, etc.), the most enduring technique has been the transfer function pioneered by Imbrie and Kipp (1971). The method involves factor analysis and multiple regression to develop equations relating microfossil abundance data in modern (core-top) samples to physical parameters such as SST, salinity, dissolved oxygen content, etc. The equations can then be applied to downcore faunal census data to estimate past oceanographic conditions. The technique has been widely applied in paleoclimate studies using a variety of fossil groups from Pliocene to Holocene sediments (eg., Kipp, 1976; Sancetta, 1979; Thunell, 1979a,b; CLIMAP, 1981, 1984; Ruddiman and Esmay, 1986; Hays et al., 1989; Dowsett and Poore, 1990; Cronin and Dowsett, 1990, Dowsett et al., 1996; 1999).

The successful application of the transfer function technique, or any other method of paleontological reconstruction, depends upon two primary factors: (1) the assumption that ecological tolerances of indicator taxa do not change over time and (2) the existence of a taxonomically stable and well-dated calibration data set. The first factor must be assumed in any reconstruction and extends to isotopic and chemical proxy methods, as well as paleontologically based techniques (Dowsett and Robinson, 1998). The second factor, the modern calibration data set, is more problematic. When reconstructing mid Pliocene SST, Dowsett and Poore (1990) were able to use a calibration data set whose samples represented conditions during the last 30 ky. The middle Pliocene temperature signal was large with respect to late Pleistocene variability. In fact, the paleotemperature equations worked remarkably well on North Atlantic last glacial maximum (LGM) and last interglacial maximum (LIM) data sets (Dowsett, 1991).

When reconstructing Holocene SST exhibiting millenial and sub-millenial scale variability, the calibration data set must be isochronous. Dowsett et al. (2002) showed that in general, the last 1500 years of faunal variability in the Gulf of Mexico region is relatively constant. Our goal is to develop and document a factor analytic planktic foraminifer transfer function capable of reconstructing Holocene temperature changes in the Gulf of Mexico. We have selected a suite of accelerator mass spectrometer (AMS) ¹⁴C dated core-top samples (Dowsett et al., 2003) and factor analyzed the associated faunal assemblages. A set of equations were developed (transfer function GOM2) that relate modern physical oceanography to the faunal data. In this paper we outline the development of GOM2 and use it to delineate a record of surface temperature from piston cores in the northern and western Gulf of Mexico.