Sea surface temperature at the Last Glacial Maximum: A reconstruction using the modern analog technique



[1] We have applied the modern analog technique (MAT) to reconstruct sea surface temperature (SST) at the Last Glacial Maximum (LGM) using a database of 292 planktonic foraminiferal samples from the Atlantic, Indian, and Pacific Oceans. LGM SST maps reveal many important patterns and gradients which reflect the climate and circulation patterns of the LGM. One of the most important results is the 2°–6°C mean annual cooling in the central to eastern tropical Atlantic and up 8°C cooling in the eastern tropical Pacific. These changes resulted in an increased east-west SST gradient in the equatorial regions of these two basins, compared to modern. SSTs in eastern boundary currents of all three ocean basins were cooler than modern at the LGM. Additionally, results indicate more zonal flow patterns in the Kuroshio/North Pacific Current and Gulf Stream/North Atlantic Drift. These changes indicate more intense midlatitude-to-equator SST gradients at the LGM, which likely resulted in more intense circulation and increased zonal winds. SSTs in much of the tropical and northern Indian Ocean were within 2°C of modern temperatures. Consistent with the results of previous studies, results indicate little change in much of the western tropics and the subtropics of all ocean basins.

1. Introduction

[2] The sea surface is the interface between the ocean and atmosphere and, as such, is a key player in both oceanic and atmospheric circulation. Sea surface temperature (SST) represents an important boundary condition for both these circulation systems because it determines the direction and rate of heat and moisture transfer between ocean and atmosphere. The time period of the Last Glacial Maximum (LGM) is of interest to oceanographers and climate modelers due to the significantly different boundary conditions from modern in terms of ice caps, sea level, vegetation and continental climate. The question of how different LGM SSTs were from modern remains controversial. An accurate reconstruction of the LGM will enhance our understanding of the large-scale patterns of change of the ocean-atmosphere system, and is useful for the testing of climate models, which have been used to predict future climates. This research is focused on the tropical and subtropical oceans since these regions are the primary source of heat to the atmosphere and the temperature stability of the tropics affects the results of climate models. For these reasons, an accurate reconstruction of Last Glacial Maximum SSTs is vital to our understanding of past and future climates.

[3] While a LGM sea surface temperature reconstruction was done many years ago using faunal/floral census counts of four different groups of plankton [CLIMAP Project Members, 1976, 1981], some of the results of that study have been questioned. In particular, small temperature changes in the low-latitude oceans between the modern and LGM, as found by CLIMAP, have been challenged by many researchers working on both terrestrial and marine data. Snow line depressions on tropical mountains [Webster and Streton, 1979; Rind and Peteet, 1985] and noble gas temperatures from low-latitude groundwater [Stute et al., 1995] represent terrestrial evidence of tropical cooling at the LGM of ∼5°C. In the oceans, Sr/Ca ratios in corals suggest similar tropical cooling [Guilderson et al., 1994; Beck et al., 1997]. On the other hand, oxygen isotopes of foraminifera [Broecker, 1986; Birchfield, 1987; Billups and Spero, 1996; Stott and Tang, 1996] and alkenone results from coccoliths [Lyle et al., 1992; Rostek et al., 1993, Sikes and Keigwin, 1994] are consistent with the CLIMAP results of modest tropical SST cooling at the LGM (see Broecker [1996] for detailed discussion). Terrestrial evidence of modest cooling (2°–3°C) in the western Pacific was found by Hope and Tulip [1994] based on lowland Irian Jaya pollen records at an altitude of 780 m.

[4] Each method of terrestrial and sea surface temperature reconstruction contains assumptions and uncertainties, which may contribute to discrepancies in the estimates. Many of the Pacific CLIMAP LGM SST estimates were based on averages of up to four different faunal/floral groups. These groups (diatoms, coccoliths, foraminifera and radiolaria) sometimes resulted in different SST estimates for a single sample, and thus values were eliminated and/or averaged to make the final estimate. Our study, by contrast, uses only one faunal group (planktonic foraminifera) for the estimates. This approach has the drawback of limiting the geographical extent of our reconstruction since carbonate sediments are affected by dissolution. A second potential source of error in previous biotic reconstructions was taxonomic inconsistency, among numerous workers. We have used a new and expanded core top database that is taxonomically consistent, with most samples checked by a single worker. Our study represents a significant reanalysis of both old and new foraminiferal data. We also use a modified version of the modern analog technique [Hutson, 1980; Prell, 1985], which has been shown to provide better SST estimates than the transfer function technique developed by Imbrie and Kipp [1971] if the calibration database is adequate [Ortiz and Mix, 1997; Gonzalez-Donoso and Linares, 1998]. The MAT has several advantages over the Imbrie and Kipp method. Most importantly, SST estimates provided by the MAT have higher correlation coefficients with observed SSTs than those produced by CLIMAP transfer functions [Prell, 1985; Ortiz and Mix, 1997, Pflaumann et al., 1996]. The geographic location of the analog samples provides additional information useful for reconstruction of the paleoceanographic regime of the sample site. A distinct advantage of the MAT over the transfer function method is that the MAT can be less sensitive to estimation biases introduced by poorly preserved samples.

[5] This study provides a SST reconstruction of the LGM ocean based on faunal distributions of planktonic foraminifera in 292 deep-sea sediment samples. We define the Last Glacial Maximum as the interval of maximum global glaciation, as indicated by marine oxygen isotopes (i.e. the approximate midpoint of stage 2). We present the first large-scale LGM SST reconstruction since the CLIMAP reconstruction. Since we have used only foraminiferal data in the reconstruction, the Southern Ocean and portions of the Pacific will not be reconstructed. These regions are lacking carbonate sediments due to dissolution. Much of the controversy surrounding the CLIMAP reconstruction may be due to extrapolation of results outside of the geographical extent of the data [Broccoli and Marciniak, 1996]. To create an objective reconstruction, we do not extrapolate significantly from the foraminiferal data. Maps of the LGM are presented along with maps showing point to point differences in SST estimates between this reconstruction and the CLIMAP reconstruction (Figure 7). Given significant improvements in our data sets and SST estimation techniques, we hope to answer the following questions:

  1. How much did the tropics cool at the LGM?
  2. How do our improved SST estimates compare with the CLIMAP results? Do regional differences exist between our reconstruction and CLIMAP?
  3. What changes in circulation dynamics are indicated in the low-latitude temperature gradients?

2. Data

[6] All samples used in this reconstruction consist of faunal counts of planktonic foraminifera from deep-sea sediment samples. As mentioned above, our data sets are an improvement over previous data sets in that all core top calibration samples and 83% of the LGM samples have been reviewed by a single worker for taxonomic consistency. The exceptions are as follows: (1) 39 Pacific LGM samples from Thompson [1981] were unavailable for recounting but follow the taxonomy of Parker [1962] and are consistent with our taxonomy. Since Thompson's core top taxonomy was consistent with ours, we assume the LGM samples are consistent as well. (2) We have also included 10 LGM samples from Ortiz et al. [1997] in the region of the California Current. While these samples were not verified at Brown, we believe that the taxonomy used is consistent with ours (J. Ortiz, personal communication, 2000). The taxonomy of A. Martin et al. (manuscript in preparation, 2000) was used for all other samples. This combines taxonomy of Parker [1962] and that of Be [1967]. We have eliminated the artificially created P-D intergrade [Kipp, 1976] and identify 40 species and coiling varieties in the >150 μm size fraction. Samples were prepared as outlined by Imbrie and Kipp [1971]. We have counted at least 300 specimens in 86% of the samples and at least 200 specimens in the remaining samples. Many samples are those originally used in the CLIMAP Project Members [1981] study.

2.1. Brown Last Glacial Maximum Database (BLGM)

[7] Our LGM database (BLGM) consists of 292 samples located mostly in the Atlantic and Indian Oceans with fewer samples in the Pacific (Figure 1a) [Prell et al., 1999]. Eight Atlantic samples and one Pacific sample do not have adequate modern analogs using a dissimilarity cutoff value of 0.25 (see methods section). Our reconstruction, therefore is based on 283 LGM samples. Of these 283 samples, 270 have 10 modern analogs with dissimilarity values less than the 0.25 cutoff. Compared to CLIMAP Project Members [1981], areas of database expansion include the Indian Ocean where we have 93 LGM samples. This more than doubles the number of samples used in previous Indian Ocean reconstructions [Prell et al., 1980; CLIMAP Project Members, 1981]. We have also expanded the LGM database in the critical region of the Pacific warm pool. LGM depth levels in the cores were located using a variety of techniques including C14 dating, δ18O measurements, carbonate stratigraphy and biostratigraphy. Details of chronostratigraphy are discussed by CLIMAP Project Members [1981]. We have not revised most of the CLIMAP stratigraphy and have accepted the LGM depth levels used in that study except in cases where the stratigraphy has been updated. In new cores, we have identified LGM depth levels using δ18O, and for many cores, we have examined samples immediately above and below the identified LGM level to confirm our selection of this level.

Figure 1.

(a) Location and dissimilarity values of 292 samples in Brown LGM database (BLGM). Asterisks indicate LGM samples which have no modern analogs. (b) Location of 1219 core top samples in the Brown Foraminiferal Database (BFD).

2.2. Brown Foraminiferal Database (BFD)

[8] The modern calibration database (BFD) consists of 1219 samples located throughout the oceans (Figure 1b, data available at or through Brown University [Prell et al., 1999]. Of the 1219 samples, 456 are from the Atlantic, 387 are from the Indian and 376 are from the Pacific Oceans. Coverage in the central North and South Pacific is sparse due to the shallow lysocline in this ocean and resulting lack of carbonate sediments both in the modern and LGM. All BFD samples have been reviewed by April Martin (Brown University) for taxonomic consistency. The accuracy of this reconstruction depends upon the modern data set covering a wide range of hydrographic regimes and preservation states. The BFD contains samples representing the complete spectrum of cold and warm season SSTs found in the global ocean as well as samples from water depths of 200 to 5500 m. This coverage provides satisfactory modern analogs for both well and poorly preserved samples.

[9] The Global Ocean Surface Temperature Atlas (GOSTA) was used for modern (calibration) SSTs [Bottomley et al., 1990]. This data set consists of 30 year (1951–1980) monthly mean SSTs gridded on a 1° × 1° scale. Further information on data processing details may be found in [Bottomley et al., 1990]. We calculated mean annual SSTs by averaging the 12 monthly mean SSTs.

3. Methods

3.1. Modern Analog Technique

[10] The modern analog technique was used for all reconstructions. Based on the results of previous experiments [Trend and Prell, 1996, Trend-Staid, 1999], a global composite species list of 32 species was developed and used in all calculations (Table 1). Previous studies have found the squared chord distance dissimilarity coefficient to reflect most accurately differences in foraminiferal faunal composition [Prell, 1985; Trend-Staid, 1999] and thus we have used this coefficient for all calculations. Application of the MAT requires selection of a dissimilarity cutoff, which is a dissimilarity value above which we reject the sample as an adequate analog. Thus if the closest modern analog to our subject sample has a dissimilarity above the cutoff, we have a no-analog situation, which means that no samples in the modern database closely resemble the sample in question. Selection of this cutoff value has, in the past, been somewhat arbitrary. Waelbroeck et al. [1998] used no cutoff value, but rather selected the number of “best analogs” based on the shape of the dissimilarity versus analog ranking curve. While this approach gives improved correlation coefficients, the results are not particularly sensitive to changes in this “shape” parameter. To calibrate the optimal dissimilarity cutoff value, we combined Q mode factor analysis (QFA) with the MAT to find a cutoff.

Table 1. Brown Foraminiferal Database
Mean Abundance, %Standard DeviationMinimum AbundanceMaximum Abundance
  • a

    Bold samples indicate those used for final composite species list (CSL).

Orbulina universa0.851.98041.18
Globigerinoides conglobatus1.111.69011.40
G. ruber(pink)1.072.82026.26
G. ruber(white)18.3716.84077.92
G. tenellus0.811.30011.34
G. sacculifer (without sac)3.994.37026.05
G. sacculifer (with sac)2.783.24023.57
G. sacculifer(total)    
Sphaeroidinella dehiscens0.320.87013.12
Globigerinella adamsi0.070.3608.07
G. aequilateralis2.832.75017.96
Globigerina calida1.421.4008.03
G. bulloides8.9211.41072.06
G. falconensis3.054.83035.74
Globigerinella digitata0.490.86013.51
Globigerina rubescens0.611.08015.06
Turborotalia humilis0.040.2003.42
Globigerina quinqueloba0.983.60032.99
Neogloboquadrina pachderma(left)6.1119.60098.49
N. pachderma(right)5.5010.67067.61
N. dutertrei7.9212.28087.95
Globoquadrina conglomerata0.551.52013.23
Globorotaloides hexagona0.260.6607.71
Pulleniatina obliquilaculata5.1311.69088.54
Globorotalia inflata7.2213.12081.99
G. trucatulinoides(left)1.202.49017.91
G. trucatulinoides(right)0.921.87015.55
G. crassaformis0.270.76010.51
G. carssula/crotonensis0.030.1201.70
G. hiruta0.561.2508.90
G. scitula0.460.9107.23
G. anafracta0.010.0400.64
G. menardii4.046.55056.02
G. tumida2.385.98058.89
G. menardii neoflexuosa0.090.3405.19
Candeina nitida0.080.2704.34
Globigerinita glutinata9.509.42053.40
Globorotalia theyeri0.050.1801.60
Globigerinita iota0.010.0601.06
Berggrenia pumilio0.000.0400.61
Berggrenia uvula0.000.0000.00

[11] QFA is a technique in which a data set is described in terms of end-member compositions. The data are then described as linear mixtures of these end-members. In QFA (see Imbrie and Kipp [1971] for discussion) the species are grouped into a small number of assemblages, which are statistically independent. Each modern sample is then described as having varying contributions from each assemblage. The degree of contribution of an assemblage to a sample is called its factor loading with respect to that assemblage. The square of the factor loading gives the percent contribution so that a loading of ∼0.71 or higher (>50%), indicates that the sample is dominated by that factor (assemblage). We would like to understand how dissimilarity values relate to assemblage dominance. Our strategy is to find a dissimilarity value which ensures that the analog sample is dominated by the subject sample assemblage.

[12] We performed a QFA on the 1219 samples in the core top database (BFD). The factor analysis was run using CABFAC [Klovan and Imbrie, 1971] with a seven-factor varimax rotation. For each factor, we calculated a loading for each sample in the database; indicating how well correlated each particular modern sample is with each factor. A loading of 1.0 would indicate that the sample is exactly representative of the factor in question. The sample with the highest loading for each factor is designated as an end-member. The locations of the seven end-member samples are shown in Figure 2. These first seven factors account for 90.31% of the variation in the data. End-member samples represent the tropical, subtropical, subpolar, polar and gyre margin assemblages of Imbrie and Kipp [1971] and a coastal upwelling assemblage [Prell, 1984]. Additionally, we identify a tropical dissolution assemblage [Hutson and Prell, 1980] in the western Pacific. The end-member samples represent seven statistically independent assemblages and all samples in the database may be described as linear combinations of them. Factor scores for the seven assemblages are shown in Table 2. The factor scores show the contribution of each species to each assemblage.

Figure 2.

Location of end-member samples found from a Q mode factor analysis of the core top database. V14-004 is tropical, V19-240 is subtropical, RIS-24 is gyre margin, RC27-009 is coastal upwelling, V27-098 is polar, ERDC-128 is tropical dissolution, and V23-079 is subpolar end-member.

Table 2. Varimax Factor Score Matrix for Core Top Databasea
TropicalSubtropicalGyre MarginUpwellingPolarTropical DissolutionSubpolar
  • a

    Bold values indicate the primary species comprising each assemblage.

 O. universa0.0280.0270.0050.0150.0040.005−0.025
 G. conglobatus0.0480.0000.0240.0150.0040.0240.003
 G. ruber pink0.0700.0020.0060.0440.010−0.015−0.020
 G. ruber white0.9100.067−0.0610.0270.019−0.0850.077
 G. tenellus0.0360.004−0.014−0.020−0.005−0.0130.005
 G. sacculifer T0.303−0.0350.1680.0910.0320.138−0.095
 S. dehiscens0.0010.0010.0230.0050.0010.0330.004
 G. adamsi0.004−0.0020.0010.0000.0000.000−0.001
 G. aequilateralis0.1100.0040.0240.0080.0020.111−0.013
 G. calida0.0560.0240.007−0.010−0.0050.0010.009
 G. bulloides−0.0900.201−0.016−0.7800.114−0.0150.143
 G. falconensis0.0300.224−0.062−0.143−0.041−0.0280.191
 G. digitata0.0110.0090.0230.004−0.0010.0070.004
 G. rubescens0.027−0.003−0.010−0.017−0.003−0.004−0.003
 G. quinquiloba−0.0080.007−0.015−0.0280.108−0.001−0.074
 N. pachyderma L0.005−0.020−0.0010.0300.9800.0080.029
 N. pachyderma R−0.0340.3430.0120.0660.056−0.023−0.876
 N. dutertrei0.0040.0660.863−0.066−0.012−0.1270.004
 G. conglomerata0.011−0.0120.041−0.0050.0010.029−0.007
 G. hexagona0.006−0.0080.014−0.015−0.002−0.004−0.003
 P. obliquilaculata0.0100.0180.0270.004−0.0040.9390.015
 G. inflata−0.0100.858−0.0280.118−0.0270.0330.194
 G. trucatulinoides L0.0100.136−0.027−0.013−0.002−0.0060.106
 G. trucatulinoides R0.0390.060−0.0140.025−0.0030.0000.028
 G. crassaformis0.0080.0080.0140.0070.0010.0020.006
 G. hirsuta0.0110.060−0.0110.004−0.006−0.0030.031
 G. scitula0.0120.032−0.007−0.005−0.006−0.0030.008
 G. menardii0.039−0.0390.406−0.062−0.0040.0240.036
 G. tumida−0.0120.0060.2040.0180.0030.2260.056
 G. flexuosa0.000−0.0030.008−0.006−0.0010.0000.001
 C. nitida0.005−0.001−0.0010.0010.000−0.001−0.001
 G. glutinata0.202−0.120−0.077−0.574−0.0890.077−0.314
 G. iota−0.001−0.0010.000−0.007−0.001−0.0010.002
Factor loading0.9900.9660.964−0.9730.9860.963−0.862

[13] We used the MAT to compare these end-member samples to the entire BFD and calculate a dissimilarity coefficient between each end-member and every other sample in the database. Factor loading versus dissimilarity is plotted in Figure 3. The area below the solid horizontal line is the region of samples which have a factor loading of ≥0.71 indicating that they contain at least 50% of the subject sample assemblage. We want to select a dissimilarity value which ensures dominance of the subject sample assemblage. The dissimilarity cutoff should be selected such that we minimize the acceptance of samples which have a factor loading of <0.71. In general, a cutoff of 0.25 (solid vertical line in Figure 3) is equivalent to factor loading of 0.8 or higher for most end-member samples. Based on these results, we use a conservative cutoff of 0.25 for all calculations. SST estimates are based on the average SSTs of the top ten analog samples. If fewer than ten samples have dissimilarity values less than the cutoff, only those samples are averaged. Of the 283 LGM samples which have modern analogs, only 13 samples have fewer than 10 analogs with dissimilarity values <0.25. Average dissimilarity values for each LGM sample are presented in Figure 1a.

Figure 3.

Factor loadings for each sample in the BFD versus dissimilarity between each sample and the respective end-members. Horizontal line indicates factor loading of 0.7 (sample contains 49% of factor). Vertical line indicates 0.25 dissimilarity. Use of a 0.25 dissimilarity rarely results in acceptance of a sample which, in fact, is not dominated by that factor.

[14] Experiments with various averaging schemes indicate little difference in SST estimates resulting from the average of the top 5 versus top 10 analogs [Trend-Staid, 1999]. No weighting function is used when averaging the samples. This approach differs from previous methods which have weighted analogs based on similarity [Prell, 1985; Pflaumann et al., 1996] or on inverse geographic distance [Pflaumann et al., 1996]. We have found that SST estimates are relatively insensitive to weighting based on similarity. While inverse geographic distance provides high correlation coefficients between modern observed and estimated temperatures, this approach constrains the paleotemperatures to resemble modern temperatures, to some extent.

[15] In identifying analog samples, we have placed some limitations on the search. For subject samples located 20°N to 20°S, all core top samples in the BFD are available as analogs. For subject samples located north of 20°N latitude, we have limited the analog search to core top samples 10°S to 90°N and for those subject samples located south of 20°S, we have limited the analog search to core top samples 10°N to 90°S. This convention is used because some samples from opposite hemispheres contain similar faunal compositions, while SSTs differ by 2° or more. This may be due to geographic isolation of the fauna. Because of this phenomenon, lumping northern and southern populations would add uncertainty to the estimates. Using the convention of split hemispheres prevents averaging of the SSTs from different hemispheres.

[16] The ability of the MAT to recover the modern calibration data (not using the subject sample as an analog for itself) is one test of the accuracy of the technique. Using the modern database of 1219 samples we calculate the standard error (SEE) for mean annual SST estimates to be 1.02°C. While the standard error for the core top data set, as a whole, is relatively small, the standard deviation of the estimates for individual samples is larger. In particular, samples located near oceanographic fronts tend to have higher standard deviations due to the selection of analog samples from both sides of the front.

[17] Since LGM SST estimates are calculated based on averaging the SSTs of the top 10 analogs, we calculate a standard deviation of the SST estimates for each LGM sample. We have greater confidence in estimated SSTs of samples characterized by smaller standard deviations than anomalies (LGM minus modern) than those samples for which the standard deviation is larger than the anomaly. For samples with relatively highstandard deviations, spatial coherence of the SST estimate with surrounding samples lends confidence to the reconstruction.

[18] A comparison of this application of the MAT to previous uses of this technique for SST reconstruction is presented in Table 3. Only this study and Prell [1985] have applied the technique to a global data set. Prell [1985] used the MAT to verify CLIMAP Project Members [1981] results but did not create a new SST reconstruction. Pflaumann et al. [1996], developed the SIMMAX technique but did not use it for SST reconstruction. The SIMMAX technique differs from our approach by weighting analogs based on their geographic proximity to the subject sample. The revised analog method (RAM) of Waelbrock et al. [1998] differs from our approach by not using a fixed dissimilarity cutoff or fixed number of analogs.

Table 3. Applications of the MAT
StudyMethodNumber of SpeciesCalibration DatabaseDissimilarity MeasureNumber of AnalogsCutoffRegion
  • a

    Not available in published literature.

This studyMAT321219squared chord100.25global
Prell [1985]MAT331145squared chord10noneglobal
Anderson et al. [1989]MATN/Aa1145squared chord10noneCoral Sea
Thunell et al. [1994]MATN/Aa1145squared chord100.4western Pacific
Pflaumann et al. [1996]SIMMAX26738scalar product100.79no reconstruction
Ortiz et al. [1997]MAT271121squared chord50.2California Current
Waelbrock et al. [1998]RAMN/Aa1582squared chordvariesvariesNorth Atlantic

3.2. Interpolation and Contouring

[19] In producing the contoured maps, we have contoured objectively, to the extent that this produces reasonable oceanographic SST fields. In all cases, SST estimates at actual LGM data points were preserved, however, strategically placed control points have been used to constrain the SST field and prevent unreasonable extrapolations. The SST estimates at LGM sample locations were interpolated in Spyglass Transform® using a linear kriging interpolation algorithm and default kernel settings, which results in accurate, objective machine contours. This scheme interpolates so as to minimize the statistical variance in the data. Missing data are filled based on the nearest true data values and the weighting of the true data values drops off linearly with distance from the missing value [Davis, 1986]. While this procedure provides accurate contouring in regions where data coverage is good, it produces errors in regions with sparse data, as it tends to result in circular contours being drawn around the areas of better coverage. Since objective contouring ignores latitudinal temperature gradients, we occasionally get isotherms trending north-south, or in some other unreasonable configuration. This is clearly an artifact of contouring unevenly distributed data. To combat this effect, we have, in some cases, used control points (marked by a “C” on LGM maps). Control points are based on alkenone SST estimates, the location of continental ice sheets and sea ice limits. We assume a one to one relationship between foraminiferal and alkenone-based SST estimates. In the southeast Pacific, (15° to 18°S latitude) several LGM samples exhibit positive anomalies (LGM warmer than modern). These samples tend to bow the contours of the LGM map poleward in this region such that SSTs at 30° to 35°S show positive anomalies as high as 6°C. Since we have no evidence to suggest that LGM temperatures as far south as 30°S latitude were warmer than modern, control points based on modern SSTs have been used in this region. We have placed the 20°C isotherm at 30°S. We believe that the placement of these control points in the South Pacific results in maps which are more nearly correct than they would be using only actual LGM data points.

[20] Anomaly maps (Figures 4d, 5d, and 6d) were hand contoured to indicate regions where LGM cooling was greater than 1°C and 5°C. In the Indian and Pacific oceans, we have contoured areas where results indicate LGM SSTs greater than 1°C warmer than modern. Dashed lines indicate regions of uncertainty due to large gaps in data coverage.

Figure 4.

Annual mean SST for the Atlantic Ocean (a) modern based on gridded data, (b) modern contoured using only LGM sample locations, (c) reconstructed LGM, (d) anomaly map (LGM minus modern) where plus indicates LGM estimates are within 1°C of modern atlas values. Asterisks on Figures 4a–4c indicate samples which have no modern analogs. CLIMAP LGM sea ice boundaries were used to locate the 0°C isotherm in the South Atlantic.

Figure 5.

Annual mean SST for the Indian Ocean (a) modern based on gridded data, (b) modern contoured using only LGM sample locations, (c) reconstructed LGM, (d) anomaly map (LGM minus modern) where plus indicates LGM estimates are within 1°C of modern atlas values. CLIMAP LGM sea ice boundaries were used to locate the 0°C isotherm in the Southern Hemisphere.

Figure 6.

Annual mean SST for the Pacific Ocean (a) modern based on gridded data, (b) modern contoured using only LGM sample locations plus control points marked by “C”, (c) reconstructed LGM, (d) anomaly map (LGM minus modern) where plus indicates LGM estimates are within 1°C of modern atlas values. Asterisks on Figures 6a–6c indicate samples which have no modern analogs.

3.3. Maps

[21] An equidistant cylindrical projection is used for all maps. While this projection grossly distorts areas at high latitudes, it approximates Cartesian coordinates and thus allows for machine contouring on an x-y grid as we have done using Spyglass® data visualization software. Coastline boundaries at the LGM were somewhat different from modern coastlines due to the ∼120 m lowering of sea level, as large amounts of water were trapped in the huge ice sheets on land [Chappell and Shackleton, 1986; Shackleton and Matthews, 1977; CLIMAP Project Members, 1981]. While this sea level change resulted in significant coastline differences in Northern Australia, Indonesia and the Bering Sea, only small coastline changes occurred on most of the continents. We therefore have produced our LGM maps using modern coastlines for simplicity. True LGM coastlines in the critical region of the Indonesian Throughflow will be taken into account for interpretation of LGM results. Maps of mean annual SSTs, based on the average of monthly means, were produced to provide a first order look at the changes which took place during the LGM. These mean annual maps also facilitate comparison with estimates made using the Uk37 index. Temperature estimates made using this index appear to be closest to mean annual SST in tropics [Prahl and Wakeham, 1987; Rossell-Melé et al., 1995; Muller, 1992]. Mean annual SST maps are of particular value in the Indian Ocean where the seasonal SST peaks in May in the Arabian Sea, and reconstructions based on February/August would miss this critical period. February and August SST estimates were made as well to examine changes in seasonality and for comparison with CLIMAP Project Members [1981] estimates (Figure 7).

Figure 7.

Difference between LGM SST estimates (this study) and CLIMAP estimates for February and August. Plus indicates estimates are within 1°C of each other. Negative (circles) values indicate CLIMAP estimates are warmer than our LGM estimates. Positive (crosses) values indicate CLIMAP estimates are cooler than our LGM estimates.

4. Results

[22] We present our MAT reconstruction as a sequence of four maps for each ocean basin. The first map in each set (a) represents objectively contoured modern mean annual SST (based on monthly averages) gridded to 1° × 1° [Bottomley et al., 1990]. The second map (b) is modern mean SST objectively contoured using only LGM sample locations. These contours reflect the same sampling biases as the LGM maps, and therefore facilitate comparisons with the LGM maps. The third map (c) represents objectively contoured LGM SST estimates with the exception of the control points used in contouring (see methods section). The fourth map (d) presents the anomalies (LGM minus modern) with the symbol size scaled by magnitude. As mentioned previously, these maps have been hand contoured.

4.1. Atlantic

4.1.1. Tropical Atlantic

[23] At the LGM, warmest waters were restricted to west of 25°W longitude (Figure 4c); compared to modern where they penetrate eastward into the Gulf of Guinea (Figure 4b). Cooling in the central to eastern equatorial waters was 2 to 6°C, with greatest cooling centered on the equator at 15° to 25°W longitude in the region of modern equatorial upwelling (Figure 4d). This result is similar to results of Mix et al. [1999], who apply a new transfer function to foraminiferal data. Seasonal estimates reveal LGM seasonality was greater than modern in this region. Cold season estimates are 5° to 8°C cooler than modern while warm season estimates are within 3°C of modern. Our seasonal results agree with those of Prell et al. [1976] who used transfer functions to reconstruct equatorial Atlantic LGM SSTs. The central-eastern cooling resulted in a larger zonal temperature gradient in the tropics in both seasons than found in the modern. While our reconstruction finds tropical cooling is greatest here, high dissimilarity coefficients and several no-analog samples may indicate greater uncertainty in these LGM temperature estimates. Standard deviation of the SSTs for the closest modern analogs, however, is significantly lower than the magnitude of the estimated SST anomalies (LGM to modern); giving confidence to the reconstruction. An analysis of the fauna in these samples may suggest cold subsurface waters with relatively warm surface waters. This could indicate a shoaling of the thermocline, as has been suggested by Ravelo et al. [1990]. Highest mean annual SSTs (>28°C) in the modern tropical Atlantic are found in the Caribbean Sea. In the LGM reconstruction, waters in this region have SSTs less than 26°C (Figure 4c). Seasonal estimates indicate SSTs in the Caribbean cooled by 1.5°–4.5°C and 0.5°–2.5°C in February and August, respectively, with greatest cooling in the western Caribbean. This result is similar to the Mix et al. [1999] reconstruction. Our closest sample to Barbados (where Guilderson et al. [1994] have found a 5°C cooling based on coral records) indicates a LGM SST only 1.0° to 1.5°C lower than modern. For many Caribbean samples, the standard deviation of the SSTs for the closest modern analogs is less than the reconstructed anomalies. This result, again, gives confidence to the LGM estimates.

4.1.2. Boundary Currents

[24] Substantial cooling occurred off the coast of NW Africa where mean annual temperatures dipped to less than 16°C, compared to 21°C today (Figure 4c). This result is consistent with the alkenone results of Zhao et al. [1995] for this region, but somewhat less extreme than the cooling found by CLIMAP Project Members [1981]. The observed cooling may be due to increased upwelling which brings cool waters to the surface. Alternatively, Zhao et al. have hypothesized that cooling may be due to surges of cold water, resulting from a series of Heinrich events, which was then advected to the lower latitudes via the Canary Current. Our results indicate that the subpolar assemblage and cool waters intruded southward along the flow path of the Canary Current, however substantial LGM cooling is observed in the modern coastal upwelling region. These results suggest both increased flow in the Canary Current and increased coastal upwelling at the LGM.

[25] The Benguela Current plays a major role in interhemispheric heat transport and the flow of the global conveyor belt [Gordon, 1986]. The modern Benguela Current advects cool waters equatorward along the west coast of Africa (the Benguela Coastal Current) and northwest of the Walvis Ridge (the Benguela Oceanic Current). Our mean annual reconstruction reveals that the northwest deflection of the cool, Southern Ocean waters along the Benguela Oceanic Current was somewhat more intense at the LGM, while the Benguela coastal current was similar in strength to modern. (Figures 3b and 3c). Fairly warm waters persisted in the Gulf of Guinea and northern Angola Basin where SSTs are only 2° to 3°C lower than modern. Our Angola Basin and Benguela Coastal Current results are similar to those found by CLIMAP Project Members [1981] and a bit warmer than alkenone-based estimates of Schneider et al. [1995]. The north-south SST gradients in our reconstruction are similar to modern temperature gradients along the African coast of 8°–10°C from the equator to 30°S. These results are consistent with a surface circulation pattern similar to modern in the eastern South Atlantic but a more intense Benguela Oceanic Current.

4.1.3. Subtropics to Midlatitudes

[26] Largest changes in the subtropics to midlatitudes of the North Atlantic occur in the east where the warm waters of the North Atlantic Drift flow in a more zonal path at the LGM. Off the coast of Portugal, mean annual SSTs are ∼8°C lower at the LGM, compared to modern. LGM SSTs are lower along the northern edge of the modern North Atlantic subtropical gyre and slightly higher along the southeastern margin (Figure 4c). These changes suggest a compression or equatorward shift in the position of the gyre. In the South Atlantic gyre, greatest cooling also occurs along the gyre margins; with very slight cooling in the gyre center. Similar to the North Atlantic, this pattern suggests compression of the warm, subtropical gyre.

[27] The subtropical convergence zone (STC) is characterized by a relatively large change in SST with latitude (i.e. closely spaced isotherms). The mean position of the STC in the modern South Atlantic is about 40°S latitude and coincides with mean annual SSTs of 12° to 16°C. Using this SST range as a guide to the position of the STC, results reveal that at the LGM, its position was shifted equatorward by about 3° to 6° latitude with the largest shift in the east. This result is consistent with a compression and/or equatorward shift of the subtropical gyre.

4.1.4. High Latitudes

[28] In the North Atlantic, cold polar waters (<4°C) intruded as far south as the northern coast of Spain (Figure 4c), indicating a restricted, more zonal North Atlantic Drift; which currently advects warm waters into the Norwegian Sea. Greatest cooling was in the Northern Hemisphere midlatitudes (40° to 60°N) where some February anomalies are in excess of −10°C. These anomalies represent the southward movement of the Arctic polar front from its modern position of 60°–70°N to 42°–45°N. LGM SSTs in the Norwegian and Greenland Seas are >4°C in August, indicating at least seasonal thawing of the pack ice. Our calibration data base contains many samples with August SSTs <2°C; thus the relatively warm results are not from a lack of suitable modern analogs. Comparison of modern observations of percent sea ice cover and corresponding SSTs show that in the modern Norwegian and Greenland Seas, February SSTs of 0° to 1°C are found in regions of approximately 50% sea ice coverage and August SST of 4°–6°C are found in regions of approximately 20% sea ice coverage [Kellog, 1975; U.S. Naval Oceanographic Office, 1967; U.S. Navy Hydrographic Office, 1958]. Based on these observations, we assume sea ice coverage in the Norwegian and Greenland Sea was ≥50% in winter and ∼20% in summer. These results are consistent with recent studies [Sarnthein et al., 1998; Dokken and Hald, 1996; Hebbeln et al., 1994; Heinrich, 1988] which have found open ocean circulation during LGM summer in the Nordic Seas.

[29] LGM changes in the Southern Hemisphere high latitudes are not as easily reconstructed due to a lack of LGM samples. In particular, the LGM location of the Antarctic polar front (AAPF) is not clear from our samples. The few samples we have suggest only a slight northward movement of the AAPF. Since we have no basis to place the 0°C isotherm, we have used the CLIMAP Project Members [1981] sea ice boundaries to do so.

4.2. Indian Ocean

4.2.1. Tropics

[30] Mean annual SSTs in the tropical Indian Ocean at the LGM were quite similar to modern (Figures 5a–5c). SSTs > 28°C existed to a similar extent as in the modern ocean. The LGM August estimates, however, indicate that the region of high SSTs (>26°C) extended westward in the tropics to the African coast; in contrast to the modern temperature field where the South Equatorial Current brings cool waters northward into the Somali Current. This expansion of the warm waters westward is consistent with a weakened southwest monsoon. Our results indicate slightly warmer LGM SSTs than those found by Bard et al. [1997], who use the Uk37 index. Central and eastern equatorial SSTs were slightly cooler during August, and similar to modern during February. Small annual mean anomalies (<1°C) in the low-latitude eastern Indian Ocean and Indonesian region tend to confirm the CLIMAP estimate of little cooling in the tropics at the LGM. Ohkouchi et al. [1994] have similarly found small (1°C) LGM SST anomalies in the western Pacific warm pool using the Uk37 index. These results are inconsistent with projected SSTs estimated by Webster and Streton [1979] who used LGM snow line distributions in Australia and Indonesia to deduce that tropical SSTs in this region cooled by 5°–7°C.

4.2.2. Northern Indian Ocean

[31] As discussed previously, mean annual SST maps are of particular use in the Indian Ocean where Arabian Sea temperatures peak in May. While changes at the LGM are relatively small, the mean annual reconstruction indicates slight warming along the coast of Oman at the LGM and areas of local cooling west and south of India (Figure 5d). Largest changes occur in August when the Gulf of Aden of cooled by 2°–5°C. During August results indicate a decreased east-west SST gradient along the Oman Margin and higher than modern SSTs in the Central Arabian Sea. Higher-than-modern SSTs occur along the Oman Margin in February. The above results suggest that the overall summer climate in the region was similar to modern but that the strength of the southwest monsoon was somewhat reduced and thus resulted in a more subtropical climate in the northern and western Arabian Sea. A recent study in the northern Indian Ocean [Cayre et al., 1999], however, has found foraminiferal variability in this region to be related to variations primary productivity, rather than SST. This study found two dominant assemblages: one characterizing upwelling regions, and one characterizing nonupwelling regions. To the extent changes in SST in this region are related to changes in upwelling intensity, our results are consistent with glacial/interglacial variations in monsoon intensity, upwelling and consequent primary productivity changes. Our results indicate warmer SSTs than those found by Sonzogni et al. [1998] who use Uk37 to estimate LGM SSTs 1°–3°C cooler than modern.

[32] Mean annual SSTs in the Bay of Bengal were within 1°C of modern (Figure 5d); however, seasonal estimates reveal February SSTs 1°–3°C warmer than modern in the northern Bay of Bengal. Positive anomalies increase toward the north. Previous studies indicate that salinity was substantially higher in the northern Bay of Bengal at the LGM and that the north-south salinity gradient in this region was greatly reduced as a result of reduced river runoff from the Ganges-Brahmaputra River system [Cullen, 1981]. The distribution of foraminiferal species reflects these changes [Cullen, 1981; Cullen and Prell, 1978]. Our results, while possibly indicative of actual LGM warming in the northern Bay of Bengal, may also reflect these reduced latitudinal gradients in the surface waters at the LGM.

4.2.3. Boundary Currents

[33] Mean annual temperatures along the west coast of Australia were 2° to 3°C lower at the LGM. LGM samples in this region, however, have relatively large standard deviation values; possibly indicating lower reliability of these estimates. More significant cooling occurred during the winter (August) with the largest anomalies occurring off Port Hedland, about 17°S latitude. This result indicates that the West Australia Current (WAC), which transports cool, southern waters equatorward, penetrated farther north at the LGM than it does currently. A recent study of 10 cores along the west coast of Australia applied the FI-2 transfer function [Hutson and Prell, 1980] to planktonic foraminfera for reconstruction of SST gradients in this region at the LGM [Wells and Wells, 1994]. These results showed the greatest cooling off North West Cape (22°S–24°S) and suggested greater penetration of the WAC northward at the LGM. We find greatest cooling ∼17°S. Our results support the finding of Wells and Wells that the southward-flowing Leeuwin Current did not warm the coast at the LGM as it does today and that this allowed for equatorward penetration of the WAC.

[34] Our mean annual reconstruction reveals SSTs were higher than modern northwest of Madagascar and cooler than modern to the south, in the region of the Agulhas Current. Since the poleward flowing Agulhas Current transports warm waters to higher latitudes, lower SSTs at the LGM suggest reduced poleward flow at the higher latitudes. Our results suggest that the warm tropical waters from this current may have been recycled into the gyre at lower latitudes than in the modern. The net effect of reduced flow intensity in the Agulhas Current is to increase the north-south SST gradient along the east coast of Africa from ∼7°C in the modern to ∼10°C at the LGM.

4.2.4. Subtropical Convergence and Antarctic Polar Front

[35] The mean position of the subtropical convergence in the modern Indian Ocean is about 40°S latitude and may be identified by the close spacing of the 12°–16°C isotherms. This boundary is most intense in the west where the strong, poleward Agulhas Current meets the eastward-flowing waters of the West Wind Drift. While LGM data coverage is sparse in this region, it appears that the mean position of the subtropical convergence (STC) is shifted equatorward by about 5° latitude in the east and somewhat less in the west. This shift is accompanied by a slight compression of the isotherms, which is most evident in the central and eastern portion of the southern Indian Ocean. While we have few LGM samples in the region of the Antarctic Polar Front (AAPF), we can use the location information from the top analogs of these samples to infer movements of water masses. For LGM samples located south of 39°S latitude, most modern analogs are located 3° to 12° latitude poleward of the location of the subject samples. This result implies that the AAPF, currently located at about 50°S latitude, moved equatorward by at least several degrees of latitude. Samples in the central region have analogs located more poleward than those in the eastern region indicating that the AAPF moved more equatorward in the central and western regions. Our results are consistent with previous results which found that the STC moved about 4° latitude equatorward [Howard and Prell, 1992] and that the AAPF moved 5° to 10° latitude equatorward at the LGM [Prell et al., 1980; Howard and Prell, 1992].

4.3. Pacific

[36] We have included in the Pacific LGM maps SST estimates from two sources outside of our BLGM database. For the LGM mean annual SST maps, we have used two alkenone-based SST estimates off the coast of southern Baja California to guide the placement of the isotherms in this region where we are lacking foraminiferal data. We have also controlled the placement of isotherms in the central south Pacific by specifying the isotherms at 30°S latitude to be similar to modern as discussed in the methods section. Nonforaminferal control points are indicated on LGM maps by a “C” (Figure 6c). Due to data limitations, we have not reconstructed the higher latitudes of the Pacific. Since the tropics are the regions most in dispute, we have focused our efforts here.

4.3.1. Tropics

[37] The modern equatorial Pacific is characterized by pronounced equatorial divergence in the east which extends westward to ∼170°W longitude (Figure 6a). This divergence is captured less well when the modern SST field is contoured using only the LGM sample locations (Figure 6b). Equatorial divergence is evident in Figure 6b to about 135°W longitude. Coastal upwelling off the Peruvian Coast results in quite low eastern equatorial SSTs. Inspection of the mean annual modern and LGM SST maps (Figures 6a–6c) reveal some important changes at the LGM. Considerable cooling occurred in the eastern equatorial Pacific off Ecuador and Peru, in the region of modern equatorial and coastal upwelling. This upwelling is most pronounced in August when temperatures are as much as 10°C lower than modern. Cooler-than-modern SSTs extend west to 105°–120°W longitude. The spatial distribution of our LGM data set results in contours which do not capture the familiar equatorial divergence patterns well. Results do indicate, however, that surface cooling associated with increased divergence was limited to east of 120°W. Two samples (RC13-113, 1°39′S, 103°38′W and V21-033, 3°48′S, 92°5′W) have SST standard deviation values of similar magnitude to the anomalies and thus we are less certain of these estimates. Several equatorial samples located west of 125°W longitude have LGM SST estimates as warm as modern temperatures, however, these samples are highly dissolved and estimates may be biased toward warmer SST. Results indicate that while upwelling was stronger in the east, it may not have extended as far west at the LGM as it does in the modern equatorial Pacific. While our eastern equatorial SST estimates are similar to those of Mix et al. [1999], they found cooler than modern SSTs extended to 130°W.

[38] Highest SSTs at the LGM were similar to modern in much of the western tropical warm pool, however, the extent of this warm pool was more restricted than modern. The 28°C isotherm was shifted significantly equatorward in both hemispheres; with greatest latitudinal shifts occurring around New Guinea and the Philippines. Mean annual SSTs in the Indonesian Seas were less than 28°C whereas modern mean SSTs in this region are all above 28°C. Average LGM cooling in this region is 1°–2°C. This result is consistent with alkenone-based estimates [Ohkouchi et al., 1994]. Our results are inconsistent with alpine snow line and pollen data from the New Guinea highlands which indicate a LGM cooling of 5°–6°C [Webster and Streton, 1979]. Both February and August SST estimates indicate an eastward extent of the warmest waters (>28°C) that is similar to the modern. Our results indicate that the east-west SST gradient was stronger at the LGM, compared to modern and thus suggest that Walker circulation was strong; with more La Nina-like conditions.

4.3.2. Boundary Currents

[39] While many of the western tropical samples indicate slightly lower LGM SSTs, one subtropical sample southwest of Okinawa has an estimated mean annual SST >2°C higher than modern. Two samples to the northeast have a 1°–3°C cooling suggesting a sharp SST gradient from ∼26°–29°N latitude, south of Japan.

[40] Waters of the California Current off the coast of Oregon were significantly cooler at the LGM. SSTs were as much as 7°C lower than modern with greatest cooling in February. The modern California Current has strongest southward flow and is situated closest to the coast in late summer [Prahl et al., 1995]. This results in a southward deflection of the isotherms along the Oregon and California coasts. Our LGM reconstruction indicates that waters cooled by only 2° to 4°C near the coast in August, but cooling increased offshore such that samples located at 132°W have August SSTs 6°C lower than modern. The resulting temperature gradients in this region suggest that coastal upwelling may have been reduced at the LGM, but that equatorward advection of cool waters in the California Current was increased. Our results indicate greater cooling than those of Ortiz et al. [1997] who used the same samples and similar techniques but different calibration data. Their LGM SST estimates average 2.5°C warmer than ours, but more importantly, they found that the offshore thermal gradient at the LGM was similar to modern, whereas our results suggest a much smaller gradient during August.

[41] Larger SST gradients existed off the eastern coast of Australia at the LGM, compared to modern. While several samples south of Papua New Guinea and the Solomon Islands remained warm, those south of New Caledonia cooled by 2° to 4°C. This increased gradient could have resulted from a weakened, southward flowing East Australia Current, accompanied by a relatively strong South Equatorial Current (SEC). This combination would result in warm waters in the western tropics which were not efficiently transported to higher latitudes. Similar to the Agulhas Current of the Indian Ocean, the warm waters in the East Australia Current may have been recycled back into the gyre at lower latitudes than in the modern. Our results are consistent with those of Anderson et al. [1989] and Thunell et al. [1994]. Both studies applied the MAT to high sedimentation rate samples from the Coral Sea. Many samples from the former study are from the same cores used in our study. Mean annual SSTs just south of the tropics were 3°–4°C lower at the LGM while those in the Coral Sea were similar to modern. Anderson, et al. concluded that cool waters flowed equatorward in the Tasman Sea at the LGM and that this resulted in lower than modern SSTs to about 25°S. Our results are consistent with this interpretation.

4.3.3. High Latitudes

[42] High latitude changes in the Pacific are difficult to evaluate due to poor carbonate preservation. Results from the North Pacific indicate greater SST changes in the northeast (4°–8°C lower than modern) and smaller changes in the northwest (2°–5°C lower than modern). This pattern of increased cooling toward the east is similar to that found in the North Atlantic. Similar to changes in the North Atlantic Drift, we hypothesize that the North Pacific Current had a more zonal flow path at the LGM than modern.

5. Discussion

[43] Many LGM changes are similar across ocean basins. All three oceans experienced lower SSTs in the mid to high latitudes at the LGM. Most dramatic cooling took place in the North Atlantic (40°–60°N). All three oceans also experienced increased equatorward advection of cool waters in the eastern boundary currents, with greatest cooling during the winter season. Much of the low-latitude Indian and Pacific Oceans remained within 1°C of modern SSTs at the LGM, while substantial cooling occurred in the tropical Atlantic. Mean annual SSTs were 2°–6°C lower in the central to eastern equatorial Atlantic and 3°–8°C lower in the eastern equatorial Pacific. These results may indicate increased equatorial divergence as a result of increased trade wind intensity in both these oceans. Hostetler and Mix [1999] found increased LGM trade winds using an atmospheric GCM with similar SST boundary conditions.

[44] An important feature of our LGM maps is the greater equatorial cooling in the Atlantic than the Pacific and Indian Oceans. This result is consistent with slowing of the conveyor belt at LGM, which would slow down deep water formation and thereby reduce heat transport into the Atlantic. Since some of this heat comes from the Pacific via the Indonesian Throughflow, slowing of heat transport into the Atlantic would cause warm water to accumulate in Indo-Pacific. This would result in a temperature gradient between basins. A study of large scale transport from the Pacific to Indian Ocean [Poterma et al., 1997] found that, while net flow is generally westward, large seasonal fluctuations exist in transport direction in the Indonesian Throughflow. Flow is westward and southward in most straits from April to July and eastward and northward in most straits during January and February. Flow through the straits is driven by both pressure gradients and wind-driven transport. To the extent the southwest monsoon was weaker at the LGM than modern [Prell and Curry, 1981; Prell, 1984; Clemens and Oglesby, 1990; Anderson and Prell, 1992], net westward flow may have decreased during this time and thus helped maintain the warmth of the western warm pool.

[45] Our results of relatively warm SSTs in the tropics are inconsistent with results from coupled climate/mixed-layer ocean models with calculated SSTs [Manabe and Broccoli, 1985], however, these models do not account for ocean currents and horizontal heat transport by the ocean [Broccoli and Marciniak, 1996]. The models also assume that the depth of the mixed-layer of the ocean is globally uniform. Our results suggest that changes in ocean heat transport between the modern and LGM may have been instrumental in maintaining the reconstructed SST field at the LGM. Variations in mixed-layer depth also affect the thermal inertia of the upper ocean and accurate parameterization of this variable is important for coupled climate-ocean models.

5.1. Comparisons to CLIMAP

[46] While many of our LGM SST estimates are quite similar to those of CLIMAP (Figure 7, marked by pluses), important differences do exist. In the equatorial region of the Atlantic and Pacific, results indicate a larger east-west temperature gradient than that found by CLIMAP. Gradient differences are largest during August where our eastern Pacific SSTs are 5°–14°C lower than those found by CLIMAP and our western Pacific SSTs are 0° to 2°C higher. Increased zonal SST gradients have implications for intensity of Walker circulation at the LGM.

[47] Many of our SST estimates are higher than those found by CLIMAP. Higher estimates occur in the western tropics of all three oceans, in the Benguela and Northwest African upwelling regions, in the West Australia and Agulhas Currents, and in the Norwegian Sea. As discussed previously, our results suggest seasonal thawing of pack ice in the Norwegian Sea at the LGM, while CLIMAP placed the Atlantic sea ice boundary south of Iceland in both seasons. While we have not drawn in sea ice boundaries on our maps, we assume sea ice is present where SST is lower than 2°C. In the West Australia Current, CLIMAP found a cooling of >4°C in both seasons whereas our results only indicate 1° to 3°C cooling in this region with SSTs cooling more in August than February, relative to modern.

[48] Our LGM ocean reconstructions share many of the same large-scale features as those found by CLIMAP. Similar to CLIMAP Project Members [1981], we see increased equatorward extent of cool water in the eastern boundary currents. Our results do not indicate as extensive cooling in these eastern boundary currents as that found by CLIMAP. Our data reveal a similar westward extension of cool tongues in the equatorial Atlantic and Pacific, though our SST gradients are higher. The cooling in the eastern equatorial Pacific appears to be due to increased coastal upwelling or increased equatorial upwelling only in the east since cooling is concentrated close to the coast and the cool tongue does not extend as far west as modern.

[49] Again similar to CLIMAP, we find little cooling in large areas of the tropics and subtropics at the LGM. In fact, our results indicate even warmer SSTs in many tropical regions than those found by CLIMAP. This is particularly true in the western tropical Pacific and most of the Indian Ocean. Our results for the western Pacific warm pool are consistent with those of Thunell et al. [1994], who used the MAT on a suite of well-preserved, relatively high sedimentation rate cores in the western Pacific. Similar to CLIMAP Project Members [1981], they found cooling of less than 2°C from modern in the region of the Pacific warm pool. This result of minimal cooling in the western tropical Pacific is significant for tropical climate at the LGM, as the warm waters of the Pacific warm pool transfer a significant amount of heat and moisture to the atmosphere. Our reconstruction suggests that the low-pressure cell, currently centered over Indonesia, was nearly as strong at the LGM as modern. This low-pressure cell appears to be longitudinally restricted during February, however since SSTs higher than 28°C are necessary to support strong convection in the atmosphere. The ocean-atmosphere coupling associated with the warm pool and low-pressure cell would have modulated the LGM climate as it does today. Overall, our results suggest larger SST gradients, both within and between ocean basins, than found by CLIMAP. These increased gradients could have a potentially large effect on heat and moisture transfer to the atmosphere.

5.2. Comparisons to TEMPUS Results

[50] The TEMPUS project is an effort to revise estimates of SST at various critical points in Earth's past using the Uk37 index. This project has produced an anomaly map of modern versus LGM SSTs based using the Uk37 index [Rosell-Melé et al., 1998]. We find broad, overall agreement between the TEMPUS results and our results in much of the Atlantic and Pacific Oceans. One notable point of disagreement in the Atlantic is the Gulf of Mexico where we find 3°–5° cooling while one alkenone-based estimate indicates 7° cooling [Jasper and Gagosian, 1989]. Additionally, where we have found substantially cooler-than-modern LGM temperatures points in the eastern equatorial Pacific, one TEMPUS datum indicates no LGM cooling [Emeis et al., 1995]. True LGM values may not be represented in this record, however, due to fairly low-resolution sampling.

[51] The most substantial differences between our result and the TEMPUS results occur in the northern Indian Ocean. MAT results indicate most of this region was within 1°C of modern at the LGM, while alkenone results indicate negative anomalies of 1° to 5°C.

[52] Large disagreements exist along the Oman Margin, along the west coast of India and in the Bay of Bengal. Foraminiferal faunal changes in this region are minor and generally indicate a more tropical climate in the northern Arabian Sea and Bay of Bengal, than modern. In the seasonally dynamic Arabian Sea, it is critical to understand which season is recorded by each faunal group. Lowest SSTs in the Northern Arabian Sea occur during February while the Oman Margin and Somalian coast experience lowest SSTs during May. The monsoon climate results in enormous seasonal productivity changes in these regions, as well, with the primary productivity peaking in the summer (cool) season. Thus differences in season recorded by foraminifera versus coccolithophores (the source of alkenones) could provide some explanation for the different SST estimates.

6. Summary and Conclusions

[53] LGM maps reveal many important patterns and gradients which affect and reflect the climate and circulation patterns of the LGM. One of the most important findings is the marked cooling in the central to eastern tropical Atlantic and eastern tropical Pacific. These changes resulted in an increased east-west SST gradient in the equatorial regions of these two basins. The increased gradients both reflect and provide a positive feedback for increased zonal wind intensity at the LGM.

[54] SSTs in eastern boundary currents of all three ocean basins cooled at the LGM. These changes suggest increased equatorward flow of cool waters. Additionally, we see more zonal flow patterns in the Kuroshio/North Pacific Current and Gulf Stream/North Atlantic Drift. These changes result in more intense midlatitude to equator SST gradients at the LGM, which would also have contributed to more intense circulation and increased zonal winds.

[55] Consistent with the results of previous studies, our results indicate little change in much of the western tropics and the subtropics of all ocean basins. Analysis of the modern and LGM tropical assemblages reveals very little change occurred in these regions. Thus we have no evidence to suggest LGM SSTs in the western tropics and subtropics were substantially different from modern values.


[56] We would like to thank A. Martin and J. Donnelley for their work on the Brown foraminiferal database. Discussions with T. Herbert and T. Webb added to the work presented here. We also thank James Cullen and two anonymous reviewers for helpful comments that improved the manuscript. This research was supported by NSF grant ATM-9709769 to Warren Prell.