Sea-surface temperature estimates in the Southeast Paci c based on planktonic foraminiferal species; modern calibration and Last Glacial Maximum

Transcription

1 Marine Micropaleontology 44 (2002) 1^29 Sea-surface temperature estimates in the Southeast Paci c based on planktonic foraminiferal species; modern calibration and Last Glacial Maximum Melissa J. Feldberg, Alan C. Mix College of Oceanic and Atmospheric Sciences, Oregon State University, Corvallis, OR 97331, USA Received 19 January 2001; received in revised form 25 June 2001; accepted 9 July 2001 Abstract Estimates of sea-surface temperatures based on foraminiferal faunal species suggest that the Eastern Equatorial Pacific Ocean was 3^5 C cooler during the Last Glacial Maximum (LGM) than at present. Analysis of new cores from the Southeast Pacific reveals a likely source of ice-age cooling in variations of the Peru Current. Off southern Peru, LGM ocean temperatures were 6^8 C cooler than at present, consistent with substantial cooling on land inferred from regional glacier advances and ice-core data. In the Southeast Pacific, ice-age foraminiferal assemblages have good modern analogs, and transfer functions that define assemblages based on ancient samples yield results similar to those based on coretop samples. During the LGM, subpolar species dominate the Eastern Boundary Current off Peru and extend to the equator. In contrast, the range of the equatorial upwelling species remains roughly constant. We infer from these data and a heat budget model that equatorward advection of cool water, more than equatorial upwelling, drove LGM cooling of the Eastern Tropical Pacific Ocean. ß 2002 Elsevier Science B.V. All rights reserved. Keywords: Paleoceanography; Paci c Ocean; Peru Current; Planktonic foraminifera; Last Glacial Maximum 1. Introduction Recent estimates of sea-surface temperatures (SST) based on planktonic foraminiferal species assemblages (Mix and Morey, 1996; Mix et al., 1999), oxygen isotope values (Patrick and Thunell, 1997), and trace metal indices (Lea et al., * Corresponding author. Tel.: ; fax: address: mix@coas.oregonstate.edu (A.C. Mix). 2000) indicate that annual mean SSTs in the Eastern Equatorial Paci c were 3^5 C cooler during the Last Glacial Maximum (LGM) than at present, signi cantly cooler than reported by CLI- MAP (1981). These new results are in closer agreement with terrestrial temperature estimates for the tropics derived from ice cores (Thompson et al., 1995), groundwater records (Stute, 1995) and the extent of former mountain glaciers (Clapperton, 1993), as well as with tropical coral temperature estimates (Guilderson et al., 1994) than previous estimates, which showed little temperature change in the tropical oceans. Cool equato / 02 / $ ^ see front matter ß 2002 Elsevier Science B.V. All rights reserved. PII: S (01)

2 2 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 rial SSTs at the LGM allow climate models to better reconcile oceanic and terrestrial temperature estimates of the LGM (Rind and Peteet, 1985; Webb et al., 1997; Hostetler and Mix, 1999; Hostetler and Clark, 2000). Mix et al. (1999) suggest that the discrepancy in tropical temperature estimates between their work and that of CLIMAP (1981) is the result of two revisions to the method used to calibrate foraminiferal species to SST. The rst revision is to include ancient samples in the de nition of faunal factors used to create the calibration equation. We refer to this as the downcore calibration method. This sample selection helps to circumvent the no-analog problem, which results when species variability is greater over time than it is spatially in the modern ocean (Hutson, 1977). A second change to the CLIMAP (1981) methodology is to calibrate the equation on a regional rather than a global scale. This optimizes the sensitivity of the equation for those species present in the study area. In this study we compare the downcore and coretop calibration methods in equations for estimating regional annual average SSTs from foraminiferal species assemblages in the Southeastern Paci c. New coretop samples located near the South American margin help to expand the sparse data set in this area and to improve the calibrations. Three long sediment cores located in a transect across the Peru Current provide new LGM samples to be used for determining the extent of glacial cooling in the eastern boundary current. Evidence that the Eastern Equatorial Paci c was considerably cooler during the LGM than at present raises questions as to the possible mechanisms driving such cooling. Was there an increase in equatorward ow of cold water in the Peru Current during the LGM? Did an increase in coastal or equatorial upwelling reduce glacial SSTs? Was a combination of these mechanisms responsible for the cooler SSTs? Given the potential importance of the eastern boundary currents in contributing to SST changes and the limited previous work done in this area, new data reported here help to constrain mechanisms of climate change in the Eastern Paci c. 2. Modern oceanographic setting The Peru (Humboldt) Current is the largest eastern boundary current in the world. Stretching almost the full length of South America, it is a major conduit for the exchange of heat and nutrients between high and low latitudes in the South Paci c Ocean (Strub et al., 1998). The Peru Current makes up the eastern portion of the southern anti-cyclonic subtropical gyre (Fig. 1a). In the South Paci c, water is transported eastward in the West Wind Drift (WWD) between 40 S and 50 S latitude. Upon reaching the South American continent, the WWD splits into the northward owing Peru Current and the southward owing Cape Horn Current (Strub et al., 1998). The Peru Current ows equatorward with a velocity of V2 cm/s between 0 and 100 m depth (Scha er et al., 1995). Near 5 S the cool current increases in velocity and is de ected away from the coast to become part of the South Equatorial Current (SEC), which ows westward between 4 N and 15 S latitude (Wyrtki, 1965) (Fig. 1a). Inshore of the Peru Current is the Peru^Chile Countercurrent (PCCC), a weak and irregular surface current owing to the south and located approximately 200 km o shore (Huyer et al., 1991). Because of this current, relatively warm SSTs extend southward along the coast Peru (Fig. 1b). The Peru Coastal Current (PCC) ows equatorward, inshore of the PCCC. This relatively minor current brings cold, nutrient-rich water upwelled along the coast of Peru toward the equator (Wyrtki, 1965; Strub et al., 1998). Beneath the SEC lies the Equatorial Undercurrent (EUC), which ows eastward across the Paci c at depths of 200^250 m. The upper portion of this current is the source of 19^24 C water upwelled along the equator near the Galapagos (Wyrtki, 1981). The lower portion of the current continues eastward, upwelling cold water (11^ 14 C) o the coast of Peru (Toggweiler and Dixon, 1991). Both the SEC and EUC contribute to the equatorial cold tongue, which extends westward to 130 W longitude (Wyrtki, 1981). Equatorial upwelling is strong in the austral winter when there is intensi cation of the southeast trade

3 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 3 Fig. 1. (a) Surface and subsurface currents of the Peru Eastern Boundary Current System. Major surface currents (black dashed lines) are the West Wind Drift (WWD), the Cape Horn Current (CHC), the Peru Current (PC), the Peru Coastal Current (PCC) and the South Equatorial Current (SEC). The subsurface currents (dotted lines) are the Peru^Chile Countercurrent (PCCC, 50^ 150 m depth), the Peru Undercurrent (PUC, 100^400 m depth), and the Equatorial Undercurrent (EUC, 200^250 m depth), after Strub et al. (1998). (b) Location of multicores (squares) and trigger or piston cores (triangles) used in this study. Contours are annual average SSTs (Ocean Climate Laboratory, 1999).

4 4 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 winds and divergence is increased. In the austral summer, the trade winds diminish, upwelling is reduced and the cold tongue is less prominent (Strub et al., 1998). Upwelling along the South American coast is driven by persistent southeasterly winds resulting in o shore Ekman ow and consequent upward ow of subsurface waters. A majority of the upwelled water comes from the Poleward Undercurrent (PUC), also known as the Gunther Undercurrent. The PUC is located mainly between depths of 50 m and 300 m, and is especially strong near depths of 150 m o Peru (Huyer et al., 1991). Further o shore beneath the Peru Current, the PUC is a deeper and slower current, located between 100 m and 400 m depth (Scha er et al., 1995). The PUC, which originates in the EUC near the Galapagos Islands, is characterized by cold temperatures, high salinities, and low dissolved oxygen (Wyrtki, 1965; Strub et al., 1998). Upwelling results in cold SSTs along the coast and high productivity in these regions due to the high nutrient content of this water (Strub et al., 1998). 3. Materials Thirty-six multicores were collected along a north^south transect from beneath the Peru^Chile Current along the western margin of South America during the Genesis III cruise aboard the R/V Roger Revelle (Fig. 1b). We sampled the uppermost 2 cm of one multicore for each site. The samples were freeze-dried, weighed and washed through a 63-Wm sieve using a 0.5% sodium hexa-meta-phosphate solution. The remaining coarse fraction was oven-dried, weighed, and dry sieved at 150 Wm. The s 150-Wm fraction was then divided with a microsplitter until the sample contained approximately 300 foraminifera. Of the 36 multicore samples collected, only the 16 cores that contained greater than 275 individual foraminifera were included in this study (Table 1). Three long core sites comprise a transect across the Peru^Chile Current along the Nazca Rise o southern Peru (Fig. 1b). Core RR9702A-63TC (2901 m depth, S, W) was sampled at depth intervals of 4 cm from 0 to 162 cm. Core Y (2734 m depth, S, W) was sampled at 10-cm intervals from 20 to 270 cm. A trigger core and piston core at one site, RR9702A-69TC, and -69PC (both at 3228 m depth, S, W), were sampled at depth intervals of 10 cm from 20 to 270 cm, and 10 cm from 260 to 460 cm, respectively. The intervals from 0 to 10 cm and from 260 and 380 cm in RR9702A-69PC were devoid of planktonic Table 1 Location of new multicores used in calibration data set Multicore site Latitude Longitude Water depth (m) RR9702A-66MC RR9702A-64MC RR9702A-62MC RR9702A-60MC RR9702A-54MC RR9702A-52MC RR9702A-50MC RR9702A-48MC RR9702A-46MC RR9702A-42MC RR9702A-24MC RR9702A-14MC RR9702A-10MC RR9702A-08MC RR9702A-06MC RR9702A-01MC

5 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 5 foraminifera and therefore no samples from these intervals were included in this study. We processed the downcore samples in the same manner as the multicore samples. We added the 16 new surface-sediment samples from this study to 216 coretop samples from the database compiled by Prell (1985), plus data from Coulbourn (1980), Sverdlove (1983), and Mix et al. (1999) located in the Eastern Paci c between 50 N and 50 S latitude and 140 W and 70 W longitude to make up the database of 232 modern samples (Fig. 2). The 112 new downcore samples from RR9702A-63TC, Y , and RR9702A- 69TC and PC were added to 423 ancient samples from ve long marine cores (Table 2) to make up the database of 535 ancient samples. The depths of the LGM samples in RR9702A- 63TC and RR9702A-69PC were de ned as the depth of the rst maximum in the oxygen isotope record of each core. The isotopic stratigraphy for these cores is presented elsewhere (Feldberg and Mix, 2002). These two LGM samples were added to the existing 26 LGM samples in the Eastern Paci c (Mix et al., 1999) to complete the LGM data set. 4. Methods To assess the extent of the no-analog problem (Hutson, 1977) and to evaluate the best method for estimating mean annual SST based on foraminiferal fauna in the Eastern Paci c, we developed three transfer-function equations using (A) only the 232 modern coretop samples, (B) only the 535 ancient downcore samples, and (C) all 767 coretop and downcore samples. For each set of samples, Q-mode factor analysis was performed (Klovan and Imbrie, 1971) followed by a multiple stepwise regression to derive the relationship between the resulting factors and modern oceanographic conditions following the transfer function technique of Imbrie and Kipp (1971) Faunal data Approximately 300 planktonic foraminifera in each sample were sorted and identi ed using the taxonomy of Parker (1962). This taxonomy was straightforward with the exception of identifying species in the intergrade between Neogloboquadrina dutertrei and Neogloboquadrina pachyderma. We did not recognize the P-D intergrade category of Kipp (1976) in the samples counted for this study. Neogloboquadrina dutertrei was distinguished from N. pachyderma primarily by the presence of an umbilical tooth, more than four chambers, and a more pitted texture based on the description of Parker (1962). For samples from previous work in the region in which the P-D intergrade was identi ed, this species was grouped with N. dutertrei based on examinations of a subset of samples (Mix et al., 1999). Prior to factor analysis, we converted all species counts to percentages with closure around the 26 species given in Table 3. These are the taxonomic categories of CLIMAP (1981) with a few exceptions. The pink and white varieties of Globigerinoides ruber were grouped together as a single species. Specimens of Globigerinoides sacculifer with and without an elongated nal chamber Table 2 Location of sediment cores used here to make up the downcore data set Core site Latitude Longitude Water depth (m) Reference Y Mix and Morey, 1996 RC Mix and Morey, 1996 RC Mix and Morey, 1996 Y Mix and Morey, 1996 RR9702A This study Y This study RR9702A This study

6 6 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 Fig. 2. Locations of the 232 coretops (circles), seven long sediment cores (triangles), and 28 LGM samples (squares) included in the coretop, downcore, and downcore+coretop equations and used to estimate modern and LGM SST. were also grouped together. Globorotalia menardii and Globorotalia tumida were excluded from the data set due to their resistance to dissolution (Parker and Berger, 1971), and for consistency with the methodology of Mix et al. (1999) which included LGM samples from the Atlantic Ocean where these species are not present. The CLIMAP (1981) study did not retain Globoquadrina con-

7 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 7 Table 3 Q-mode factor score matrix for the coretop data set Species Equatorial Subpolar Subtropical EBC Orbulina universa Globigerinoides conglobatus Globigerinoides ruber (total) Globigerinoides tenellus Globigerinoides sacculifer (total) Sphaeroidinella dehiscens Globigerinella aequilateralis Globigerinella calida Globigerina bulloides Globigerina falconensis Globigerina digitata Globigerina rubescens Globigerina quinquiloba Neogloboquadrina pachyderma (s) Neogloboquadrina pachyderma (d) Neogloboquadrina dutertrei Globoquadrina conglomerata Globoquadrina hexagona Pulleniatina obliquiloculata Globorotalia in ata Globorotalia truncatulinoides (s) Globorotalia truncatulinoides (d) Globorotalia crassaformis Globorotalia hirsuta Globorotalia scitula+theyeri Globigerinita glutinata Total Coretop data explained 87% 33% 13% 24% 17% Downcore data explained 83% 68% 3% 6% 8% Downcore+coretop data explained 83% 56% 7% 8% 13% Bold numbers indicate those species which are most in uential in each factor (factor scores 90.3 and v0.3). For Neogloboquadrina pachyderma and Globorotalia truncatulinoides, s indicates sinistral coiling and d indicates dextral coiling. glomerata or Globoquadrina hexagona, both of which were rare but included in our study Factor analysis Q-mode factor analysis (Klovan and Imbrie, 1971) is used to model the maximum amount of data using the fewest statistically independent end members de ned empirically in the available samples. This eliminates redundant information in the data set and makes the regression equations more robust. Q-mode factor analysis results in a factor score matrix that describes the composition of each factor in terms of the species included in the analysis, and a loading matrix that describes the composition of the samples in terms of the factors. A log transformation of the species percentages was used to prevent the most abundant species from dominating the factors (Mix and Morey, 1996; Mix et al., 1999). Each set of factors de ned for the three data sets (coretop, downcore, and downcore+coretop) was applied to the modern samples by expressing the species assemblage of each coretop sample in terms of the three di erent sets of factor scores. To examine their oceanographic signi cance, we mapped the coretop factor loadings from all three experiments Multiple regression Equations relating modern oceanographic con-

8 8 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 Table 4 Q-mode factor score matrix for the downcore data set Species Equatorial Subpolar Subtropical EBC Orbulina universa Globigerinoides conglobatus Globigerinoides ruber (total) Globigerinoides tenellus Globigerinoides sacculifer (total) Sphaeroidinella dehiscens Globigerinella aequilateralis Globigerinella calida Globigerina bulloides Globigerina falconensis Globigerina digitata Globigerina rubescens Globigerina quinquiloba Neogloboquadrina pachyderma (s) Neogloboquadrina pachyderma (d) Neogloboquadrina dutertrei Globoquadrina conglomerata Globoquadrina hexagona Pulleniatina obliquiloculata Globorotalia in ata Globorotalia truncatulinoides (s) Globorotalia truncatulinoides (d) Globorotalia crassaformis Globorotalia hirsuta Globorotalia scitula+theyeri Globigerinita glutinata Total Coretop data explained 76% 34% 17% 19% 6% Downcore data explained 93% 27% 25% 39% 2% Downcore+coretop data explained 83% 46% 16% 18% 3% Bold numbers indicate those species that are most in uential in each factor (factor scores 90.3 and v0.3). For Neogloboquadrina pachyderma and Globorotalia truncatulinoides, s indicates sinistral coiling and d indicates dextral coiling. ditions to planktonic foraminifera species assemblages were calibrated using multiple stepwise linear regressions using the transfer function method of Imbrie and Kipp (1971). For each of the three factor de nitions, coretop loadings were regressed against modern annual average SST, seasonal SST range (warmest three months minus coldest three months), primary productivity, pycnocline depth, and mixed-layer depth. Each of the equations created for estimating oceanographic parameters from planktonic foraminifera was then applied to the modern samples and to the 28 samples from the LGM. All modern annual average data except productivity were obtained from the World Ocean Atlas 1998 (hereafter WOA98) (Ocean Climate Laboratory, 1999), which contains surface and subsurface information for the annual average and each of four seasons in 1 U1 grid boxes. Annual average values for each coretop location were determined using the value for the 1 grid box in which the sample is located. Primary productivity values were estimated from satellite color data (Coastal Zone Color Scanner) (Antoine et al., 1996) and averaged over the 15 grid boxes (0.35 latitude by 0.72 longitude) closest to the core location. This averaging removes some of the noise in the satellite data and approximates the spatial smoothing of the data in WOA98.

9 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 9 Table 5 Q-mode factor score matrix for the downcore+coretop data set Species Equatorial Subpolar Subtropical EBC Orbulina universa Globigerinoides conglobatus Globigerinoides ruber (total) Globigerinoides tenellus Globigerinoides sacculifer (total) Sphaeroidinella dehiscens Globigerinella aequilateralis Globigerinella calida Globigerina bulloides Globigerina falconensis Globigerina digitata Globigerina rubescens Globigerina quinquiloba Neogloboquadrina pachyderma (s) Neogloboquadrina pachyderma (d) Neogloboquadrina dutertrei Globoquadrina conglomerata Globoquadrina hexagona Pulleniatina obliquiloculata Globorotalia in ata Globorotalia truncatulinoides (s) Globorotalia truncatulinoides (d) Globorotalia crassaformis Globorotalia hirsuta Globorotalia scitula+theyeri Globigerinita glutinata Total Coretop data explained 80% 40% 17% 12% 10% Downcore data explained 91% 71% 9% 4% 6% Downcore+coretop data explained 90% 37% 24% 23% 6% Bold numbers indicate those species that are most in uential in each factor (factor scores 90.3 and v0.3). For Neogloboquadrina pachyderma and Globorotalia truncatulinoides, s indicates sinistral coiling and d indicates dextral coiling. 5. Results 5.1. Factor analysis Factor analysis performed on the coretop, downcore and coretop+downcore samples produced results which are similar for the three data sets. Analysis of the 232 coretop samples revealed four signi cant faunal factors, each of which is dominated by a unique combination of foraminiferal species (Table 3). These four factors explain 87% of the coretop data. The coretop factors were applied to the downcore and the downcore+coretop data sets to assess the extent to which the coretop factors describe these samples. The squares of the factor loadings divided by the number of samples in each data set to determines the fraction of information in each data set accounted for by the coretop factors. The coretop factors explained 83% of both the downcore and the downcore+coretop samples (Table 3). Holocene and Pleistocene samples from seven cores (Table 2) in the Eastern Paci c were included in the downcore data set. Analysis of these 535 samples resulted in four signi cant factors that explain 93% of the variance in the downcore data. The factors were roughly similar in species composition to those from the coretop analysis. The downcore factor accounted for 76% of the

10 10 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 Fig. 3. Spatial distribution of factor communalities for the (a) coretop, (b) downcore, and (c) downcore+coretop equations. The communalities indicate how well each sample can be described by the factors de ned with each of these data sets. The communalities are similar for all three cases, with high values in the central and eastern Paci c where coretop samples are most abundant and lower values further west where coretop samples are less common. coretop and 83% of the downcore+coretop data (Table 4). The coretop and downcore data sets were combined for a third factor analysis on the 767 samples. Four factors were again retained, explaining 90% of the variance in the downcore+coretop samples. This set of factors, which includes both the coretop and downcore samples, explains 91% of the downcore data and 80% of the coretop data (Table 5) Communalities The communality of each of the samples indicates how well that sample can be described by the given set of factors. The communalities of the coretop factors on the coretop samples were generally high ( s 0.8) averaging Samples with low communalities ( 6 0.7) were con ned to the northwest and southwest regions. Environmental conditions of those regions are not as well represented due to sparser coretop sampling (Fig. 3a). The communalities of the downcore factors applied to the coretop samples were lower than those of the coretop factors, averaging 0.76 and were markedly decreased in the samples having low coretop communalities (Fig. 3b). The communalities of the downcore+coretop factors applied to the coretop samples were between those of the other two cases with a mean of 0.80 (Fig. 3c) Faunal factors The distribution of the factor loadings in modern sediments indicates that each factor dominates an oceanographically distinct region of the Southeast Paci c Ocean and that these regions are similar for the three data sets. The factors have been named to re ect this distribution. Factor 1 has highest loadings in the warm equatorial waters and is referred to here as the Equatorial factor (Fig. 4). This factor explains 33% of the variance in the coretop data, 27% in the downcore samples, and 37% in the downcor-

11 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 11 Fig. 4. Spatial distribution of factor loadings of the Equatorial factor for the (a) coretop, (b) downcore, and (c) downcore+coretop equations. This factor explains 33% of the coretop data, 27% of the downcore data, and 37% of the downcore+coretop data. High loadings of this factor are concentrated in across the equatorial Paci c for all three of the equations. e+coretop samples. The coretop Equatorial factor is high in the species Neogloboquadrina dutertrei and Pulleniatina obliquiloculata, while the downcore and coretop+downcore Equatorial factors have only a small contribution from P. obliquiloculata. The spatial distribution of the Equatorial factor is roughly similar for each of the three data sets (Fig. 4). Factor 2, the Subpolar factor, dominates the cooler, subpolar regions of the modern Paci c Ocean (Fig. 5). The coretop Subpolar factor is comprised primarily of Globigerina bulloides, Globorotalia truncatulinoides (dextral), Neogloboquadrina pachyderma (sinistral), and Globorotalia in- ata, and accounts for 13% of the total coretop data. The downcore Subpolar factor is highest in the species G. bulloides, N. pachyderma (dextral), N. pachyderma (sinistral) and G. in ata. The positive score of N. pachyderma (dextral) in this factor relative to a negative score in the coretop factor, and the lack of G. truncatulinoides (dextral) causes this assemblage to have high loadings in both the subpolar regions and the eastern boundary current south of approximately 15 S latitude (Fig. 5b). This factor accounts for 25% of the downcore data. The coretop+downcore Subpolar factor explains 24% of the samples in this data set and is most similar in species composition and distribution to the downcore factor (Fig. 5c). Globigerinoides ruber, Globigerinoides sacculifer and Globigerinita glutinata are the primary foraminiferal species found in Factor 3 in the three data sets. This assemblage has been named the Subtropical factor, as it is most abundant in the warm oligotrophic waters north and south of the equator (Fig. 6). The Subpolar factor explains 24% of the coretop data (Fig. 6a), 39% of the downcore data (Fig. 6b), and 23% of the downcore+coretop data (Fig. 6c), and has a similar geographic range for each of the data sets. The fourth faunal factor is most prevalent in the California and Peru Currents and has thus been designated the Eastern Boundary Current

12 12 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 Fig. 5. Spatial distribution of factor loadings of the Subpolar factor which explains 13% of the coretop data (a), 25% of the downcore data (b), and 24% of the downcore+coretop equations (c). High loadings of the Subpolar factor in modern sediments are found in the high northern and southern latitudes of the Paci c and, in the case of the downcore and downcore+coretop equations, along the eastern boundary of South America south of V15 S. (EBC) fauna (Fig. 7). This factor also has high loadings in the subpolar regions in the Southern Hemisphere, suggesting that there is some geographic overlap with the Subpolar factor. The coretop EBC factor is dominated by the species Neogloboquadrina pachyderma (dextral), N. pachyderma (sinistral), and Globigerina bulloides and explains 17% of the coretop data. The downcore EBC factor has a slightly di erent species composition, in uenced primarily by G. bulloides (with positive scores) and N. pachyderma (dextral) and Globorotalia in ata (both with negative scores). The EBC factor explains only 2% of the downcore data but was retained for consistency in comparing results to those from the coretop data. Note that unlike the coretop factor, which had positive loadings in both the subpolar and coastal regions, the downcore factor has positive loadings o shore and negative loadings near the coast (Fig. 7b). The downcore EBC factor, therefore, appears to do a better job of separating the e ects of coastal and o shore upwelling than the coretop EBC factor. The downcore+coretop EBC factor is similar to that of the downcore factor but is opposite in sign and is not strongly in uenced by Globorotalia in ata. The geographic distribution of this factor appears to do the best job of depicting highly productive regions of coastal upwelling. The downcore+coretop EBC factor is also relatively abundant along the equator and here may re ect a coastal and equatorial upwelling (Fig. 7c). The downcore+coretop EBC factor has positive loadings of Neogloboquadrina pachyderma (dextral), Pulleniatina obliquiloculata, and Neogloboquadrina dutertrei, and negative loadings for Globigerina bulloides, and explains 6% of the coretop+downcore data. Retaining a fth factor from the factor analysis contributed 92% additional information to the data sets. As this factor added little additional information and did not appear to have an oce-

13 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 13 Fig. 6. Spatial distribution of factor loadings of the Subtropical factor for the (a) coretop, (b) downcore, and (c) downcore+ coretop equations. This factor explains 24% of the coretop, 39% of the downcore, and 23% of the downcore+coretop data. High loadings of this factor are associated with the warm, oligotrophic gyres north and south of the equator. anographically meaningful distribution, it was excluded. which are strongly negatively correlated (Table 6). These highly correlated variables can Multiple regression analysis We performed multiple regression analysis on the coretop factors and WOA98 annual average sea-surface temperature (SST mean ), seasonal range of temperature (SST range ), mixed-layer depth (MLD), primary productivity (PP), and pycnocline depth (PYC). MLD is de ned here as the depth at which density is units greater than the minimum (sea-surface) density. PYC is de ned as the depth of the shallowest maximum density gradient below the MLD. PP, MLD, and PYC are independent of SST range (i.e., r is not signi cantly di erent from zero at the 95% con- dence level). Other oceanographic variables were only weakly correlated to one another with the exception of PYC and MLD, which are strongly positively correlated, and SST mean and SST range, Table 6 Correlation matrix (r) documents relationships among oceanographic variables at the coretop sites used here for calibration of paleoenvironmental proxies PP MLD PYC SST range SST mean SST range PYC MLD Here, SST mean is the annual average temperature at the sea surface. SST range is range in seasonal SST variations (warmest 3 months minus coolest 3 months), MLD is mixed-layer depth, de ned as the depth at which density (c t ) is units greater than at the sea surface, PYC = pycnocline depth, de ned as the depth of the shallowest maximum in vertical density gradient, and PP is primary productivity, based on satellite color measurements (Antoine et al., 1996).

14 14 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 Fig. 7. Spatial distribution of factor loadings of the EBC factor, most prevalent in the California and Peru Currents. This factor explains 17% of the coretop data (a), 2% of the downcore data (b), and 6% of the downcore+coretop data (c). Note the change in sign of the loadings in panel b. not be estimated independently of each other using this calibration data set. The factors in all three of the equations explained v95% of the SST data (i.e., r 2 values are v0.95, Table 7). For the other variables, a smaller fraction of the variance is explained: 77^ 81% for SST range and MLD, 65^71% for PYC, and 44^58% for PP. Thus SST mean is best explained by the factors and is most likely to be estimated correctly from the regression equations. Regressions performed on annual average SST and the (A) coretop, (B) downcore, and (C) downcore+coretop factors yielded the following equations, respectively; ðaþ SST mean ¼ 4:26ðF 1 Þ 2 þ 8:16ðF 1 ÞðF 4 Þþ 17:77ðF 3 ÞðF 4 Þ39:08ðF 2 Þþ4:75ðF 3 Þ 313:78ðF 4 Þþ21:81 ð1þ Table 7 Root mean square error (RMSE) and fraction of variance explained (r 2 ) resulting from estimates of oceanographic parameters using the coretop, downcore and downcore+coretop equations Coretop equation Downcore equation Downcore+coretop equation RMSE r 2 RMSE r 2 RMSE r 2 SST mean ( C) SST range ( C) PYC (m) MLD (m) PP (g C/m 2 /yr) Here, SST mean is the annual-average temperature at the sea surface. SST range is range in seasonal SST variations (warmest three months minus coolest three months), MLD is mixed-layer depth, de ned as the depth at which density (c t ) is units greater than at the sea surface, PYC = pycnocline depth, de ned as the depth of the shallowest maximum in vertical density gradient, and PP is primary productivity, based on satellite color measurements (Antoine et al., 1996).

15 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 15 ðbþ SST mean ¼ 10:18ðF 2 Þ 2 þ 2:67ðF 3 Þ 2 þ 7:64ðF 1 ÞðF 3 Þþ9:1ðF 1 ÞðF 4 Þ36:88ðF 2 ÞðF 4 Þþ 9:81ðF 3 Þ318:55ðF 2 Þþ20:55 ðcþ SST mean ¼ 4:46ðF 1 ÞðF 3 Þþ7:03ðF 1 Þ 312:46ðF 2 Þþ6:35ðF 3 Þþ20:08 ð2þ ð3þ where SST mean is average annual sea-surface temperature, F 1 is the factor loading of the Equatorial factor, F 2 is the factor loading for the Subpolar factor, F 3 is the factor loading for the Subtropical factor, and F 4 is the factor loading for the EBC factor. All terms included in these equations are signi cant at the 95% con dence level. The regression coe cients for SST mean, SST range, MLD, PYC, and PP are given in the Appendix SST estimates: modern We used the regression equations to estimate SST from the factors for each data set. There is little correlation between the residuals of SST and the SST estimates (Fig. 8), implying that there is no systematic bias in the equations with respect to SST. The residuals are also essentially random with respect to the observed pycnocline depth (r = 0.07), productivity (r = 0.20), mixed layer depth (r = 0.21), and seasonal range of SST (r = 0.26). The standard deviations of the SST residuals (RMSE) are 1.8 C, 1.9 C and 2.0 C for the equations based on the coretop, downcore and downcore+coretop data, respectively. Observed average annual SSTs from WOA98 (Fig. 1b) and SSTs estimated at the coretop locations from each of the equations (Fig. 9) are very similar. The residuals have a roughly Gaussian distribution that is slightly skewed towards underestimating SST. In all three equations SSTs are underestimated in the western section of the study area and in the Panama Basin, and overestimated along the area roughly following the East Paci c Rise and along the South American coast (Fig. 10). Underestimation in the western region is not surprising as there are fewer samples in this area and communalities are lowest there (Fig. 3). High residuals in the Panama Basin have previously been noted using both foraminifera (Mix et al., 1999) and radiolarians (Pisias and Mix, 1997) to estimate SST. This area of high residuals corresponds with anomalously low surface salinities, which maintain an exceptionally shallow pycnocline. Low surface salinities may cause the plankton to live deeper in the water column (Fairbanks et al., 1982) and therefore a ect estimates of surface temperatures. Overestimation of SST by samples along the coastline and East Paci c Rise suggest that depth, and therefore dissolution of the carbonate tests, may be contributing to SST estimates that are too warm in these areas. However, when SST residuals are compared to the water depth of the samples, no correlation is observed (r = 0) SST estimates: LGM Samples from 28 cores located in the Eastern Paci c were used to estimate SSTs for this region. The samples are concentrated in the area between 10 N and 10 S latitude and include new samples from cores on the Nazca Rise. The three equations result in roughly similar SST patterns, with the coldest water located along the eastern boundary and warmer water o shore (Fig. 11). The downcore equation results in slightly warmer temperatures along the coast but there is no signi cant di erence between the LGM SSTs estimated by these equations. LGM samples located nearest to the equator and in the EBC are described well (communalities v0.8) by all of the factor models (Fig. 12). Given the scarcity of coretop samples located in the subtropical gyre and south of 10 S, it is not surprising that the factors do not do a particularly good job of describing the southern LGM samples. The communality values are signi cantly correlated to latitude with decreasing communalities to the south. In general, communalities for the LGM samples are similar for the three equations. Including ancient samples in the calibration improved the communalities slightly, but

16 16 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 Fig. 8. Estimated vs. observed SST (left) and SST residuals vs. observed SST (right) as determined by the coretop (panels a and e), downcore (panels b and f) and downcore+coretop (panels c and g) equations. SSTs are estimated well from the equations (r 2 v 0.95 and RMSE 9 2 C for all three equations). The essentially zero correlation between the observed SST and the SST residuals reveals that there is no systematic bias of the SST estimates based on the observed SST, except perhaps at sites where observed temperatures are s 27 C in the Panama Basin area.

17 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 17 Fig. 9. Modern average annual SSTs as estimated by the (a) coretop, (b) downcore, and (c) downcore+coretop equations. These maps show that all three of the calibration equations estimate essentially the same SSTs from modern sediments and that these estimates are generally similar to observed modern SSTs (Fig. 1b). the improvement was not substantial in this area Estimates of SST change: LGM3modern Subtracting the modern estimates from the LGM estimates for each of the three calibration equations yielded estimates of the change in average annual SST between LGM and the present. Modern estimates were used rather than WOA98 data so that inherent biases in the transfer function would be minimized (Mix et al., 1999). Due to the di erence in the number of modern and LGM data points, modern temperature estimates were interpolated onto a 1 U1 grid. The value at the grid point nearest to each LGM sample location was used as the modern estimate for that site. We determined the SST anomaly by subtracting this gridded modern temperature estimate from the LGM estimate. Maps of the di erence between LGM and modern SST estimates reveal signi cant cooling in the eastern tropical Paci c during the LGM relative to modern (Fig. 13). Temperature changes in the equatorial region are 4^6 C o the coast of Ecuador, and 1^3 C west of V95 W longitude. These results, comparable to those of Mix et al. (1999) and Pisias and Mix (1997), are considerably different from CLIMAP s (1981) estimates of only a 0^2 C decrease in SST over the Eastern Equatorial Paci c. The new LGM samples located on the Nazca Rise record a substantial cooling of 6^9 C in the Peru Current during the LGM. Cooling is greatest near the coast resulting in an increase in the thermal gradient across the Peru Current. The LGM SST estimates are 3^5 C warmer at RR9702A-63 than at RR9702A-69, compared to a coretop gradient of 0.5^1.0 C. This enhanced thermal gradient likely indicates increased strength of equatorward advection. The three equations yield similar LGM cooling patterns relative to modern estimates (Fig. 13). Along the equator, SSTs from the downcore and

18 18 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 Fig. 10. SST residuals (estimated SST minus observed SST) for the (a) coretop, (b) downcore, and (c) downcore+coretop equations. Residuals are random with respect to water depth (r = 0). The di erences between estimated and observed SSTs systematically negative in the Panama Basin and Gulf of Tehuantepec, where low surface salinities maintain an exceptionally shallow pycnocline. This may a ect SST estimates by increasing the habitats available for foraminifera to live in cooler subsurface waters. downcore+coretop equations are similar to each other but slightly cooler than those derived from the coretop equation. Temperature anomalies in the Peru Current agree well for all three equations. 6. Discussion 6.1. Comparison of calibration equations One goal of this study was to determine if a regional calibration and the inclusion of ancient samples in a calibration equation would improve SST estimates for the LGM in the EBC of the Southeast Paci c. A comparison of results from the three equations used in this study shows that the statistical error in the calibration is slightly higher when ancient samples are included in the factor de nitions (1.8 vs. 2.0 C). When the downcore samples are added to the calibration, the extent to which the factors describe the LGM samples (i.e., the communalities) increases. With the exception of the equatorial band, however, the various methods yield similar results, within statistical error. Temperatures estimated from the three equations employed here are similar because the LGM samples have good modern analogs in the Southeast Paci c. Because the region used for calibration included coretop samples from 50 N to 50 S in the Eastern Paci c, the modern variability of the fauna was su cient to describe the LGM samples in the eastern boundary current. The addition of 16 modern samples from the Southeast Paci c to this study increased the range of modern oceanographic conditions represented by the coretop data. The two samples with the lowest communalities for the coretop calibration at the LGM did not improve to acceptable values even

19 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 19 Fig. 11. LGM average annual SSTs as estimated by the (a) coretop, (b) downcore, and (c) downcore+coretop equations applied to 28 samples. The three equations results in roughly similar SST patterns, with the coldest water located along the Eastern Boundary and warmer water o shore. when the ancient samples were added (Fig. 12). This indicates that there is another controlling factor on these samples that makes them poorly represented in the calibration data, regardless of the fact that ancient samples were used LGM cooling of the Eastern Equatorial Paci c Foraminiferal transfer functions indicate significant cooling in the Eastern Equatorial and Southeast Paci c during the LGM (Mix et al., 1999). Presently, there are two primary sources of cold water responsible for forming the equatorial cold tongue: (a) equatorial upwelling of cold water from the EUC into the SEC and (b) advection of cold water from the Peru Current into the SEC (Wyrtki, 1981). Changes in one or both of these mechanisms are therefore the likely source of glacial cooling in the Eastern Equatorial Pacific. We expect that an increase in equatorial upwelling during the LGM would be accompanied by an increase in the Equatorial faunal factor. In the modern ocean, this factor displays a pattern centered on, and roughly symmetrical about, the equator (Fig. 4). The LGM Equatorial factor is also high on the equator with loadings essentially the same as for the coretops (Fig. 14a). There is little di erence in the distribution of this factor between the present and the LGM, suggesting little change in equatorial upwelling from modern conditions. Previous studies have addressed the issue of LGM equatorial upwelling with con icting results. Using oxygen isotopes of planktonic foraminifera from a core in the eastern Paci c (TR163-31B, 03 58PS, 83 58PW), Patrick and Thunell (1997) infer a glacial decrease in the temperature gradient between the sea surface and the thermocline. They cite this decrease as evidence for a shoaling thermocline in the Eastern Equato-

20 20 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 Fig. 12. Communalities of LGM samples with the (a) coretop, (b) downcore, and (c) downcore+coretop equations. Communalities are higher closer to the equator and decrease further to the south. Incorporation of ancient samples in the calibration (b and c) improved the mean communalities slightly, but the improvement was not substantial in this area. rial Paci c and conclude that it is most likely the result of an increase in upwelling along the equator, although their site is not on the equator. Andreason and Ravelo (1997) studied changes to the thermocline depth in the Eastern Equatorial Paci c using foraminiferal transfer functions. They found only small changes between LGM and modern mixed-layer depths and concluded that divergent equatorial upwelling could not have caused signi cant cooling at the sea surface (Andreason and Ravelo, 1997). This result agrees with that of Romine (1982), who nds a decrease in the radiolarian faunal factor associated with the EUC during the LGM. Studies of biological productivity have also attempted to evaluate the strength of equatorial upwelling during the LGM. Loubere (1999) infers decreased biological productivity from benthic foraminiferal assemblages, and suggests a reduced supply of nutrients to the surface waters of the Eastern Equatorial Paci c during the LGM. Based on lack of a correlation between these productivity estimates and records of SST in the Eastern Equatorial Paci c, Loubere (2000) indicates that cooling was not the result of increased upwelling on glacial^interglacial timescales. These inferences based on benthic foraminifera may be in con ict with productivity estimates based on organic carbon accumulation rates (Lyle et al., 1988). Lyle et al. (1992) used alkenone temperature estimates to evaluate the change in upwelling ux in the central Paci c. They assumed that cooling, if caused by upwelling, should correspond to high organic carbon burial due to increased biological productivity. During the LGM, these records show a decrease in both SST and productivity relative to modern values, opposite the relationship predicted for upwelling, suggesting that an increase in upwelling was not responsible for the glacial cooling (Lyle et al., 1992). Based on these studies and our observation of little or no change in the equatorial upwelling

21 M.J. Feldberg, A.C. Mix / Marine Micropaleontology 44 (2002) 1^29 21 Fig. 13. Temperature anomalies (estimated LGM SST minus estimated modern SST) as estimated by the (a) coretop, (b) downcore, and (c) downcore+coretop equations. SSTs were 6^9 C cooler in the EBC and 3^5 C cooler along the equator of the Eastern Paci c during the LGM than at present. fauna, we conclude that a mechanism other than upwelling must be largely responsible for LGM cooling of the Eastern Equatorial Paci c. The second mechanism for cooling the tropical Paci c is the advection of cool water from the Peru Current into the South Equatorial Current. Presently, there are two sources of cool water in the Peru Current: surface waters originating in the eastern boundary current (Strub et al., 1998) and water from the Poleward Undercurrent upwelled along the coast of Peru, via the Peru Coastal Current (Toggweiler and Dixon, 1991). If either northward ow in the Peru Current or coastal upwelling was increased during the LGM relative to today, the eastern boundary could have brought more and/or colder water to the equator and may have been responsible for cooling the tropics. Results from cores on the Nazca Rise indicate that the Peru Current experienced substantial cooling during the LGM (Fig. 13). A comparison of the distribution of the Subpolar factor in the modern ocean (Fig. 5) and at the LGM (Fig. 14b) reveals that in the modern ocean this assemblage is dominant in the colder waters south of approximately 35 S, while years ago this factor was abundant in the waters of the PCC. This indicates that cold water (with which the subpolar fauna is associated) was advected northward in the eastern boundary current at greater rates during glacial times than at present. We infer that advection of Peru Current water into the equatorial cold tongue is at least partially responsible for LGM cooling in the Eastern Equatorial Paci c. Evidence for increased advection in the Peru Current during the LGM agrees well with other studies from this region. At ODP Site 846 located southeast of the Galapagos Archipelago, an LGM increase of Globigerina bulloides was interpreted as an increase in the advection of cold water o the eastern boundary into the SEC, due to either an increase in the intensity of the Peru Current or