Comparison of Imbrie&Kipp Transfer Function and Modern ...

Comparison of Imbrie-Kipp transfer function and modern

analog temperature estimates using sediment trap and core

top foraminiferal faunas

J. D. Ortiz' and A. C. Mix

College of Oceanic and Atmospheric Sciences, Oregon State University, Corvallis

Abstract. We evaluate the reliability of statistical estimates of sea surface temperature (SST) derivedfrom planktonic foraminiferal faunas using the modern analog method and the Imbrie-Kipp method.Global core top faunas provide a calibration data set, while modern sediment trap faunas are used forvalidation. Linear regression of core top predicted SST against atlas SST generated slopes close to onefor both methods. However, the Jmbrie-Kipp transfer function temperature estimates had an intercept13°C warmer than modern analog estimates and 1.7°C warmer than recorded atlas SST. The RMS errorfor the core top data set using the modern analog method (1.5°C) was smaller than that of the Jmbrie-Kipp method (1.9°C). SST errors for the sediment trap faunas were not statistically different from thoseof the core top data set, regardless of method. Developing Imbrie-Kipp transfer functions for limitedregions reduced the RMS variability but introduced residual structure not present in the global Imbrie-Kipp transfer function. Dissolution simulations with the sediment trap sample which generated thewarmest SST residual for both methods suggests that the loss of delicate warm water foraminifera frommidlatitude sediments may be the cause of this thermal error. We conclude that (1) the faunal structure ofsediment trap and core top assemblages are similar; (2) both methods estimate SST reliably for modernforaminiferal flux assemblages, but the modern analog method exhibits less bias; and (3) both methodsare relatively robust to samples with low communality but sensitive to selective faunal dissolution.

PALEOCEANOGRAPHY, VOL. 12, NO.2, PAGES 175-190, APRIL 1997

Introduction

Because sea surface temperature (SST) provides an importantclimate diagnostic on a variety of spatial and temporal scales,we assess the reliability of paleotemperature reconstructionsderived from planktonic foraminiferal faunas. The twotechniques most commonly used to estimate paleotemperaturefrom these microfossils are the transfer function method ofImbrie awl Kipp [1971] and the modern analog method ofHuison [1980] as modified by Prell [1985]. Do these twomethods provide consistent, unbiased paleotemperatureestimates?

A potential problem is most evident in the temperaturedilemma of the low-latitude last glacial maximum (LGM) oceanwhere microfossil paleotemperature reconstructions differ fromother methods by as much as 2°-3°C and are inconsistent withsome climate model results [Rind and Peteet, 1985]. Statisticalreconstructions based on microfossils [e.g., CLIMAP, 1976,1981] suggest that low - latitude SST was at most 2°-3°C cooler

'Now at Lamosn-Doherty Earth Observatosy of Columbia University,Palisades, New York.

Copyright 1997 by the American Geophysical Union.

Paper number 96PA02878.O883-83O5/976PA-O2878$l2.00

175

than modern. This 2°-3°C cooling was indicated by severaltypes of microfossils [Molfino et al. 1982].

Broecker [1986] compared these findings with planktonicforaminiferal 6'O and concluded the 6'°O data was alsoreasonably consistent with a 2°-3°C cooling. Stott and Tang[1996] reached a similar conclusion in their study of tropical8180 measurements. The recent study of Broccoli and Marciniak[1996] suggests that point by point comparisons of climatemodel output and observations reduces some of the low-latitudeSST discrepancy. Unfortunately, they could not make a cleardetermination as to the validity of one approach over the other.

In contrast, results derived from coral SrjCa paleo-thermometry [e.g., Beck et al. 1992; Guilderson et al. 1994]and a variety of terrestrial based methods [e.g., Rind and Peteet,1985] suggest that low-latitude glacial SST may have been asmuch as 5°-6°C cooler than modem. Because of this apparentdiscrepancy, we reevaluate the reliability of statistically basedmicrofossil SST estimates derived from planktonicforaminiferal faunas using the transfer function method ofImbrie and Kipp [1971] and the modern analog method ofHutson [1980] as modified by Prell [1985]. Our analysis differsfrom previous sediment based studies in that we estimate SSTfor sediment trap assemblages caught in the water column aswell as for core tops. This test is designed to assess whethersmoothing and modification involved with the generation ofthe fossil record induces bias in foraminiferal faunal SSTestimates.

PALEOCEANOGRAPHY, VOL. 12, NO. 2, PAGES 175-190, APRIL 1997

Comparison of Imbrie-Kipp transfer function and modern

analog temperature estimates using sediment trap and core

top foraminiferal faunas

J. D. Ortiz • and A. C. Mix

College of Oceanic and Atmospheric Sciences, Oregon State University, Corvallis

Abstract. We evaluate the reliability of statistical estimates of sea surface temperature (SST) derived from planktonic foraminiferal faunas using the modem analog method and the Imbrie-Kipp method. Global core top faunas provide a calibration data set, while modem sediment trap faunas are used for validation. Linear regression of core top predicted SST against atlas SST generated slopes close to one for both methods. However, the Imbrie-Kipp transfer function temperature estimates had an intercept 1.3øC warmer than modem analog estimates and 1.7øC warmer than recorded atlas SST. The RMS error for the core top data set using the modem analog method (1.5øC) was smaller than that of the Imbrie- Kipp method (1.9øC). SST errors for the sediment trap faunas were not statistically different from those of the core top data set, regardless of method. Developing Imbrie-Kipp transfer functions for limited regions reduced the RMS variability but introduced residual structure not present in the global Imbrie- Kipp transfer function. Dissolution simulations with the sediment trap sample which generated the warmest SST residual for both methods suggests that the loss of delicate warm water foraminifera from midlatitude sediments may be the cause of this thermal error. We conclude that (1) the fauna] structure of sediment trap and core top assemblages are similar; (2) both methods estimate SST reliably for modem foraminifera] flux assemblages, but the modem analog method exhibits less bias; and (3) both methods are relatively robust to samples with low communality but sensitive to selective faunal dissolution.

Introduction

Because sea surface temperature (SST) provides an important climate diagnostic on a variety of spatial and temporal scales, we assess the reliability of paleotemperature reconstructions derived from planktonic foraminifera] faunas. The two techniques most commonly used to estimate paleotemperature from these microfossils are the transfer function method of

Imbrie and Kipp [1971] and the modern analog method of Hutson [1980] as modified by Prell [1985]. Do these two methods provide consistent, unbiased paleotemperature estimates?

A potential problem is most evident in the temperature dilemma of the low-latitude last glacial maximum (LGM) ocean where microfossil paleotemperature reconstructions differ from other methods by as much as 2ø-3øC and are inconsistent with some climate model results [Rind and Peteet, 1985]. Statistical reconstructions based on microfossils [e.g., CLIMAP, 1976, 1981] suggest that low - latitude SST was at most 2ø-3øC cooler

1Now at Lamont-Doherty Earth Observatory of Columbia University, Palisades, New York.

Copyright 1997 by the American Geophysical Union.

Paper number 96PA02878. 0883-8305/97/96PA-02878512.00

than modem. This 2ø-3øC cooling was indicated by several types of microfossils [Molfino et al. 1982].

Broecker [1986] compared these findings with planktonic foraminiferal 8180 and concluded the 8180 data was also

reasonably consistent with a 2ø-3øC cooling. Stott and Tang [1996] reached a similar conclusion in their study of tropical 8180 measurements. The recent study of Broccoli and Marciniak [1996] suggests that point by point comparisons of climate model output and observations reduces some of the low-latitude SST discrepancy. Unfortunately, they could not make a clear determination as to the validity of one approach over the other.

In contrast, results derived from coral St/Ca paleo- thermometry [e.g., Beck et al. 1992; Guilderson et al. 1994] and a variety of terrestrial based methods [e.g., Rind and Peteet, 1985] suggest that low-latitude glacial SST may have been as much as 5ø-6øC cooler than modern. Because of this apparent discrepancy, we reevaluate the reliability of statistically based microfossil SST estimates derived from planktonic foraminiferal faunas using the transfer function method of lmbrie and Kipp [1971] and the modern analog method of Hutson [1980] as modified by Prell [1985]. Our analysis differs from previous sediment based studies in that we estimate SST for sediment trap assemblages caught in the water column as well as for core tops. This test is designed to assess whether smoothing and modification involved with the generation of the fossil record induces bias in foraminiferal faunal SST estimates.

175

176 ORTIZ AND MIX: SEDIMENT TRAP-CORE TOP COMPARISON

Methods

Core Top and Sediment Trap Data Sets

Prell [1985] tested the modem analog method of Hutson[1980] and demonstrated that with slight modification, it yieldsessentially identical SST results to the transfer functionapproach of Imbrie and Kipp [1971]. These tests wereconducted using core top foraminiferal faunas from the threemajor ocean basins. Unfortunately, validation ofpaleotemperature proxy methods based on core top data alonehas several potential sources of bias. These include (1)stratigraphic sampling errors, (2) dissolution errors, (3)bioturbation which mixes together foraminiferal shells fromdifferent times, and (4) the possibility that the statisticalcorrelation to SST observed in the core top faunas is notpresent in the living faunas but rather is a secondary artifact,caused, for example, by the correlation of SST with otherenvironmental variables.

The first three problems may be expressed as either randomor systematic errors in the SST estimates derived from eithermethod. We assess random error by calculating the RMS errorfor each method. We assess systematic bias in each method bydetermining the slope and intercept of actual versus estimatedSST regressions and by testing residuals for statisticallysignificant trends. To address the fourth problem, weassembled a validation data set of foraminiferal faunas from 13sediment trap locations (Figure 1) including our three sites at42°N in the California Current and a previously unpublisheddata set from a site in the equatorial Pacific, Manganese NoduleProject (MANOP) Site C.

We compare the sediment trap validation data set with acalibration data set of 1121 core tops (Figure 2) composed of

o 30 60 90 120 150 180 150 120 90 60 30 090 90

60

30

w

0

30

60

those from PrelI [1985] and selected samples from Parker andBerger [1971], Coulbourn Ct al. [1980], and Thompson [1981].Samples in these data sets were screened to exclude duplicatesand samples with erroneous locations (A. E. Morey and A. C.Mix, personal communication, 1996). To account fordifferences in the taxonomy used by various workers, weemployed a subset of 27 taxonomic species and lumped sometaxonomically similar species into morphologic groups. Thefour morphologic groups we employed are (1) G. ruber (total),which includes both the pink and white varieties; (2) G.sacculifer (total), which includes G. trilobus and G. sacculifer;(3) G. menardii (total), which includes G. menardii, G. menardiiflexuosa (=neoflexuosa), and G. tumida; and (4) N. dutertrei(which includes the "Neogloboquadrina pachyderma -

Neogloboquadrina dutertrei (P-D) intergrade" category of Kipp[1976]). We believe the P-D intergrade category to beconspecific with N. dutertrei based on our plankton tow andsediment trap studies [Ortiz and Mix, 1992; Ortiz et al. 1995].

The sediment trap data are of known modem age andessentially free of dissolution. Table 1 provides informationregarding the locations of the sediment trap faunas and theirsources, while Table 2 lists the flux-weighted foraminiferalfaunas from these locations. The sediment trap samples havethe disadvantage of short integration times, ranging fromseveral months to 6 years. Poorly resolved interannualvariation may affect the results. To assess this effect, themodem analog method and Q-mode factor analysis providequantitative estimates of how different the sediment trap faunasare from the sedimentary faunas.

Unlike the core top data set, the sediment trap faunal countsare not all based on the >150 im size fraction. Ideally, thesediment trap samples should be processed in a manner which is

60

30

0

30

60

90 I I I I I I I I I I 900 30 60 90 120 150 180 150 120 90 60 30 0

LONG 1U DE

FIgure 1. Locations of the 13 sediment trap faunas used in the paper. Numbers next to each trap locationcorrespond to map indices in Table 1. References for each published sample are listed in Table 1.


Methods

Core Top and Sediment Trap Data Sets

Prell [1985] tested the modem analog method of Hutson [1980] and demonstrated that with slight modification, it yields essentially identical SST results to the transfer function approach of lmbrie and Kipp [1971]. These tests were conducted using core top foraminiferal faunas from the three major ocean basins. Unfortunately, validation of palcotemperature proxy methods based on core top data alone has several potential sources of bias. These include (1) stratigraphic sampling errors, (2) dissolution errors, (3) bioturbation which mixes together forarniniferal shells from different times, and (4) the possibility that the statistical correlation to SST observed in the core top faunas is not present in the living faunas but rather is a secondary artifact, caused, for example, by the correlation of SST with other environmental variables.

The first three problems may be expressed as either random or systematic errors in the SST estimates derived from either method. We assess random error by calculating the RMS error for each method. We assess systematic bias in each method by determining the slope and intercept of actual versus estimated SST regressions and by testing residuals for statistically significant trends. To address the fourth problem, we assembled a validation data set of foraminiferal faunas from 13

sediment trap locations (Figure 1) including our three sites at 42øN in the California Current and a previously unpublished data set from a site in the equatorial Pacific, Manganese Nodule Project (MANOP) Site C.

We compare the sediment trap validation data set with a calibration data set of 1121 core tops (Figure 2) composed of

those from Prell [1985] and selected samples from Parker and Berger [1971], Coulbourn et al. [1980], and Thompson [1981]. Samples in these data sets were screened to exclude duplicates and samples with erroneous locations (A. E. Morey and A. C. Mix, personal communication, 1996). To account for differences in the taxonomy used by various workers, we employed a subset of 27 taxonomic species and lumped some taxonomically similar species into morphologic groups. The four morphologic groups we employed are (1) G. ruber (total), which includes both the pink arid white varieties; (2) G. sacculifer (total), which includes G. trilobus and G. sacculifer; (3) G. menardii (total), which includes G. menardii, G. menardii fiexuosa (=neofiexuosa), and G. turnida; and (4) N. dutertrei (which includes the "Neogloboquadrina pachyderma - Neogloboquadrina dutertrei (P-D)intergrade" category of Kipp [1976]). We believe the P-D intergrade category to be conspecific with N. dutertrei based on our plankton tow and sediment trap studies [Ortiz and Mix, 1992; Ortiz et al. 1995].

The sediment trap data are of known modem age and essentially free of dissolution. Table 1 provides information regarding the locations of the sediment trap faunas and their sources, while Table 2 lists the flux-weighted foraminiferal faunas from these locations. The sediment trap samples have the disadvantage of short integration times, ranging from several months to 6 years. Poorly resolved interannual variation may affect the results. To assess this effect, the modem analog method and Q-mode factor analysis provide quantitative estimates of how different the sediment trap faunas are from the sedimentary faunas.

Unlike the core top data set, the sediment trap faunal counts are not all based on the >150 g.rn size fraction. Ideally, the sediment trap samples should be processed in a manner which is

0 30 60 90 120 150 180 150 120 90 60 30 0 9O 9O

60 6O +

4 + 30 30 .7 5 ,",• 8

i,i + + C3 10 :::) 0 + 9' 0

30 30

'11 12 13,

60 • i• 60 90 I I I I I I I I 90 0 50 60 90 120 150 180 150 120 90 60 30 0

LONGITUDE

Figure 1. Locations of the 13 sediment trap faunas used in the paper. Numbers next to each trap location correspond to map indices in Table 1. References for each published sample are listed in Table 1.

60

30

waD 0

30

60

900 30 60 90 120 150 180 150 120 90 60 30

LONGITUDE

FIgure 2. Locations for the 1121 core top samples in the global core top data set.

identical to that of the core top sediments. In our studies wehave measured both the 125-150 and >150 p.m size fractions.This allows us to easily compare our samples to >150 pmsediments or >125 pm sediment trap samples. We recommendthat future sediment trap and plankton tow studies follow thisprocedure or focus on the >150 pm size fraction. Whilevariations in sieve size introduce some uncertainty in thesediment trap to core top comparison, the residual errors weobtained are not correlated with sieve size variations. Any biasintroduced by this source of error is thus not systematic. Thesepotential sources of error will be evaluated further in thediscussion to follow.

Q-Mode Factor Analysis and Imbrle-Kipp TransferFunctions

As a first step in the comparison of core top and sedimenttrap faunas, we calculated a Q-mode factor model for the globalcore top database following Imbrie and Kipp [1971]. Q-modefactor analysis provides an objective, quantitative means ofsimplifying complex data sets. This is accomplished bydecomposing the core top data matrix (Uct) into a factor scorematrix (Fat). a factor loading matrix (Bct), and an error matrix(Eci:

Uct = Bt Fct Ect (1)

the core top percent abundance data, is based on counts

of n=27 planktonic foraminiferal taxa in N=1 121 coretops.Each core top fauna in Uct is row normalized so that its sum ofsquares is unity. The elements of the factor score matrix, Fct,describe m varimax rotated assemblages composed of weighted

contributions from each of the 27 taxa (see Table 3). A varimaxrotation is applied to Fct so that the resulting assemblagesremain close to the centroid of the sample data. This rotationhas the advantage of producing orthogonal assemblages withgenerally positive coefficients that are more easily interpretedthan an unrotated solution. The elements of Bct describe therelative contribution of each variinax assemblage to each coretop sample. The model fit is described by its communality (thesum of squares of Bct or the normalized vector length), which isa linear measure of the fraction of information retained fromeach sample [Imbrie and Kipp, 1971]. A perfect model fit isachieved when communality goes to 1, and Ect=O.

We use the transpose of the core top factor score matrix(FTct) to evaluate the structure of the sediment trap data matrix(Ust) by determining a factor loading matrix (Bst) for thesediment trap data set:

U5 FTct = Bst + Et (2)

This assumes that the structure of the core top assemblagesapply to the sediment trap faunas. if this assumption is notcorrect, the resulting sample communality will be low, and theelements of Est will be nonzero. If the distribution of the coretop assemblages are controlled by their environment, then tshould produce distinct patterns when plotted against acontrolling environmental variable such as SST [Imbrie andKipp, 1971]. However, if the fundamental structure of thesediment trap faunas differs from that of the core top faunas,then the distribution pattern of B5t with respect to SST will notmatch that of Bet.

The core top factor loadings in Bct were regressed againstSST to develop a stepwise least squares transfer function.

ORTIZ AND MIX: SEDIMENT TRAP-CORE TOP COMPARISON 177

0 30 60 90 120 150 180 150 120 90 60 30 090 90


9O

6O

3O

3O

6O

9O 0

3O

+ +

60 90 120 150 180 150 120 90 60 30 0 9O

6O

3O

LONGITUDE

Figure 2. Locations for the 1121 core top samples in the global core top data set.

identical to that of the core top sediments. In our studies we have measured both the 125-150 and >150 [lxn size fractions.

This allows us to easily compare our samples to >150 [lxn sediments or >125 •m sediment trap samples. We recommend that future sediment trap and plankton tow studies follow this procedure or focus on the >150 [lxn size fraction. While variations in sieve size introduce some uncertainty in the sediment trap to core top comparison, the residual errors we obtained are not correlated with sieve size variations. Any bias introduced by this source of error is thus not systematic. These potential sources of error will be evaluated further in the discussion to follow.

Q-Mode Factor Analysis and Imbrie-Kipp Transfer Functions

As a first step in the comparison of core top and sediment trap faunas, we calculated a Q-mode factor model for the global core top database following Imbrie and Kipp [1971]. Q-mode factor analysis provides an objective, quantitative means of simplifying complex data sets. This is accomplished by decomposing the core top data matrix (Uct) into a factor score matrix (Fct), a factor loading matrix (Bct), and an error matrix (Ect):

Uct = Bet Fct + Ect (1)

Uct, the core top percent abundance data, is based on counts of n=27 planktonic foraminiferal taxa in N=1121 coretops. Each core top fauna in Uct is row normalized so that its sum of squares is unity. The elements of the factor score matrix, F ct, describe m varimax rotated assemblages composed of weighted

contributions from each of the 27 taxa (see Table 3). A varimax rotation is applied to Fct so that the resulting assemblages remain close to the centroid of the sample data. This rotation has the advantage of producing orthogonal assemblages with generally positive coefficients that are more easily interpreted than an unrotated solution. The elements of Bet describe the relative contribution of each varimax assemblage to each core top sample. The model fit is described by its communality (the sum of squares of Bet or the normalized vector length), which is a linear measure of the fraction of information retained from

each sample [Imbrie and Kipp, 1971]. A perfect model fit is achieved when communality goes to 1, and Ect=0.

We use the transpose of the core top factor score matrix (F?ct) to evaluate the structure of the sediment trap data matrix (Ust) by determining a factor loading matrix (Bst) for the sediment trap data set:

Ust FTct = Bst + Est (2)

This assumes that the structure of the core top assemblages apply to the sediment trap faunas. ff this assumption is not correct, the resulting sample communality will be low, and the elements of Est will be nonzero. If the distribution of the core top assemblages are controlled by their environment, then Bst should produce distinct patterns when plotted against a controlling environmental variable such as SST [Imbrie and Kipp, 1971]. However, if the fundamental structure of the sediment trap faunas differs from that of the core top faunas, then the distribution pattern of Bst with respect to SST will not match that of Bct.

The core top factor loadings in Bct were regressed against SST to develop a stepwise least squares transfer function.


0\000

. , . , j";l .jU'nO 00 In In 00 CV) InN In in In N N C C N In N NAAAAAAAAAAAAA

o o 00000 N 0 In 00 mN N N N N N N cfl N 0%'n NN .'Cl

000000000%00%N00 IO

ON 00 NO in N 0000 c %D iflC N In 0 In N 000000 N Iflin in vi N 00 N N

O 0' 0% In 00 In N In ON 'C In C Cfl In C C C) 0%

c,cn--

. u U

u

v-i '0 N 000% 0N m

I6

I

J

Following Imbrie and Kipp [1971], we include squared and crossproduct tenns for each factor in the stepwise regression. Thesetms are included in the final regression if their partial F valueexceeds the critical 5% significance threshold. Squared andcross product terms are included so that our results can becompared with equations of the type developed for CLIMAP[1976, 1981] and because each factor may exhibit nonlinear,parabolic responses to temperature and/or interactive effects[Imbrie and Kipp, 1971]. The regression coefficients from theglobal Imbrie-Kipp transfer function are used with the sedimenttrap factor loadings in B5t to determine Imbrie-Kipptemperature estimates for each of the sediment trap faunas.

The transfer function SST is compared with seasonallyweighted SST from each trap location obtained from Levitus[1982]. The SST weighting for each trap temperature isdetennined from its duration and deployment season. Thisprocedure is necessary because some of the sediment trapdeployments did not sample the entire annual cycle. Assigningan annual average temperature to a subannual forarniniferalfauna would introduce an unrealistic temperature bias to ourcomparisons.

The Modern Analog Method

We calculate modern analog SST from the sediment trapfaunas using the core tops as the calibration data set. We thencompare these SST estimates with measured SST values fromLevitus [1982] at the sediment trap locations. As a final test ofthe modern analog method, we estimated modern analog SST foreach core top in the database by comparison against everyother core cop in the database. This global calculation issimilar to the separate ocean basin calculations of Prell [1985].It allows us to evaluate the level of variation in the sedimenttrap versus core top comparison against the level of variationobserved in the core top data set.

The modem analog method compares planktonicforaminiferal assemblages in samples with unknownenvironmental conditions (SST in this application) with coretops from locations with known environmental conditions.The two basic assumptions for the modern analog method aresimilar to those of the Imbrie-Kipp transfer function. The firstassumption is that similar foraminiferal faunal assemblages areproduced by similar suites of environmental conditions. Thesecond assumption is that SST is the environmental variablewhich determines variation in foraminiferal assemblages or iscorrelated with environmental variables which determineforaminiferal variation.

The first asswnpcion holds well for most of the world ocean.One notable exception to the rule occurs at very high northernand southern latitudes. At these climatic extremes, theforaminiferal fauna becomes essentially monospecific,dominated by left-coiling N. pachyderma. However, highsouthern latitudes are cooler by l°-2°C than high northernlatitudes. Because foraminiferal faunas attain essentiallymonospecific status well before extremely cold southern oceanSST values are reached, use of a global data set to predictmodem analog SST at extremely high latitudes can giveerroneous results ckn to the averaging of SST from bothhemispheres. To avoid this problem, we follow the standard


AAAAAAAAAAAAA

Following lmbrie and Kipp [1971], we include squared and cross product terms for each factor in the stepwise regression. These terms are included in the final regression if their partial F value exceeds the critical 5% significance threshold. Squared and cross product terms are included so that our results can be compared with equations of the type developed for CLIMAP [1976, 1981] and because each factor may exhibit nonlinear, parabolic responses to temperature and/or interactive effects [lmbrie and Kipp, 1971]. The regression coefficients from the global Imbrie-Kipp transfer function are used with the sediment trap factor loadings in Bst to determine Imbrie-Kipp temperature estimates for each of the sediment trap faunas.

The transfer function SST is compared with seasonally weighted SST from each trap location obtained from Levitus [1982]. The SST weighting for each trap temperature is determined from its duration and deployment season. This procedure is necessary because some of the sediment trap deployments did not sample the entire annual cycle. Assigning an annual average temperature to a subannual foraminiferal fauna would introduce an unrealistic temperature bias to our comparisons.

The Modern Analog Method

We calculate modern analog SST from the sediment trap faunas using the core tops as the calibration data set. We then compare these SST estimates with measured SST values from Levitus [1982] at the sediment trap locations. As a final test of the modem analog method, we estimated modem analog SST for each core top in the database by comparison against every other core top in the database. This global calculation is similar to the separate ocean basin calculations of Prell [ 1985]. It allows us to evaluate the level of variation in the sediment

trap versus core top comparison against the level of variation observed in the core top data set.

The modem analog method compares planktonic foraminiferal assemblages in samples with unknown environmental conditions (SST in this application) with core tops from locations with known environmental conditions. The two basic assumptions for the modern analog method are similar to those of the Imbrie-Kipp transfer function. The first assumption is that similar foraminiferal faunal assemblages are l:noduced by similar suites of environmental conditions. The second assumption is that SST is the environmental variable which determines variation in foraminiferal assemblages or is correlated with environmental variables which determine foraminiferal variation.

The farst assumption holds well for most of the world ocean. One notable exception to the rule occurs at very high northern and southern latitudes. At these climatic extremes, the foraminiferal fauna becomes essentially monospecific, dominated by left-coiling N. pachyderma. However, high southern latitudes are cooler by 1ø-2øC than high northern latitudes. Because foraminiferal faunas attain essentially monospecific status well before extremely cold southern ocean SST values are reached, use of a global data set to predict modem analog SST at extremely high latitudes can give erroneous results due to the averaging of SST from both hemispheres. To avoid this problem, we follow the standard

Tab

le 2

. Flu

x-W

eigh

ted

Perc

ent A

bund

ance

for

the

Sedi

men

t Tra

p Fo

ram

inif

eral

Fau

nas

as N

umbe

red

in T

able

1

9 7

aSom

e ra

re s

peci

es th

at w

ere

repo

rted

in th

e or

igin

al s

ourc

es a

re o

mitt

ed h

ere

from

the

flux

wei

ghte

d pe

rcen

ts b

ecau

se th

ese

taxa

wer

e no

t uni

forn

ily r

epor

ted

by a

ll w

orke

rs. T

hese

cat

egor

ies

wer

e es

sent

ialy

trea

ted

as u

nide

ntif

ied.

bG. r

uber

(to

tal)

= G

. rub

er (

whi

te)

and

G. r

uber

(pi

nk);

G. s

accu

1fer

(to

tal)

= G

. sac

culjf

er a

nd G

. tri

lobu

s; G

. nw

nard

ii (t

otal

) =

G. m

enar

dii,

G. t

umid

a,an

dG

. men

ardi

i neo

flex

uosa

.

Map

Loc

atio

ns

12

34

56

78

910

1112

13

0. w

zive

rsa

2.9

1.8

8.6

37.0

4.9

1.1

2.7

0.3

0.4

0.2

0.0

0.0

0.0

G. c

onql

obat

us0.

00.

00.

00.

00.

00.

32.

00.

80.

30.

60.

00.

00.

0G

. rub

er (

tota

l)"

0.0

1.8

4.8

13.0

2.5

33.4

14.8

38.4

14.9

21.6

0.0

0.0

0.0

G. t

enel

us0.

00.

00.

00.

00.

00.

017

.30.

00.

11.

00.

00.

00.

0G

. sac

culif

er (

tota

l)"

0.0

0.0

0.0

0.0

0.0

0.4

5.2

10.0

2.9

9.6

0.0

0.0

0.0

S. d

ehis

cens

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.3

0.0

0.0

0.0

0.0

0.0

G. a

equi

laie

rahs

0.0

0.0

0.0

0.0

1.4

4.6

0.4

5.6

1.4

4.5

0.0

0.0

0.0

G. c

alid

a0.

01.

80.

06.

00.

00.

00.

90.

50.

66.

20.

00.

00.

0G

. bui

loid

es2.

99.

715

.213

.035

.432

.20.

90.

313

.38.

20.

00.

00.

0G

. fac

lone

nsis

0.0

20.4

4.8

6.0

0.0

0.0

0.0

0.0

0.0

1.4

0.0

0.0

0.0

G.d

igila

ta0.

00.

00.

00.

00.

30.

00.

40.

00.

10.

00.

00.

00.

0G

. rub

esce

r,.s

0.0

0.0

0.0

0.0

oo1.

123

.31.

503

2.5

0.0

0.0

0.0

G. q

uinq

uelo

ba58

.21.

86.

73.

024

.10.

00.

20.

81.

90.

00.

00.

06.

3L

eft-

coili

ng N

. pac

hyde

rma

16.8

27.4

6.7

0.0

0.2

0.0

0.0

0.0

0.0

0.0

90.0

96.0

92.0

Rig

ht-c

oilin

g N

. pac

hyde

rma

11.8

14.2

21.9

7.0

17.5

0.0

0.0

0.0

3.0

0.2

10.0

4.0

1.7

N. d

uter

trei

0.0

17.6

16.2

2.0

4.9

0.9

0.2

5.6

27.2

9.4

0.0

0.0

0.0

G. c

ongl

o,ne

rata

0.0

0.0

0.0

0.0

0.0

0.0

0.9

0.0

0.3

2.0

0.0

0.0

0.0

G. h

exag

ona

0.0

0.0

0.0

0.0

1.4

o.o

4.3

0.0

1.3

2.0

0.0

0.0

0.0

P. o

bliq

uilo

cula

ta0.

00.

00.

00.

00.

00.

70.

00.

00.

38.

20.

00.

00.

0G

. inf

lata

0.0

0.0

0.0

0.0

0.0

1.2

0.0

0.0

0.0

0.0

0.0

0.0

0.0

Lef

t-co

iling

G. t

runc

atul

inoi

des

0.0

0.0

0.0

0.0

0.0

9.3

0.0

0.0

0.0

0.0

0.0

0.0

0.0

Rig

ht-c

oilin

g G

. tru

ncat

u1ir

oide

s0.

00.

00.

00.

00.

00.

00.

00.

00.

00.

00.

00.

00.

0G

. cra

ssaf

orm

is0.

00.

00.

00.

00.

00.

20.

00.

00.

00.

00.

00.

00.

0G

. hi,y

uta

0.0

0.0

0.0

0.0

0.0

3.9

0.0

0.0

0.0

0.0

0.0

0.0

0.0

G. s

citii

la0.

21.

88.

63.

00.

00.

02.

56.

40.

21.

50.

00.

00.

0G

. men

ardi

i (to

tal)

"0.

00.

00.

00.

00.

00.

00.

43.

64.

72.

30.

00.

00.

0G

. glu

tinat

a7.

21.

86.

77.

05.

210

.817

.023

.523

.56.

50.

00.

00.

0


.g

o. o.o. ooooooooo. o.o.o.o.o.o.o.o.o.o.o.o.o.o.o. o

••oo••oo•oo•oooooooo••

N N •: K•4 d d d d d d • d d d d c5 •:c5 d d d d d N cSK

,4 d• dd d •: d N dd,4d dd dd dd,4c• dd •d dd

•: d N d d d,-4 d• dd d •: dK •: d,4d dd d d d d d•

• d•dddd••dd•dKNdddddddd•dK

00 ,4 d 'dddd,4c•dd4•:Kdddddddd4d,4

N • • c:5 c• d c:5 d N d d c:5 • ,• 4 c:5 c:5 c5 d d d d c:5 d c5 c:5•

180 ORTIZ AND MIX: SEDIMENT TRAP-CORE TOP COMPARESON

Table 3. Factor Scores for the Seven-Factor, Varimax Rotated,Q-Mode Factor Model Based on 1121 Core Tops

procedure of Preli [1985] and compare high-latitude (northern)southern hemisphere trap samples against (northern) southernhemisphere core top samples only.

Assessment of the second assumption is more problematic.For relatively small spatial scales and short timescales,absolute abundance assemblages of planktonic foraminiferafrom plankton tow and seasonal sediment traps demonstrategreater control by biological factors (food availability andlight) than temperature [Fairbanks and Wiebe, 1980; Watkinset al. 1996; Ortiz and Mix, 1992; Ortiz et al. 1995]. Here wefocus on the relative abundance of foraminiferal assemblages atlarge spatial scales, integrated over time scales of generallylonger than the annual cycle. Under such conditions,foraminiferal core top assemblages exhibit relatively strongrelationships to SST [e.g., Imbrie and Kipp, 1971; PrelI,1985]. We wish to determine whether the correlation to SST isalso present in the global data set of temporally averagedsediment trap faunas.

Temperature estimates based on the modern analog methoddepend on averages of sample-by-sample comparisons. Hutson[1980] used the cosine theta angle as a measure of themultivariate distance (i.e., dissimilarity) between twoforaminiferal faunas. Empirical pollen studies suggest that thesquared chord distance provides more reliable results thanseveral other dissimilarity measures including the cosine theta

Dnninant species.G. rube, (total) = G. rube, (white) and G. rube, (pink); G. saccu1fer (total) = G. saculjfer and G. trilobMs; G. menardii (Total) =G. ,nenardii, G. twnida, and G. menardii neoflexuosa.

1ota1 information explained: 91.5%.

angle [Prentice, 1980; Overpeck et al. 1985]. The use of thesquared chord distance tends to increase the importance of rarespecies and decrease the importance of abundant ones so thatdifferences in the abundant species alone do not dominate thedissimilarity estimate. Prell [1985] demonstrated that thesquared chord distance worked well with foraminiferal faunas.Accordingly, we use the squared chord distance between thetarget sample and each core top in the database as a measure ofintersample dissimilarity:

m

d=[p 1t2 1/22it -pit I

k=1In this notation, d11 is the squared chord distance between the ithand jth samples, while p. and p.1 represent the fractionalpercentages of the kth species in samples i and j. The 27taxonomic categories used to calculate the squared chorddistance are the same as those used in the Q-mode factoranalysis described above (Table 3). Just as the communalityprovides a quantitative estimate of how well a foraminiferalfauna is explained by a factor model, the average sampledissimilarity provides a quantitative estimate of the similaritybetween a target fauna and its closest analogs. Empiricalstudies suggest that values of d, < 0.25 yield reliable analogs(W. Prell, personal communication, 1996). We tested this rule

(3)

Taxon

Factor 1,G.rub.T.IGsac. T.

Factor 2.G. men. T.

Factor 3.N.pac.L

Factor 4.G.inf.

Factor 5,N.dul.

Factor 6,G. bulIJG.glut.

Factor 7.P.obliq.

0. w&iversa 0.02 0.00 0.00 0.03 0.00 0.01 0.01G. conglobahts 0.05 0.01 -0.00 0.00 -0.00 0.00 0.04G. rube, (total) 0.90' -0.05 0.01 0.06 -0.05 -0.11 -0.11G. tenalus 0.04 -0.01 -0.01 -0.00 -0.00 0.02 0.00G. sacculifer (total) 0.31' 0.07 0.03 -0.02 0.04 -0.08 0.02S. dehircens 0.00 0.03 0.00 0.00 -0.00 -0.00 0.01G. aequiloieralLc 0.10 0.01 -0.01 -0.00 0.01 0.02 0.11G. caliihi 0.05 -0.01 -0.01 0.01 0.01 0.01 0.06G. bulloides -0.03 0.06 0.17 0.13 -0.06 0.81' -0.16G.facloneans 0.05 -001 -0.04 0.13 -0.03 0.05 -0.02G. digitata 0.01 0.02 0.00 0.01 .0.00 0.00 -0.00G. rubescerss 0.03 -0.01 -0.01 -0.01 -0.00 0.02 0.01G. quinqueloba -0.01 -0.01 0.12 0.01 0.01 0.06 -0.01Left-coiling N. pachyderma 0.02 -0.00 0.97' -0.03 -0.01 -0.07 0.04Right-coiling N. pachwie,,na -0.04 -0.09 0.04 0.26 0.10 0.05 0.04N. duterirci 0.02 0.03 0.02 0.08 0.98' 0.01 0.00G. conglomerala 0.01 0.04 0.00 -0.01 0.01 0.00 0.03G. hexagona 0.02 0.01 -0.00 -0.01 0.01 0.02 -0.01P. obliquioculata 0.01 0.12 0.03 0.03 -0.04 -0.03 0.93'G.inflata -0.02 0.03 -0.02 0.92' -009 -0.07 0.03Left-coilingG. truncatulinoides 0.02 0.01 -0.01 0.09 -0.02 0.00 -0.03Right-coilingG. truncalulinoides 0.02 0.01 001 0.10 -0.03 0.01 -0.02G. crassaformis 0.01 0.01 0.00 0.02 0.00 -0.01 -0.00G.hirsuta 0.01 0.00 -0.01 0.05 -0.01 -0.01 -0.01G.sciwia 0.01 -0.00 -0.00 0.03 0.00 0.02 -0.00G. ,nanardii (total) 0.05 0.97' -0.03 -0.04 -0.01 0.04 -0.07G. glutinata 0.26 -0.14 -0.12 -0.12 0.05 0.55' 0.27Infonnation per Factot 27.4% 12.1 % 8.4% 12.5% 14.6% 11.5% 5.1%


Table 3. Factor Scores for the Seven-Factor, Varimax Rotated, Q-Mode Factor Model Based on 1121 Core Tops

Factor 1, Factor 6, G. rub. T./ Factor 2, Factor 3, Factor 4, Factor 5, G. bull./ Factor 7,

Taxon G•vac. T. G. men. T. N. pac. L. G. inf. N. dut. G. glut. P. obliq.

O. universa 0.02 0.00 0.00 0.03 0.00 0.01 0.01

G. conglobatuSo 0.05 0.01 -0.00 0.00 -0.00 0.00 0.04 G. ruber (total) 0.90' -0.05 0.01 0.06 -0.05 -0.11 -0.11 G. tenelus 0.04 -0.01 -0.01 -0.00 -0.00 0.02 0.00

G. sacculiœer (total) ø 0.31' 0.07 0.03 -0.02 0.04 -0.08 0.02 S. dehiscens 0.00 0.03 0.00 0.00 -0.00 -0.00 0.01

G. aequilateralis 0.10 0.01 -0.01 -0.00 0.01 0.02 0.11 G. calma 0.05 -0.01 -0.01 0.01 0.01 0.01 0.06 G. bulloides -0.03 0.06 0.17 0.13 -0.06 0.81' -0.16

G. faclonensis 0.05 -0.01 -0.04 0.13 -0.03 0.05 -0.02 G. di•#ata 0.01 0.02 0.00 0.01 -0.00 0.00 -0.00 G. rubescerts 0.03 -0.01 -0.01 -0.01 -0.00 0.02 0.01

G. ½uincueloba -0.01 -0.01 0.12 0.01 0.01 0.06 -0.01 Left-coiling N. pachyderma 0.02 -0.00 0.97' -0.03 -0.01 -0.07 0.04 Ri•at-coiling N. pachwlerma -0.04 -0.09 0. 04 0.26 0.10 0. 05 0.04 N. dutertrei 0.02 0.03 0.02 0.08 0.98' 0.01 0.00

G. conl•lomerata 0.01 0.04 0.00 -0.01 0.01 0.00 0.03 G. hexagona 0.02 0.01 -0.00 -0.01 0.01 0.02 -0.01 P. obli•uiloculata 0.01 0.12 0.03 0.03 -0.04 -0.03 0.93' G. i .n//ata -0.02 0.03 -0.02 0.92' -0.09 -0.07 0.03 Left-coilingG. truncatulinoides 0.02 0.01 -0.01 0.09 -0.02 0.00 -0.03 Right-coilingG. truncatulinoides 0. 02 0.01 0. 01 0.10 -0.03 0. 01 -0.02 G. crassaformis 0.01 0.01 0.00 0.02 0.00 -0.01 -0.00 G. hirsuta 0.01 0.00 -0.01 0.05 -0.01 -0.01 -0.01 G. scitula 0.01 -0.00 -0.00 0.03 0.00 0.02 -0.00 G. menardii (total) ø 0.05 0.97' -0.03 -0.04 -0.01 0.04 -0.07 G. •lut/nata 0.26 -0.14 -0.12 -0.12 0.05 0.55' 0.27 Information per Factoff 27.4% 12.1% 8.4% 12.5% 14.6% 11.5% 5.1%

•Dominant species. G. ruber (total) = G. ruber (white) and G. tuber (pink); G. sacculifer (total) = G. saculifer and G. trilobus; G. menardii (Total) = G. menardii, G. turnida, and G. menardii neoflexuosa.

•Fotal information explained: 91.5%.

procedure of Prell [1985] and compare high-latitude (northern) southern hemisphere trap samples against (northern) southern hemisphere core top samples only.

Assessment of the second assumption is more problematic. For relatively small spatial scales and short timescales, absolute abundance assemblages of planktonic foraminifera from plankton tow and seasonal sediment traps demonstrate greater control by biological factors (food availability and light) than temperature [Fairbanlcs and Wiebe, 1980; Watkins et al. 1996; Ortiz and Mix, 1992; Ortiz et al. 1995]. Here we focus on the relative abundance of foraminiferal assemblages at large spatial scales, integrated over time scales of generally longer than the annual cycle. Under such conditions, foraminiferal core top assemblages exhibit relatively strong relationships to SST [e.g., lmbrie and Kipp, 1971; Prell, 1985]. We wish to determine whether the correlation to SST is also present in the global data set of temporally averaged sediment trap faunas.

Temperature estimates based on the modem analog method depend on averages of sample-by-sample comparisons. Hutson [1980] used the cosine theta angle as a measure of the multivariate distance (i.e., dissimilarity) between two foraminiferal faunas. Empirical pollen studies suggest that the squared chord distance provides more reliable results than several other dissimilarity measures including the cosine theta

angle [Prentice, 1980; Overpeck et al. 1985]. The use of the squared chord distance tends to increase the importance of rare species and decrease the importance of abundant ones so that differences in the abundant species alone do not dominate the dissimilarity estimate. Prell [1985] demonstrated that the squared chord distance worked well with foraminiferal faunas. Accordingly, we use the squared chord distance between the target sample and each core top in the database as a measure of intersample dissimilarity:

rn

diJ • [Pill/2 _ 1/2•2 (3) = - PjI• 1 k=l

In this notation, dij is the squared chord distance between the ith and jth samples, while P ik and P jk represent the fractional percentages of the kth species in samples i and j. The 27 taxonomic categories used to calculate the squared chord distance are the same as those used in the Q-mode factor analysis described above (Table 3). Just as the communality provides a quantitative estimate of how well a foraminiferal fauna is explained by a factor model, the average sample dissimilarity provides a quantitative estimate of the similarity between a target fauna and its closest analogs. Empirical studies suggest that values of dij < 0.25 yield reliable analogs (W. Prell, personal communication, 1996). We tested this role

of thumb by calculating d for each sample in the core top dataset in comparison with all others. While values of du are notnormal in distribution, natural log transformed values of d areapproximately log normal. We note that 97% of the faunas intho core top data set have an average d., value for their top five

analogs that is <0.20 and that 99% of the faunas have averaged values <0.26. W. L. Prell (personal communication, 1996)reports similar experience with applications of this method toforaminiferal faunas. We thus choose 0.20 as a conservativecutoff limit.

0khZ AND M1X SEDIMENT TRAP-CORE TOP COMPARISON 181

Factor 1 (G. rub. T., G. sac: 27.4%) Factor 2 (G. men. T.: 12.1%)1.00- 1.00-

bO 0.75- C 0.75 -00 0.50- 0 0.50--I-0 025- I- 0.25 -

rJ000- ri..

0.00--0.25 - -0.25 - I I I I I I I

-5 0 5 10 15 20 25 30 -5 0 5 10 15 20 25 30

SST (°C) SST (°C)

Factor 3 (N. pac. Left: 8.4%) Factor4(G. inf.: 12.5%)bfJ

1.00-0.75 - bO

C

1.00-0.75 -

0 0.50- .0 0.50-I- 0.25 - I-0 0.25 -

000- 0.00--0.25 -,

-5I

0I

5I I

10 15I

20I I

25 30-0.25-, , , ,

-5 0 5 10 15 20 25 30SST (°C) SST (°C)

Factor 5 (N. dut.: 14.6%) Factor 6 (G. bull., G. glut.: 11.5%)

C

1.00-0.75 -

bfC

1.00-

0.75 -

0 0.50- 0.50-

I.-0 0.25- 0.25-

0.00- 0.00-

-0.25 -,-5

I

0I

5I I

10 15I

20I I

25 30

-0.25-, ,

-5 0, , , ,

5 10 15 20 25 30

SST (°C) SST (°C)

Factor 7 (P. obliq.: 5.1%) Communality1.00-1.00

C 0.75 - 0.75 -0 0.50-I- 0.25- E

E

0.50-

0.00- 0L) 0.25 -

-0.25 - I I I I I I I 0.00-,-5 0 5 10 15 20 25 30 -5 0 5 10 15 20 25 30

SST (°C) SST (°C)

Sediment Traps C Core Tops

Figure 3. Factor loadings and communalities versus sea surface temperature for the global Q-mode factormodel described in the text. Core tops are small open squares, while sediment traps are large solid squares.Abbreviated names list the dominant species for each factor. Figures 3a-3g correspond to factors 1-7. Figure 3hdisplays sample communalities.

f


Factor 1 (G. rub. T., G. sac' 27.4%) 1.00

0.75

0.50

0.25

0.00

-0.25 I I I I I I

-5 0 5 10 15 20 25 30

SST (øC)

Factor 2 (G. men. T.' 12.1%) 1.•

.E 0.75 r2• cl • :•' o 0.50 :c• • .

•- 0.25 t ¸

• -.'-'-... '..:' :....'i¾:' • 0.00 -0.25 i i i I I I

-5 0 5 10 15 20 25 30

SST (øC)

1.00

0.75 1 0.50

0.25

0.00

-0.25

-5

Factor 3 (N. pac. Left' 8.4%) ß

•2 >: . -:-•, t" ß • ,.'•<'.., .::•'• g•.!?...- .

-":.?' *::"::::"'"::•i•'i•

0 5 10 15 20 25 30

SST (øC)

Factor 4 (G. inf.' 12.5 %) 1.00

.. - .•;:: .:.:'q-..': 0.25

-0.25 -5 0 5 10 15 20 25 30

SST (øC)

1.oo 0.75

0.50

0.25

0.00

-0.25

-5

Factor 5 (N. dut.' 14.6%)

' •.;•gY'-')2' '"."?.':,'::'i e •._ ..'..•7 ::• .......

'•. "•'•' '-• '• .:' :•..'.i:::::...•.i.:i• - •, .•<'-•;-..! -..-:::, i.:'•: ....• .'.'.'"..,,:.!:.;:;.-. • •...' . •. ,•,,. :.: . :•.:.. ..::;. : '{• •"- •.: ..... '. ..:. - ', .:•' . .-.:'.•i::•:,-.-..::::?.,

.:: ...... ß ::.:,.!.• ....,:::.....•!:'•'_:•.•i?:•iii,,•,,i:"•::::•.':;•iiiii!' ß :.,y,... .., _.:.......:::::::::--:.:• ',: .•:'.:::•: ..:.: ::.:.:.'...•,•,.:

0 5 10 15 20 25 30

SST (øC)

Factor 6 (G. bull., G. glut.' 11.5%) 1.00

0.75 - '*•5 :-' "'- :""• ;' -•" ..... " ,.•:gg.;5.':.q C_._"•: -" )-- 'r• •- -::...?• ...... ':•:' -•

-. ;..:j..' :!!.!...,::. ':i l -0.25 I I i i i I* -5 0 5 10 15 20 25 30

SST (øC)

1.00

.E 0.75

o 0.50

• 0.25

• 0.00

-O.25

Factor 7 (P. obliq.' 5.1%)

g • 0.75 h ß .

• .::-?;.f".. •;.•. ß - y,. ¾,:/.

• % .... :'....::::'.'5,. - .i-..iff: i:::;:i:: .......... :'::i:-:i:"'

-:• ':5 :i!•i •;•::•+,, .:.i:i:•i:•i!•?:.:.::• ......... .... -.'" :: $::.•:..:. ::::::::::::::::::::::::::::::::::::::::::::::::::::::::: .:::: ......

-•" --o:'•,'-3 '" • ...... .- •.

0.50

0.25 1 0.00

-5

I

-5 0 5 10 15 20 25 30 0

SST (øC)

I Sediment Traps Core Tops

Communality

:%. ½ :: ...,:. =

I I I I

5 10 15 20 25 30

SST (øC)

Figure 3. Factor loadings and communalities versus sea surface temperature for the global Q-mode factor model described in the text. Core tops are small open squares, while sediment traps are large solid squares. Abbreviated names list the dominant species for each factor. Figures 3a-3g correspond to factors 1-7. Figure 3h displays sample communalities.

of thumb by calculating d o for each sample in the core top data set in comparison with all others. While values of d o are not normal in distribution, natural log transformed values of d o ß are approximately log normal. We note that 97% of the faunas in the core top data set have an average d 0 value for their top five

analogs that is <0.20 and that 99% of the faunas have average d 0 values <0.26. W. L. Prell (personal communication, 1996) reports similar experience with applications of this method to foraminiferal faunas. We thus choose 0.20 as a conservative cutoff limit.

182 ORTIZ AND MDC: SEDIMENT TRAP-CORE TOP COMPARISON

Using the values of as the selection cTiteria, the SSTvalues from the five core tops least dissimilar to the targetsample are averaged to estimate the modern analog SST for thatsample. We use arithmetic, rather than weighted averages,when calculating analog SST estimates. Tests we haveconducted demonstrate that SST averages weighted by theindividual dissimilarity estimates as proposed by Hiason[1980] are not significantly different from arithmetic averages.We did not experiment with more sophisticated geographicweighting methods [e.g., Pflaumann et al. 1996].

Results

Q-Mode Factor Analysis

Seven factor assemblages together account for 91.5% of theinformation in the global core top data set. The addition of aneighth factor assemblage would have increased the totalinformation explained by only 2.4% to 93.9%. Models withfew assemblages generated groupings that were ecologicallyless distinct. Table 3 lists the factor scores for the sevenassemblages we chose to retain. The seven factors groupspecies with similar distributions into assemblages whichindividually account for 5-27% of the total information. Theseassemblages correspond roughly to oceanographicenvironments. For example, factor 1, which is dominated byG. rube, and G. sacculifer, is important in warm, oligotrophicregions, while factor 3, composed of left-coiling N.pachyderina, is indicative of cold, high-latitude extremes.

To compare the sediment trap and core top foraminiferalfaunas, we apply the core top factor scores to the sediment trapdata and extract factor loadings for each sample. Plots of factorloading against SST from Levitus [1982] serve as a usefulindication of sample environments (Figure 3). In general,factor loadings from the two data sets have similar trends withrespect to SST. This is most clear for factors 1, 3, 5, and 6(Figure 3).

Potential differences between the two data sets do exist. Theamplitude of the factor loadings for factors 2, 4, and 7 arediminished in the sediment trap assemblages relative to thecore top assemblages. The dominant species in these factorsare relatively rare in the sediment traps and abundant in some ofthe core top sediments. This could be a sampling problem ormay reflect a bias in the sediments. All of the species involvedhave heavily calcified shells, suggesting that dissolution mayenrich them in the sediments relative to their abundance in theoverlying water column.

The model communality measures how much of theinformation content of the foraminiferal faunas is explained bythe factor model (Figure 3h). For both core top and sedimenttrap faunas the factor model explains adequate amounts ofsample information at latitudinal extremes (SST <9°C or >20°C)but explains little information in the midlatitude faunas at SSTfrom 9°-15°C. Note, however, that this pattern is seen in thecommunalities of both the core top and sediment trap faunas,suggesting that a similar process contributes to the effect inboth data sets. We explore how low sample communality mayaffect the SST estimates derived from the Imbrie-Kipp andmodem analog methods in the discussion section.

Table 4. Coefficients for the Global Imbrie-KippTransfer Function

Global Imbrie-KIpp Transfer Function

A statistically significant SST transfer function wasdeveloped using the core top data set. The terms for this globallmbrie-Kipp transfer function are listed in Table 4. Thisregression, based on 1121 samples, is significant at p<0.Ol,has an r2 of 0.93, and an RMS error of 1.9°C. The core top SSTestimates have a slope of 0.93±0.02 and an intercept of1.7±0.4 at the 95% confidence limit with respect to Levitus[1982] SST (Figure 4a). The slope of the core top regression issignificantly different from one (with 95% confidence), giventhe sample size of 1121 core tops.

To compare the core top and sediment trap assemblages, weapplied the core top transfer function to the sediment trap factorloadings (Figure 4, Table 5). When the transfer function isapplied to the sediment trap samples, the SST estimates followa slope of 0.92±0.16 and an intercept of 4.2±2.8 at the 95%confidence limit with respect to the Levitus [1982] SST values.The RMS error for the sediment trap SST estimates with thismethod was 2.6°C. The slope of the sediment trap relationship(0.92±0.16) is not significantly different from that of the coretops, or from one. However, as was the case for the core tops,the 95% confidence interval on the regression interceptindicates a warm bias in the sediment trap SST estimates (Figure4b). A two-sided t-test of the 3.0°C average Imbrie Kipp-SSTsediment trap residual demonstrates that this bias is significantat psO.Ol (degrees of freedom (4t) =12, RMSE=2.6°C, t-value=4.18 > [email protected]=3.055). The sediment trap SSTestimates also appear to exhibit a residual trend as a function ofthe factor model communality. However, it is unlikely that thisfeature is statistically significant: The trend is not present inthe core top SST residuals and occurs only at very lowcommunality in the sediment trap residuals (Figure 4c).

Term Coefficient

Intercept 27.6Factor 3 -46.2Factor 4 -26.2Factor 3 squared 21.2Factor 4 squared 18.6Factor 5 squared -1.1Factor 6 squared -3.3Factor 7 squared 3.0Factor 1 x factor 3 72.8Factor 1 x factor 4 4.5Factor 1 x factor 6 5.4Factor 2 x factor 3 55.9Factor 2 x factor 4 13.4Factor 2 x factor 6 9.3Factor 2 x factor 7 -3.8Factor 3 x factor 4 21.4Factor 3 x factor 5 20.0Factor 3 x factor 6 10.6Factor 4 x factor 6 -3.1Factor 4 x factor 7 19.7Factor 5 c factor 6 -6.0


Using the values of dij as the selection criteria, the SST values from the five core tops least dissimilar to the target sample are averaged to estimate the modern analog SST for that sample. We use arithmetic, rather than weighted averages, when calculating analog SST estimates. Tests we have conducted demonstrate that SST averages weighted by the individual dissimilarity estimates as proposed by Hutson [1980] are not significantly different from arithmetic averages. We did not experiment with more sophisticated geographic weighting methods [e.g., Pfiaumann et al. 1996].

Results

Q-Mode Factor Analysis

Seven factor assemblages together account for 91.5% of the information in the global core top data set. The addition of an eighth factor assemblage would have increased the total information explained by only 2.4% to 93.9%. Models with few assemblages generated groupings that were ecologically less distinct. Table 3 lists the factor scores for the seven

assemblages we chose to retain. The seven factors group species with similar distributions into assemblages which individually account for 5-27% of the total information. These assemblages correspond roughly to oceanographic environments. For example, factor 1, which is dominated by G. ruber and G. sacculifer, is important in warm, oligotrophic regions, while factor 3, composed of left-coiling N. pachyderma, is indicative of cold, high-latitude extremes.

To compare the sediment trap and core top foraminiferal faunas, we apply the core top factor scores to the sediment trap data and extract factor loadings for each sample. Plots of factor loading against S•F from Levitus [1982] serve as a useful indication of sample environments (Figure 3). In general, factor loadings from the two data sets have similar trends with respect to SST. This is most clear for factors 1, 3, 5, and 6 (Figure 3).

Potential differences between the two data sets do exist. The

amplitude of the factor loadings for factors 2, 4, and 7 are diminished in the sediment trap assemblages relative to the core top assemblages. The dominant species in these factors are relatively rare in the sediment traps and abundant in some of the core top sediments. This could be a sampling problem or may reflect a bias in the sediments. All of the species involved have heavily calcified shells, suggesting that dissolution may enrich them in the sediments relative to their abundance in the

overlying water column. The model communality measures how much of the

information content of the foraminiferal faunas is explained by the factor model (Figure 3h). For both core top and sediment trap faunas the factor model explains adequate amounts of sample information at latitudinal extremes (SST <9øC or •20øC) but explains little information in the midlatitude faunas at SST from 9ø-15øC. Note, however, that this pattern is seen in the communalities of both the core top and sediment trap faunas, suggesting that a similar process contributes to the effect in both data sets. We explore how low sample communality may affect the SST estimates derived from the lmbrie-Kipp and modern analog methods in the discussion section.

Table 4. Coefficients for the Global Imbrie-Kipp Transfer Function

Term Coefficient

Intercept 27.6 Factor 3 -46.2 Factor 4 -26.2 Factor 3 squared 21.2 Factor 4 squared 18.6 Factor 5 squared -1.1 Factor 6 squared -3.3 Factor 7 squared 3.0 Factor I x factor 3 72.8 Factor I x factor 4 4.5 Factor I x factor 6 5.4 Factor 2 x factor 3 55.9 Factor 2 x factor 4 13.4 Factor 2 x factor 6 9.3 Factor 2 x factor 7 -3.8 Factor 3 x factor 4 21.4 Factor 3 x factor 5 20.0 Factor 3 x factor 6 10.6 Factor 4 x factor 6 -3.1 Factor 4 x factor 7 19.7 Factor 5 • factor 6 -6.0

Global Imbrie-Kipp Transfer Function

A statistically significant SST transfer function was developed using the core top data set. The terms for this global Imbrie-Kipp transfer function are listed in Table 4. This regression, based on 1121 samples, is significant at p<<0.01, has an r: of 0.93, and an RMS error of 1.9øC. The core top SST estimates have a slope of 0.93+0.02 and an intercept of 1.7+0.4 at the 95% confidence limit with respect to Levitus [1982] SST (Figure 4a). The slope of the core top regression is significantly different from one (with 95% confidence), given the sample size of 1121 core tops.

To compare the core top and sediment trap assemblages, we applied the core top transfer function to the sediment trap factor loadings (Figure 4, Table 5). When the transfer function is applied to the sediment trap samples, the SST estimates follow a slope of 0.92+0.16 and an intercept of 4.2+2.8 at the 95% confidence limit with respect to the Levitus [1982] SST values. The RMS error for the sediment trap SST estimates with this method was 2.6øC. The slope of the sediment trap relationship (0.92.•.16) is not significantly different from that of the core tops, or from one. However, as was the case for the core tops, the 95% confidence interval on the regression intercept indicates a warm bias in the sediment trap SST estimates (Figure 4b). A two-sided t-test of the 3.0øC average Imbrie Kipp-SST sediment trap residual demonstrates that this bias is significant at p<<0.01 (degrees of freedom (dr)=12, RMSE=2.6øC, t- value--4.18 > t-crit•0.01=3.055). The sediment trap SST estimates also appear to exhibit a residual trend as a function of the factor model communality. However, it is unlikely that this feature is statistically significant: The trend is not present in the core top SST residuals and occurs only at very low communality in the sediment trap residuals (Figure 4c).

34-30-26-22-18-14-10-6-2-

-2-,-2 2 6 10 14 18 22 26 30 34

5-

0-

-5-

-10-

ORTIZ AND MIX: SEDIMElff ThAP-CORE TOP COMPARISON 183

I I

-2 2 6 10 14 18 22 26 30 34Levitus SST (°C)

15-

10-

5-

0-

-5-

-10-

-15-, I I I I I I I I I

-2 2 6 10 14 18 22 26 30 34Levitus SST (°C)

15-

Levitus SST (°C) Levitus SST (°C)

15- 15-

- 1.0 0.8 0.6 0.4 0.2 0.0 -15-, . . .

Communality 0.0 0.1 0.2 0.3 0.4 0.5 0.6Sediment Traps o Core Tops Dissimilarity

Figure 4. Results of the global Imbrie-Kipp transfer U Sediment Traps .i Core Topsfunction. (a) Levitus [1982] sea surface temperature (SST)versus predicted SST. Diagonal lines mark a 1:1 relationship Figure 5. Results of the global modem analog method.(long dashed line), and least squares regressions for the core top Dashed and solid lines are as defmed in Figure 4. (a) Levitus(solid line) and sediment trap (short dashed line) samples. (b) [1982] SST versus predicted SST. (b) Levitus SST versusLevitus SST versus residual SST. (c) Residual SST versus factor residual SST. (c) Residual SST versus average samplemodel communality. Lower communalities denote weaker dissimilarity. Greater dissimilarity denotes less similarfactor model fit. Note reversed communality axis. samples.

34-30-26-

1D 22--C 18-0 14-

10-6-2-

-2-, , , , , , , ,

-2 2 6 10 14 18 22 26 30 34

a Imbrie-Kipp0.


34

30-

26--

22-

18-

14-

10-

6-

2-

-2

-2 2

a Imbrie-Kipp •, •.:.,.

•,•--•;-•, -" -• .,7.[? :i:?. ' •... - ß

ß '...-' •.'.."•&-2 -%'

e I I I I I I I I 6 10 14 1822263034

Levims SST (øC) 15

10

• ..,• ,.% f"::: ß - . ;.'".',. •.• (¾.,...:...•? :..-:}

=5- <"=" ..... 'ø:'"" "" •' • """'"'•"" ' ••''- ß ':- . .' L".O ;;., ,:.,., .... ,:0 "' .d.., o

-10-

-15 -2 2

I I I I I I I

6 10 14 1822263034

Levims SST (øC)

10

&• ..... .,.:, 5 ?•:"'%,.., '-"-:.•: '"•'

,-...--•:.... -.•, -•,..:-._'•. :.-. .... -5 <'..•:5..•.•..,:., .,...-'

"o'o o o -10

15 • • •-1.0 0.8 0.6 0.4 0.2

Communality Sediment Traps

0.0

Core Tops

Figure 4. Results of the global Imbrie-Kipp transfer function. (a)Lev/tus [1982] sea surface temperature (SST) versus predicted SST. Diagonal lines mark a 1:1 relationship (long dashed line), and least squares regressions for the core top (solid line) and sediment trap (short dashed line) samples. (b) Levims SST versus residual SST. (e) Residual SST versus factor model communality. Lower eommunalities denote weaker factor model fit. Note reversed communality axis.

34

30-

26-

22-

18-

14-

10-

6-

2-

-2

-2 2

a Modem Anal.•,g... • ,......,_•,....•/ ß .•::•'• •"•• •':• ....-.ff:'..{• ß ... ,--, •. ,.,,.•:...,.• ,,,' ....

..... r.'.'•.=.,$•' .., - 7-., ":'""

•'" ..L,:• ;a .4 '?"•

.4:•-v• .---•

I I I I I I I I

6 10 141822263034

Levims SST (øC)

Levims SST (øC)

15

[' 10-

E -10

15 • 0.0 0.1

I I I I

0.2 0.3 0.4 0.5 0.6

I

Dissimilarity

Sediment Traps .•:• Core Tops Figure 5. Results of the global modem analog method. Dashed and solid lines are as defined in Figure 4. (a) Levitus [1982] SST versus predicted SST. (b) Levims SST versus residual SST. (c) Residual SST versus average sample dissimilarity. Greater dissimilarity denotes less similar samples.


Table 5. Global Imbrie-Kipp and Modern Analog Temperature Estimates for the Sediment Trap Faunas

Modern Analog Results

We assessed potential bias in the modern analog SSTestimates in the same manner as with the Imbrie-Kipp transferfunction. The modem analog SST estimates for the sedimenttrap samples are listed in Table 5. We determined modemanalog SST for every sample in the core top database bycomparing each sample against all other samples in the data set(Figure 5a). These core top MAT SST estimates have an RMSerror of 1.5°C. The modem analog core top SST estimates havea slope of 0.98±0.01 and an intercept of 0.4±0.3 relative toLevitus [1982] SST.

When the core tops are used to estimate MAT SST values foreach of the sediment trap faunas, the resulting estimates havean RMS error of 2.2°C and follow a slope of 0.95±0.15 and anintercept of 1.9±2.5 relative to Levitus [1982]. The modemanalog method produced slopes and intercepts with nostatistically significant difference between the core top andsediment trap data sets. The intercepts for the modem analogSST estimates were smaller than those for the global Imbrie-Kipp transfer function (Fable 6). Unlike the result using theImbrie-Kipp method, a two-sided t-test of the 1.2°C averageMAT SST sediment trap residual demonstrates that this offset isnot statistically different from zero (df-12, RMSE = 2.2°C, t-value=1.89 < t-critt)0.05=2.179). Using the core top data set,the slope for the modem analog method was closer to unitythan the slope for the global Imbrie-Kipp transfer function.

Table 6. Statistical Comparison of the Two Methods

'Signifrant difference from 1 for slopes and from zero for intercepts at p <0.05."Significant difference fmm zero atp <<0.01.

Modem analog SST residuals displayed no significant trends asa function of Levitus SST (Figure Sb) or average modern analogdissimilarity (Figure 5c). l'his was true for both the core topand sediment trap modem analog SST estimates.

Discussion

These two methods of estimating paleoceanographictemperature have received considerable scrutiny over the pasttwo decades. However, not all of these studies have reached thesame conclusions regarding the relative applicability of thetwo methods. PreIl [1985] compared the two methods usingcore top and glacial maximum samples from the Atlantic,Indian, and Pacific basins. He concluded that they providedsimilar results within each basin and that transfer functionscalibrated for a specific basin worked best in that basin. Prell[1985] did not present results for a global comparison, nor didhis work include any modem sediment trap samples. Andersonat al. [1989] made downcore comparisons of the two methods inthe Coral Sea and also concluded the two methods providedsimilar results. Their findings accentuated the discrepanciesbetween marine SST estimates which suggest little cooling ofthe LGM low-latitude Pacific and terrestrial temperatureestimates which imply a much larger low-latitude thermaldecrease.

Working with a calibration data set composed of 499 Pacific

SiteAverage

DissimilaritySample

CommunalityActual

Temperature

ModemAnalog

Temperature

ModemAnalogResidual

hnbrie-KippTemperature

Imbrie-KippResidual

Gulf of Alaska 0.34 0.16 8.5 9.5 1.0 12.9 4.4Nearshore 0.34 0.70 12.4 12.7 03 13.8 1.4Midway 0.22 0.63 13.4 14.0 0.6 16.2 2.7Gyre 0.58 0.23 14.4 18.5 4.2 23.7 9.3San Pedm Basin 0.23 0.61 15.5 11.7 -3.8 15.4 -0.1Sargasso Sea 0.27 0.83 23.1 23.4 0.4 26.6 3.6Central Pacific 0.34 0.39 26.3 28.5 2.2 29.5 3.1Troica1 Atlantic 0.13 0.94 26.8 26.7 -0.1 29.6 2.8Panama Basin 0.11 0.97 26.9 27.5 0.6 24.6 -2.3MANOP Site C 0.06 0.97 26.3 28.1 1.8 28.9 2.6King George Basin 0.08 0.94 -0.6 2.0 2.6 2.9 3.5North Weddell Sea 0.04 0.95 -0.9 2.0 2.9 3.3 4.2Mend Rise 0.05 0.96 -0.5 1.7 2.2 3.4 3.9

Global Jmb,ie-Kipp 0.0 1.9 0.93±0A2' 1.7±0.4 3.0" 2.6 0.92±0.16 4.2±2.8kTransfer functionModem analog method 0.0 1.5 0.98±0.01 0.4±0.3 1.2 2.2 0.95±0.15 1.9±2.5Temperature estimate

Core top Faunal Comparison (N=1121) Sediment Trap Faunal Comparison (N=13)

Mean RMS Mean RAnalysis Type Residual Error Slope Intercept Residual Error Slope Intercept


Table $. Global lmbrie-Kipp and Modem Analog Temperature Estimates for the Sediment Trap Faunas

Modem Modem

Average Sample Actual Analog Analog Imbrie-Kii• Imbrie-Kipp Site Dissimilarity Communality Temperature Temperature Residual Temperature Residual

Gulf of Alaska 0.34 0.16 8.5 9.5 1.0 12.9 4.4 Nearshore 0.34 0.70 12.4 12.7 0.3 13.8 1.4

Midway 0.22 0.63 13.4 14.0 0.6 16.2 2.7 Gvre 0.58 0.23 14.4 18.5 4.2 23.7 9.3 San Pedro Basin 0.23 0.61 15.5 11.7 -3.8 15.4 -0.1

Sar•asso Sea 0.27 0.83 23.1 23.4 0.4 2&6 3.6 Central Pacific 0.34 0.39 26.3 28.5 2.2 29.5 3.1

Tropical Atlantic 0.13 0.94 26.8 26.7 -0.1 29.6 2.8 Panama Basin 0.11 0.97 26.9 27.5 0.6 24.6 -2.3 MANOP Site C 0.06 0.97 26.3 28.1 1.8 28.9 2.6

King George Basin 0.08 0.94 -0.6 2.0 2.6 2.9 3.5 North Weddell Sea 0.04 0.95 -0.9 2.0 2.9 3.3 4.2 Maud Rise 0.05 0.96 -0.5 1.7 2.2 3.4 3.9

Modern Analog Results

We assessed potential bias in the modern analog SST estimates in the same manner as with the Imbrie-Kipp transfer function. The modern analog SST estimates for the sediment trap samples are fisted in Table 5. We determined modem analog SST for every sample in the core top database by comparing each sample against all other samples in the data set (Figure 5a). These core top MAT SST estimates have an RMS error of 1.5øC. The modern analog core top SST estimates have a slope of 0.98+0.01 and an intercept of 0.4+0.3 relative to Levitus [1982] SST.

When the core tops are used to estimate MAT SgI' values for each of the sediment trap faunas, the resulting estimates have an RMS error of 2.2øC and follow a slope of 0.95+0.15 and an intercept of 1.9•_2.5 relative to Levitus [1982]. The modem analog method produced slopes and intercepts with no statistically significant difference between the core top and sediment trap data sets. The intercepts for the modern analog SST estimates were smaller than those for the global Imbrie- Kipp transfer function (Table 6). Unlike the result using the Imbrie-Kipp method, a two-sided t-test of the 1.2øC average MAT SST sediment trap residual demonstrates that this offset is not statistically different from zero (df=12, RMSE = 2.2øC, t- value=l.89 < t-orit•0.05=2.179). Using the core top data set, the slope for the modern analog method was closer to unity than the slope for the global Imbrie-Kipp transfer function.

Modem analog SST residuals displayed no significant trends as a function of Levims SST (Figure 5b) or average modem analog dissimilarity (Figure 5c). This was true for both the core top and sediment trap modem analog SST estimates.

Discussion

These two methods of estimating paleoceanographic temperature have received considerable scrutiny over the past two decades. However, not all of these studies have reached the

same conclusions regarding the relative applicability of the two methods. Prell [1985] compared the two methods using core top and glacial maximum samples from the Atlantic, Indian, and Pacific basins. He concluded that they provided similar results within each basin and that transfer functions

calibrated for a specific basin worked best in that basin. Prell [1985] did not present results for a global comparison, nor did his work include any modem sediment trap samples. Anderson et al. [ 1989] made downcore comparisons of the two methods in the Coral Sea and also concluded the two methods provided similar results. Their findings accentuated the discrepancies between marine SST estimates which suggest little cooling of the LGM low-latitude Pacific and terrestrial temperature estimates which imply a much larger low-latitude thermal decrease.

Working with a calibration data set composed of 499 Pacific

Table 6. Statistical Comparison of the Two Methods

Core top Faun• Comp•iso• (N= 1121)

Mean RMS

Analysis Type Residual Error Slope Intercept

Global Imbrie-Kipp 0.0 1.9 0.93_+0.02' 1.7_+0.4' Transfer function

Modem analog method 0.0 1.5 0.98_+0.01 0.4_+0.3 Temperature estimate

'Significant difference from 1 for slopes and from zero for intercepts at p < 0.05. bSignificant difference from zero at p << 0.01.

Sediment Trap Faunal Comparison (N= 13)

Mean RMS

Residual Error Slope Intercept

3.0 • 2.6 0.92__+0.16 4.2+__2.8'

1.2 2.2 0.95__+0.15 1.9+__2.5

core tops, Le [1992] compared these methods at two sites in thewestern Pacific. The primary conclusion of I.e [1992] was thatthe Imbrie-Kipp method provided consistent, reliable SSTestimates and that the modem analog method did not.Methodological differences between the study of [1992] andthose of PrelI [1985] and our study must be addressed. Le[1992] compared the two methods using 18 and 33 species formodem analog calculations and a subset of 24 species fortransfer function calculations. This introduces a secondvariable into the comparison, making it difficult to determine ifthe obtained results are fundamental or potentially related todifferences in the species lists. Le [1992] made use of a smallercalibration data set than Prell [1985] or this study and evaluatedthe methods by comparing downcore SST estimates from twocores in a relatively small, low-latitude region. The first ofthese factors decreases the range of hydrographic variationincluded in the calibration data set, while the second decreasesthe range of temperature variation over which the two methodswere assessed. While evaluating the temporal response of thetwo methods provides indications of their precision, it cannotassess their accuracy at reconstructing true SS1' variationbecause of the lack of a priori knowledge of the truepaleotemperatures. For these reasons, we employ global datasets of coretop and sediment trap faunas as a means of assessingboth the accuracy and precision of the two methods over theobserved global SST range.

Comparisons of Both Methods

To evaluate the temperature biases associated with themodem analog and global hnbrie-Kipp SST estimates, wecalculated simple linear regressions of actual SST againstestimated SST for both the coretop and sediment Irap data sets(Figures 4 and 5). Four regression statistics (slope, intercept,RMS error, and mean residual value) are summarized in Table 6.A slope significantly different from 1 indicates residual trendsin the SST estimates. A regression intercept or a mean residualvalue significantly different from zero indicates a constantoffset in the SST estimates relative to the actual SST if theslope of the regression is not significantly different fromunity. Larger RMS errors indicate greater random error in theSST estimates. Thus minimal temperature bias is displayedwhen slopes are close to 1, intercepts are close to zero, andRMS errors are small.

Temperature estimates derived from the coretop data set(N=1 121) using the modem analog method came closest to thisideal, with a slope of 0.98±0.01, an intercept of 0.4±0.3, andan RMS error of 1.5°C (Table 6). The variability associatedwith the slope and intercept of the regression are within the95% confidence interval of the statistic. Using the same coretop data set but estimating SST with the global Imbrie-Kipptransfer function method resulted in greater temperature bias.This can be seen by comparing the coretop statistics down eachcolumn in Table 6. The statistics for the coretop global Imbrie-Kipp transfer function yield a slope of 0.93±0.02, which issignificantly different from one, an intercept of 1.7±0.4, whichis significantly different from zero, and an RMS error of 1.9°C,almost 0.5°C larger than the RMS error for the modem analogmethod (Table 6).

Temperature bias in the Imbrie-Kipp method for the sedimenttrap faunas was expressed as a slope of 0.92±0.16, a nonzero


intercept of 4.2±2.8, and an RMS error of 2.6°C. The statisticsfor the modem analog method using the sediment trap data setwere a slope of 0.95±0.15, an intercept of 1.9±2.5. and anRidS error of 2.2°C (Table 6). The confidence intervalsassociated with the much smaller sediment trap data set (N=13)are wider than those of the larger, coretop data set (N=1 121).As a result, using the sediment traps, both methods producedSST estimates with slopes that were not significantly differentfrom unity relative to observed SST (with 95% confidence).However, using the small sample two-sided f-test, we

demonstrate that the mean residual value of the lmbrie-Kippmethod was significantly different from zero (with 99%confidence), while the mean residual value produced by themodern analog method was not significantly different fromzero. This was the same pattern observed for the regressionintercepts with the much larger core top data set.

The greatest potential systematic bias we observed usingeither the core top or sediment trap data was related to thedifferences in the intercepts of the actual versus estimated SSTregressions for the two methods. Using the core top data set,the Imbrie-Kipp method generated an intercept of 1.7±0.4°Crelative to Levitus [1982] SST. This value is 1.3°C warmer thanthe 0.4°C±0.3 intercept of the modern analog method for thecoretop data set (Table 6).

Because Imbcie-Kipp transfer functions are often developedfor specific oceanic regions, we generated two additional factormodels and Imbrie-Kipp transfer functions using (1) onlysamples from sites >8°C and (2) only samples from sites>20°C. The RMS error in the >8°C equation was 1.9°C, whilethat for the >20°C equation was 1.3°C. Despite RMS errorssmaller or equal to the global Imbrie-Kipp equation, bothregional equations had more significant, systematictemperature bias than the global relationship (Figure 6). Thisfmding suggests the need for great caution when regionallmbrie-Kipp transfer functions are employed. Extreme SSTerrors can occur if a downcore foraminiferal fauna was generatedwhen the true paleotemperature was outside the calibrationrange of the data set (i.e., during "no-analog" situations).Temperature estimates within the thermal bounds of the data setmay also be questionable due to the systematic residual bias.

The results in Figure 5, Table 5, and Table 6 indicate that themodem analog method provides relatively unbiased estimatesof SST over a range of almost 30°C. This result is particularlyimpressive when one considers the relatively poor quality ofthe core top analogs which were identified for many of thesediment trap assemblages. In 7 of the 13 cases, the averagedissimilarity coefficient was >0.20, the critical threshold wediscuss in the methods section. While we do not recommendrelaxing this constraint in downcore studies, the robustness ofthe method is demonstrated by the fact that even when pushedwell beyond the scope of its geologic application parameters,it still provided relatively accurate SSI' values for most of thesediment trap localities.

The two most severe SST errors in the sediment trapcomparison occurred for traps in the San Pedro Basin (-3.8°C)and at the Multitracers Gyre site (-i-4.2°C). Both of these sitesare located in the California Current region where coretops arerelatively scarce and calcite dissolution is heavy. At least someof the temperature error associated with these two traps mustarise from inadequacy of the coretop calibration data set. Whileboth of the errors described above are serious, similar

ORTIZ AND MIX: SEDIMENr TRAP-CORE TOP COMPARISON 185

core tops, Le [1992] compared these methods at two sites in the western Pacific. The primary conclusion of Le [1992] was that the Imbrie-Kipp method provided consistent, reliable $ST estimates and that the modem analog method did not. Methodological differences between the study of Le [ 1992] and those of Prell [1985] and our study must be addressed. Le [1992] compared the two methods using 18 and 33 species for modem analog calculations and a subset of 24 species for transfer function calculations. This introduces a second

variable into the comparison, making it difficult to determine if the obtained results are fundamental or potentially related to differences in the species lists. Le [ 1992] made use of a smaller calibration data set than Prell [ 1985] or this study and evaluated the methods by comparing downcore SST estimates from two cores in a relatively small, low-latitude region. The first of these factors decr•es the range of hydrographic variation included in the calibration data set, while the second decreases

the range of temperature variation over which the two methods were assessed. While evaluating the temporal response of the two methods provides indications of their precision, it cannot assess their accuracy at reconstructing true SST variation because of the lack of a priori knowledge of the true paleotemperatures. For these reasons, we employ global data sets of coretop and sediment trap faunas as a means of assessing both the accuracy and precision of the two methods over the observed global SST range.

Comparisons of Both Methods

To evaluate the temperature biases associated with the modem analog and global Imbrie-Kipp SST estimates, we calculated simple linear regressions of actual SST against estimated SST for both the coretop and sediment trap data sets (Figures 4 and 5). Four regression statistics (slope, intercept, RMS error, and mean residual value) are summarized in Table 6. A slope significantly different from 1 indicates residual trends in the SST estimates. A regression intercept or a mean residual value significantly different from zero indicates a constant offset in the SST estimates relative to the actual SST if the

slope of the regression is not significantly different from unity. Larger RMS errors indicate greater random error in the SST estimates. Thus minimal temperature bias is displayed when slopes are close to 1, intercepts are close to zero, and RMS errors are small.

Temperature estimates derived from the coretop data set (N=1121) using the modern analog method came closest to this ideal, with a slope of 0.98+0.01, an intercept of 0.4+0.3, and an RMS error of 1.5øC (Table 6). The variability associated with the slope and intercept of the regression are within the 95% confidence interval of the statistic. Using the same core top data set but estimating SST with the global Imbrie-Kipp transfer function method resulted in greater temperature bias. This can be seen by comparing the coretop statistics down each column in Table 6. The statistics for the coretop global Imbrie- Kipp transfer function yield a slope of 0.93+0.02, which is significantly different from one, an intercept of 1.7_+0.4, which is significantly different from zero, and an RMS error of 1.9øC, almost 0.5øC larger than the RMS error for the modern analog method (Table 6).

Temperature bias in the Imbrie-Kipp method for the sediment trap faunas was expressed as a slope of 0.92+0.16, a nonzero

intercept of 4.2_-t:2.8, and an RMS error of 2.6øC. The statistics for the modern analog method using the sediment trap data set were a slope of 0.95+0.15, an intercept of 1.9+2.5, and an RMS error of 2.2øC (Table 6). The confidence intervals associated with the much smaller sediment trap data set (N= 13) are wider than those of the larger, coretop data set (N=I 121). As a result, using the sediment traps, both methods produced SST estimates with slopes that were not significantly different from unity relative to observed SST (with 95% confidence). However, using the small sample two-sided t-test, we demonstrate that the mean residual value of the Imbrie-Kipp method was significantly different from zero (with 99% confidence), while the mean residual value produced by the modem analog method was not significantly different from zero. This was the same pattern observed for the regression intercepts with the much larger core top data set.

The greatest potential systematic bias we observed using either the core top or sediment trap data was related to the differences in the intercepts of the actual versus estimated SST regressions for the two methods. Using the core top data set, the Imbrie-Kipp method generated an intercept of 1.7+0.4øC relative to Levitus [1982] SST. This value is 1.3øC warmer than the 0.4øC+0.3 intercept of the modem analog method for the coretop data set (Table 6).

Because Imbrie-Kipp transfer functions are often developed for specific oceanic regions, we generated two additional factor models and Imbrie-Kipp transfer functions using (1) only samples from sites >8øC and (2) only samples from sites >20øC. The RMS error in the >8øC equation was 1.9øC, while that for the >20øC equation was 1.3øC. Despite RMS errors smaller or equal to the global Imbrie-Kipp equation, both regional equations had more significant, systematic temperature bias than the global relationship (Figure 6). This finding suggests the need for great caution when regional Imbrie-Kipp transfer functions are employed. Extreme SST errors can occur if a downcore foraminiferal fauna was generated when the true paleotemperature was outside the calibration range of the data set (i.e., during "no-analog" situations). Temperature estimates within the thermal bounds of the data set may also be questionable due to the systematic residual bias.

The results in Figure 5, Table 5, and Table 6 indicate that the modern analog method provides relatively unbiased estimates of SST over a range of almost 30øC. This result is particularly impressive when one considers the relatively poor quality of the core top analogs which were identified for many of the sediment trap assemblages. In 7 of the 13 cases, the average dissimilarity coefficient was >0.20, the critical threshold we discuss in the methods section. While we do not recommend

relaxing this constraint in downcore studies, the robustness of the method is demonstrated by the fact that even when pushed well beyond the scope of its geologic application parameters, it still provided relatively accurate SST values for most of the sediment trap localities.

The two most severe SST errors in the sediment trap comparison occurred for traps in the San Pedro Basin (-3.8øC) and at the Multitracers Gyre site (+4.2øC). Both of these sites are located in the California Current region where coretops are relatively scarce and calcite dissolution is heavy. At least some of the temperature error associated with these two traps must arise from inadequacy of the coretop calibration data set. While both of the errors described above are serious, similar

186 ORTIZ AND MIX: SEDIMENT ThAP-CORE TOP COMPARISON

30-

F-'Cl)c20-

10-

0

, , -10-, , ,26 30 34 -2 2 6 10 14 18 22 26 30 34

Levitus SST (°C) Levitus SST (°C)

15- 30

10- F-20

5-

::,1.0 0.8 0.6 0.4 0.2 0.0 1.0 0.8 0.6 0.4 0.2 0.0

Communality Communality

o Sediment Traps (in calibration range)

Sediment Traps (outside calibration range)

+ Core Tops (in calibration range)

Figure 6. Results of the culled Imbrie-Kipp transfer functions. Dashed and solid lines are as defmed in Figure4. Figures 6a, 6c, and 6e show transfer function based on all core tops >8°C. Figures 6b, 6d. and 6f showtransfer function based on all core tops >20°C. Note reversed communality axis.

34-30-26-22-18-14-10-6-2-

- _I I U U I I I U

-2 2 6 10 14 18 22 26 30 34Levitus SST (°C)

34-30-

Cl) 26-.l)rI 22-

18-14-

-ua) 10-

6-2-

-2-, I I I I I I I

15-

E- 10-C,,Cl)

C

0--u -5-C)

-15-i-2

,

2 6,

10 14 18,

22

-2 2 6 10 14 1822263034Levitus SST (°C)

0

S S IS

'S

h inihiie-Kipp, >20°CS :1

186 ORTIZ AND MIX: SEDIM• TRAP-CORE TOP COMPARISON

34 34 30 30J._ b Imbrie-Kipp, >20øC.

• .• ." . ,g'3 TM - q 14 q 14•....""•."

• 6 • 6T , , ' /

-2 2 6 1o 14 1822263o34 -2 2 6 lO 14 1822263o34 Levi•s SST (øC) Levitus SST (øC)

o -10-

-15

lO-C

2..-*.- ,o ß •': ,.,,,, ':'-b,•:'"" ,.-4'-. 5",,. '½',,•!•,•0 .-•..."-' ,*.'.{!4:' ß

4- '?,":-;•? '"• "•"' '"•- "'•- ß '.., '"'•?-? ' '::'"' ' ? 4'"*"'

ß -t-' 4- -3-' ,4.

I I I I I I I I

-2 2 6 10 14 1822263034

Levitus SST (øC)

30 •

d 20-

10-

0-

Levitus SST (øC)

15

lO i:.' - .... 0

5 -..:,.-. ::.-'t.,""•,•: :•'

-15 1.0 0.8 0.6 0.4 0.2 0.0

Communality

30 f

ca :20-

.• o o o e

-10

1.0 0.8 0.6 0.4 0.2 •.•

Communality

o Sediment Traps (in calibration range)

Sediment Traps (outside calibration range) Core Tops (in calibration range)

Figure 6. Results of the culled Imbrie-Kipp transfer functions. Dashed and solid lines are as defined in Figure 4. Figures 6a, 6c, and 6e show transfer function based on all core tops >8øC. Figures 6b, 6d, and 6f show transfer function based on all core tops >20øC. Note reversed communality axis.

12

10-

1 8-

4-

2-

-2

6-

0-

b

Slope of -0.4°C/year explains 45%of the >125 jim error variance.

0 1

I I I I I

I I I I I I

1 2 3 4 5 6 7

Trap deployment duration (years)

2 3 4 5 6 7


100 m sieve0 125 pmsieveo 150 j.tm sieve

Figure 7. Absolute magnitude of the sediment trap SSTresiduals for (a) the hnbrie-Kipp method and (b) the modemanalog method as a function of the sediment trap deploymentduration and seive size. Slope in Figure lb is based on a leastsquares regression of the >125 jim samples only.

situations are avoidable in downcore applications by carefulattention to the quality of the modem analogs as indicated bythe magnitude of the dissimilarity coefficient. In a sedimentapplication, erroneous SST estimates such as those at Gyre andthe San Pedro Basin would not be predicted if a criticaldissimilarity threshold of 0.20 were used as a cutoff criteria.

Assessing Potential Bias in the Sediment TrapData Set

We observed that the basic structure of the sediment trap andcore top faunas relative to SST were comparable by applyingthe coretop factor model to the sediment trap data set. Thegreatest differences between the two data sets occurred inmidlathude faunas which had relatively poor coznmunalities andlarge modern analog dissimilarities relative to coretops (Table5). These differences between the two data sets may occurbecause (1) the global factor model does not adequately resolveforazniniferal faunal variability in the midlatitudes, (2) delicate,soluble individuals in the sediment trap faunas are unlikely tobe weli preserved in the geologic record, or (3) the differencebetween the sediment trap and coretop data set is due to sievingartifacts and/or the duration of trap deployment.

We will address the third point before dealing with the othertwo possibilities. If the difference between the two data setswere due to artifact alone, we predict that the absolutemagnitude of the sediment trap residuals would increase as sievesize moved farther from the standard >150 pm sieve size andwould decrease with increasing trap duration. Perhapssurprisingly, the absolute SST residuals from both the Imbrie-Kipp and modern analog methods do not indicate anysystematic dependence on sieve size (Figure 7). Indeed the>100 jim sieved samples that were deployed for the shortestoverall duration exhibit smaller error than most of the >125 jimand >150 pm sieved samples. One explanation for this curiousresult is that these samples were collected from sediment trapsdeployed in tropical regions where small species are relativelyuncommon [e.g., Bradshaw, 1959]. A second example servesto illustrate this point further. Small species, particularly G.quinqueloba, are present in high relative abundance in the >125pm integrated sediment trap samples from the Gulf of Alaskaand the San Pedro basin. This accounts in large part for the lowcommunality and high dissimilarities of these two integratedsediment trap samples (Fable 5). While the extremedissimilarity and low communality for the Gulf of Alaskasample might indicate its temperature estimate should be verypoor, it actually produces a temperature estimate with less errorthan the San Pedro trap sample. The observation that coretopsfrom the vicinity of the Gulf of Alaska core are somewhat morecommon than those from the San Pedro Basin in the coretopdata set provides a plausible explanation. In short, sieve sizedoes not appear to exhibit any systematic effects on the SSTresiduals in the sediment trap data set.

Three key points can be made with respect to sediment trapduration. First, the Imbrie-Kipp method produced residuals thatwere independent of sediment trap duration (Figure 7a). Thisresult seems plausible, as the errors in the Imbrie-Kipptemperature regression are largely a function of the structure ofthe coretop faunas that determine the terms in the transferfunction. Additionally, the use of factor analysis acts to filterthe sediment trap observations of random variations. Incontrast, modem analog SST residuals tend to decrease withlonger trap teployments (Figure 7b). Again, such a trend seemsplausible. Longer integration times should result in sampleswith increasing similarity to fossil faunas. The trend accountsfor roughly 45% of the error variance in the seven >125 pmsieved samples. Interestingly, errors that are not statistically

ORTIZ AND MIX: SEDIMENT TRAP-CORE TOP COMPARISON 187ORTIZ AND MIX: SEDIMENT TRAP-CORE TOP COMPARISON 187

12

10-

a

I I I I I I

1 2 3 4 5 6 7


12

10-

[- < 4

• 2

0

-2

Slope of-0.4øC/year explains 45% of the >125 !xm error variance.

0 1 2 3 4 5 6 7


100 gm sieve

125 gm sieve

o 150 gm sieve

Figure 7. Absolute magnitude of the sediment trap SST residuals for (a) the Imbrie-Kipp method and (b) the modem analog method as a function of the sediment trap deployment duration and seive size. Slope in Figure 7b is based on a least squares regression of the >125 !xm samples only.

situations are avoidable in downcore applications by careful attention to the quality of the modem analogs as indicated by the magnitude of the dissimilarity coefficient. In a sediment application, erroneous SST estimates such as those at Gyre and the San Pedro Basin would not be predicted if a critical dissimilarity threshold of 0.20 were used as a cutoff criteria.

Assessing Potential Bias in the Sediment Trap Data Set

We observed that the basic structure of the sediment trap and core top faunas relative to SST were comparable by applying the coretop factor model to the sediment trap data set. The greatest differences between the two data sets occurred in midlatitude faunas which had relatively poor communalities and large modem analog dissimilarities relative to coretops (Table 5). These differences between the two data sets may occur because (1) the global factor model does not adequately resolve foraminiferal faunal variability in the midlatitudes, (2) delicate, soluble individuals in the sediment trap faunas are unlikely to be well preserved in the geologic record, or (3) the difference between the sediment trap and coretop data set is due to sieving artifacts and/or the duration of trap deployment.

We will address the third point before dealing with the other two possibilities. If the difference between the two data sets were due to artifact alone, we predict that the absolute magnitude of the sediment trap residuals would increase as sieve size moved farther from the standard >150 !lm sieve size and would decrease with increasing trap duration. Perhaps surprisingly, the absolute SST residuals from both the Imbrie- Kipp and modem analog methods do not indicate any systematic dependence on sieve size (Figure 7). Indeed the >100 pxn sieved samples that were deployed for the shortest

overall duration exhibit smaller error than most of the > 125 gm and >150 Ilm sieved samples. One explanation for this curious result is that these samples were collected from sediment traps deployed in tropical regions where small species are relatively uncommon [e.g., Bradshaw, 1959]. A second example serves to illustrate this point further. Small species, particularly G. quinqueloba, are present in high relative abundance in the > 125 lira integrated sediment trap samples from the Gulf of Alaska and the San Pedro basin. This accounts in large part for the low communality and high dissimilarities of these two integrated sediment trap samples (Table 5). While the extreme dissimilarity and low communality for the Gulf of Alaska sample might indicate its temperature estimate should be very poor, it actually produces a temperature estimate with less error than the San Pedro trap sample. The observation that coretops from the vicinity of the Gulf of Alaska core are somewhat more common than those from the San Pedro Basin in the coretop data set provides a plausible explanation. In short, sieve size does not appear to exhibit any systematic effects on the SST residuals in the sediment trap data set.

Three key points can be made with respect to sediment trap duration. First, the Imbrie-Kipp method produced residuals that were independent of sediment trap duration (Figure 7a). This result seems plausible, as the errors in the Imbrie-Kipp temperature regression are largely a function of the structure of the coretop farinas that determine the terms in the transfer function. Additionally, the use of factor analysis acts to filter the sediment trap observations of random variations. In contrast, modem analog SST residuals tend to decrease with longer trap •eployments (Figure 7b). Again, such a trend seems plausible. Longer integration times should result in samples with increasing similarity to fossil faunas. The trend accounts for roughly 45% of the error variance in the seven >125 sieved samples. Interestingly, errors that are not statistically


different from zero are achieved by the two sediment trap faunaswith deployment duration of 4-6 years.

This result casts doubt on the conventional wisdom thatargues sediment traps are not directly comparable to core topsbecause of their differences in integration time (months toyears in sediment traps versus decades to millennia in coretops). One way to explain this result is that variance at anannual cycle dominates living foraminiferal assemblages,while variance at interannual time scales is significantlysmaller. If so, a few years of sediment trap deployment wouldcapture the long-term mean assemblage with sufficientprecision for calibration with temperature data collected overthe last few decades, or for comparison with geologic samplesthat accumulated in the last few thousand years.

To summarize, sieve size does not appear to play a dominantfactor in determining the magnitude of the SST errors recordedby these sediment trap faunas. Sediment trap duration does notappear to contribute significantly to the Imbrie-Kipp SSTerrors but could account for up to 45% of the variance in themodem analog temperature estimates. Sediment trap records ofmore than 4 years duration appear sufficient to provide averageassemblages analogous to those in geologic samples. Becausesignificant sources of error between the sediment trap and coretop sediments remain, we explore the two remaining alternativehypotheses: (1) the global factor model does not adequatelyresolve foraminiferal faunal variability in the midlatitudes and(2) delicate, soluble individuals in the sediment trap faunas areunlikely to be well preserved in the geologic record.

The first possibility seems unlikely. For example, severalof the factors exhibit high factor loadings in the midlatitudetemperature range (Figure 3), suggesting model terms ofsignificance to these regions have been isolated. Dismissingthe first possibility leads us to conclude that the secondpossibility, dissolution, may play a role in the sediment trapversus core top differences. Although the results wereconsiderably noisy, the sediments appear to be enriched inrobust species which remain after the dissolution of morefragile species from the sediment trap faunas. Despite thesedifferences, the coretop calibration data set estimated accurateSST for most of the sediment trap faunas. In the final section,we explore the error at Gyre in closer detail as a means ofevaluating this potential dissolution bias in the sedimentrecord.

Assessing Potential Dissolution Bias in the CoreTop Data Set

The Gyre sediment trap is located in the California Current,650 km off the southern Oregon coast. On a seasonal basis,this site shifts from a summer/fall subtropical fauna dominatedby 0. wziversa and G. ruber to a diverse winter/spring fauna richin right-coiling N. pachyderma, N. dutertrei, G. gltainaa, G.quinqueloba, G. bulloides, and G. falconensis. The remainder ofthe species in this fauna include G. calida, T. humilis, and G.scitula [Ortiz and Mix, 1992]. The resulting flux-weightedannual average fauna is composed of 37% 0. universa and 13%G. rube,'.

In the underlying core top sediments of the northeastPacific, 0. universa and G. ruber accounts for 0-5% of theforaminiferal assemblage [Coulbourn et al. 1980]. An annuallyaveraged abundance of 37% for 0. universa exceeds

126 124

Predicted SST for 0% simulated dissolution


0 Predicted SST for 75% simulated dissolution


Figure 8. Effects of numerical dissolution scenarios onmodem analog temperature estimates from the Multitracerssediment traps. Error bars of ±1.5°C apply in all cases but forclarity are shown only for 0% and 75% simulations. See textfor further details.

expectations for its relative abundance in the underlying recentsediments by at least 32%. Moreover, the 37% abundance for0. universa exceeds the maximum abundance for this speciesrecorded in the entire global coretop data base by 19%. The13% abundance for G. ruber exceeds expectations for its relativeabundance in the underlying sediments by at least 8%. Becausethese two species are indicative of warmer waters than theremainder of the species in the Gyre fauna, and both arerelatively sensitive to dissolution, it seems likely theycontribute heavily to the 4.2°C warm temperature bias at theGyre site.

To evaluate the contribution of these two species to thisproblem, we recalculated modern analog SST for the threesediment traps in the Multitracers transect after "numericallydissolving" the sediment trap faunas (Figure 8). Weaccomplished this by making very simple assumptions of howdissolution might affect these samples. These assumptions arethat 0. universa and G. ruber would be removed at equal rates atall three sites and that no dissolution of other species wouldoccur. This provides a very simple scenario for evaluating theimpact of these two species on estimated SST while holding allother variables constant. Sensitivity studies were carried outon faunas with 50%, 75%, and 90% of the individuals of these

0-I I

134 132 130 128

Longitude (°W)

Levitus SST (°C)


different from zero are achieved by the two sediment trap faunas with deployment duration of 4-6 years.

This result casts doubt on the conventional wisdom that

argues sediment traps are not directly comparable to core tops because of their differences in integration time (months to years in sediment traps versus decades to millennia in core tops). One way to explain this result is that variance at an annual cycle dominates living foraminiferal assemblages, while variance at interannual time scales is significantly smaller. If so, a few years of sediment trap deployment would capture the long-term mean assemblage with sufficient precision for calibration with temperature data collected over the last few decades, or for comparison with geologic samples that accumulated in the last few thousand years.

To summarize, sieve size does not appear to play a dominant factor in determining the magnitude of the SST errors recorded by these sediment trap faunas. Sediment trap duration does not appear to contribute significantly to the Imbrie-Kipp SST errors but could account for up to 45% of the variance in the modem analog temperature estimates. Sediment trap records of more than 4 years duration appear sufficient to provide average assemblages analogous to those in geologic samples. Because significant sources of error between the sediment trap and core top sediments remain, we explore the two remaining alternative hypotheses: (1) the global factor model does not adequately resolve foraminiferal faunal variability in the midlatitudes and (2) delicate, soluble individuals in the sediment trap faunas are unlikely to be well preserved in the geologic record.

The first possibility seems unlikely. For example, several of the factors exhibit high factor loadings in the midlatitude temperature range (Figure 3), suggesting model terms of significance to these regions have been isolated. Dismissing the first possibility leads us to conclude that the second possibility, dissolution, may play a role in the sediment trap versus core top differences. Although the results were considerably noisy, the sediments appear to be enriched in robust species which remain after the dissolution of more fragile species from the sediment trap faunas. Despite these differences, the coretop calibration data set estimated accurate SST for most of the sediment trap faunas. In the final section, we explore the error at Gyre in closer detail as a means of evaluating this potential dissolution bias in the sediment record.

Assessing Potential Dissolution Bias in the Core Top Data Set

The Gyre sediment trap is located in the California Current, 650 km off the southern Oregon coast. On a seasonal basis, this site shifts from a summer/fall subtropical fauna dominated by O. universa and G. ruber to a diverse winter/spring fauna rich in right-coiling N. pachyderrna, N. dutertrei, G. glutinata, G. quinqueloba, G. bulloides, and G. falconensis. The remainder of the species in this fauna include G. calida, T. humilis, and G. scitula [Ortiz and Mix, 1992]. The resulting flux-weighted annual average fauna is composed of 37% O. universa and 13% G. ruber.

In the underlying core top sediments of the northeast Pacific, O. universa and G. ruber accounts for 0-5% of the foraminiferal assemblage [Coulbourn et al. 1980]. An annually averaged abundance of 37% for O. universa exceeds

21

18

• 15

12

9

O

'.... '.... ß ,. '...

ß .... -.., ß .. -...

ß .. %

•'** ,,o

i I I I

134 132 130 128 126 124

Longitude (øW)

Levitus SST (øC)

........ 1 ........ Predicted SST for 0% simulated dissolution

........ & ........ Predicted SST for 50% simulated dissolution

........ O ........ Predicted SST for 75% simulated dissolution

........ • ........ Predicted SST for 90% simulated dissolution

Figure 8. Effects of numerical dissolution scenarios on modem analog temperature estimates from the Multitracers sediment traps. Error bars of +1.5øC apply in all cases but for clarity are shown only for 0% and 75% simulations. See text for further details.

expectations for its relative abundance in the underlying recent sediments by at least 32%. Moreover, the 37% abundance for O. universa exceeds the maximum abundance for this species recorded in the entire global coretop data base by 19%. The 13% abundance for G. ruber exceeds expectations for its relative abundance in the underlying sediments by at least 8%. Because these two species are indicative of warmer waters than the remainder of the species in the Gyre fauna, and both are relatively sensitive to dissolution, it seems likely they contribute heavily to the 4.2øC warm temperature bias at the Gyre site.

To evaluate the contribution of these two species to this problem, we recalculated modem analog SST for the three sediment traps in the Multitracers transect after "numerically dissolving" the sediment trap faunas (Figure 8). We accomplished this by making very simple assumptions of how dissolution might affect these samples. These assumptions are that O. universa and G. ruber would be removed at equal rates at all three sites and that no dissolution of other species would occur. This provides a very simple scenario for evaluating the impact of these two species on estimated SST while holding all other variables constant. Sensitivity studies were carded out on faunas with 50%, 75%, and 90% of the individuals of these

species removed from the faunal list. The various speciespercentages were recalculated following each "dissolution" stepto preserve percent abundance closure, and then modern analogSST estimates were generated for the new fauna.

Removal of up to 50% of the individuals of these two speciesresults in SST estimates which overlap within errors with theinitial modem analog SST estimates for these three samples(Figure 8). Thus the modem analog method is quite robust tomoderate dissolution. Removal of 75% of the individuals ofthese two species from the Gyre fauna resulted in SST estimateswhich were similar to the historically recorded SST at that site.In the more coastal sites, removing greater than 50% of thesetwo species resulted in SST underestimates of l°-3°C. Thisresult suggests the Gyre site thermal bias derives from the highrelative abundance of 0. isniversa and G. ruber in the sedimenttraps at the Gyre site, a no-analog condition in the trapsrelative to the regional coretops due to the removal bydissolution of these fragile species from the fossil assemblage.Unfortunately, simply culling these species from the list usedin the calculation of sample dissimilarities does not solve thispreservation problem. We conducted additional experiments inwhich we excluded these two species from the taxonomic listand then recalculated modern analog SST. In the absence ofthese species from the dissimilarity function matrix, analogsfor slightly more northern latitudes are selected, and theresulting modern analog SST underestimates the true SST.

Conclusions

We reassessed the utility of the Imbrie-Kipp and modemanalog methods of estimating paleotemperature fromforaminiferal faunas. Our approach differs from previoussediment based calibration studies because we use global coretop faunas for calibration and sediment trap faunas forvalidation. Our results can be summarized as follows.

The basic structure of the sediment trap and coretop faunalassemblages are comparable. The greatest difference occurs inmidlatitude faunas where the sediment traps have poorcomxnunalities relative to coretops. These differences mayarise from the scarcity of coretops in midlatitude regions aswell as the presence of delicate, soluble forms in the sedimenttraps which are unlikely to be well preserved in sediments.Despite these differences, the coretop calibration data seteffectively estimated SST for most of the sediment trap faunas.

The modem analog method exhibited less systematic andrandom bias than the Imbrie-Kipp method over the full range ofglobal SST.

Regional Imbne-Kipp transfer functions exhibited greatersystematic bias and equal or smaller random bias than theglobal Imbrie-Kipp transfer function we developed.

Acknowledgments. We thank the captain and crew of R/V Wecomaand the Muhutracers sediment trap group for their efforts during theMuhitracers sediment trap croises. Early versions of the manuscriptwere improved by comments from N. Pisias, P. Wheeler, M. Abbott, andD. Birkes. We thank W. Prell, D. Andreasen, and two anonymousreviewers for their helpful reviews. Funding for this project wasprovided by a NASA Graduate student fellowship to the first author andby NSF funding to the Multitracers project. Curation of the Multitracers

sediment trap samples at the NORCOR Marine Geological Repositosy at

OSU was provided by a grant from the NSF. This is Lamont-DohertyEarth Observatosy contribution 5598.

ReferencesAnderson, D.M., W.L Prell, and NJ. Barratt, Estimates of sea surface

temperature in the coral sea at the last glacial maximum,Paleoceanography, 4, 615-627, 1989.

Beck, J.W., R.L Edwards, E. Ito, F.W. Taylor, I. Recy, F. Rougerie, P.Joanno and C. Heinin, Sea-surface temperature from coral skeletalstronsiwn/calcium ratios, Science, 257,644-647, 1992.

Bradshaw, I., Ecology of living planktonic foraminifera in the north andequatorial Pacific Oceans, Contrib. Cushman Found., Foraminfera1Rat., 10,25-64, 1959.

Broccoli, AJ., and E.P. Marciniak, Comparing simulated glacial climateand paleodata: A reexamination, Paleoceanography, 11, 3-14, 1996.

Broecker, W.S., Oxygen isotope constraints on surface oceantemperatures, Quat. Rat., 26, 121-134, 1986.

Climate: Long-Range Investigation, Prediction, and Mapping (CLIMAP)Project Members, The surface of the ice age Earth, Science, 191,1131-1137, 1976.

CLIMAP Project Members, Seasonal reconstructions of the earth'ssusface at the last glacial maximum, Geol. Soc. Am. Map Chart Ser..MC-36, 1-18, 1981.

Coulbown, W.T., FL, Parker, and W.H. Berger, Faunal and solutionpatterns of planktcnic foraminifera in surface sediments of the NorthPacific, Mar. Micropaleontol.. 5, 329-399, 1980.

Deuser, W.G., Seasonal variation in isotopic composition and deep-waterfluxes of the tests ci perennially abundant planktonic foraminifera ofthe Sargasso Sea: Results from sediment-trap collections and theirpaleoceanographic significance, I. Foraminferal Res., 17, 14-27,1987.

Deuser, W.G., and E.H. Ross, C., Seasonally abundant planktomcforaminifera of the Sargasso Sea: succession, deep-water fluxes,isotopic compositions, and paleoceanographic implications, I.Foranwsjferal Ram., 19, 268-293. 1989.

Donner, B., and G. Wefer, Flux and stable isotopic composition of!'leogloboquad.rina pachyderma and other planktonic foraminifers inthe Southern Ocean (Atlantic sector), Deep Sea Res., Pan 1, 41, 1733-1743, 1994.

Fairbanks R., and P.H. Wiebe, Fomaminiferal and chlorophyll maximum:Vertical distribution, seasonal succession, and paleoceanographicsignificance, Science, 209, 1524-1526, 1980.

Guilderson, T., R.G. Fairbanks, and J.L. Rubenstone, Tropicaltemperature variations since 20,000 years ago: Modulatinginterhemnispheric climate change. Science, 263, 663-665, 1994.

Hutaon, W.H., The Aguihas Current during the Late Pleistocene: Analysisof modem faunal analogs, Science, 207, 64-66, 1980.

Imbrie, 1., and N.G. Kipp, A new micmpaleontological method forquantitative paleoclimatology: Application to a Late PleistoceneCaribbean core, in The Late Cenozoic Glacial Ages, edited by K.Turekian, pp. 71-181, Yale Univ. Press, New Haven, Conn., 1971.

Kipp, N.G., New transfer function for estimating past seasurfaceconditions from sea level distribution of planktonic foraminifera in theNorth Atlantic, Geol. Soc. Am. Me,n.,v. 18, 3-41, 1976.

Le, J., Palaeotemperature estimation methods: Sensitivity test on twowestern equatorial pacific cores, Quaternary Science Reviews, 11,801-820, 1992.

Levitus, S., Climatological atlas of the world ocean. NOAA Prof. Pap. 13,173 pp. US. Ckwi. Print. Off., Washington, D.C. 1982.

Molfino, B., N.G. Kipp, and J. Morley, Comparison of foraminiferal,Coccolithophorid, and Radiolarian paleotemperature equations:Assemblage coherency and estimate concordancy, Qua:. Res., 17, 279-313, 1982.

Ortiz, J.D., and A.C. Mix, The spatial distribution and seasonal successionof planktonic foraminifera in the California Current off Oregon,September 1987- 1988, in Upwelling Systemr: Evolution Since the EarlyMiocene, edited by C.P. Summerhayes, W.L Prell, and K.C. Emeis,197-213, Geol. Soc. Spec. PubI.. 64, 1992.

Ortiz, J.D., A.C. Mix, and R.W. Collier, Environmental control of livingsymbiotic and asymbiotic planktonic foraminifera in the CaliforniaCurrent, Paleoceanography, 10, 987-1009, 1995.

ORTIZ AND MIX: SEDIMENT TRAP-CORE TOP COMPARISON 189ORTIZ AND MIX: SEDIMENT TRAP-CORE TOP COMPARISON 189

species removed from the faunal list. The various species percentages were recalculated following each "dissolution" step to preserve percent abundance closure, and then modern analog SST estimates were generated for the new fauna.

Removal of up to 50% of the individuals of these two species results in SST estimates which overlap within errors with the initial modern analog SST estimates for these three samples (Figure 8). Thus the modern analog method is quite robust to moderate dissolution. Removal of 75% of the individuals of

these two species from the Gyre fauna resulted in SST estimates which were similar to the historically recorded SST at that site. In the more coastal sites, removing greater than 50% of these two species resulted in SST underestimates of 1ø-3øC. This result suggests the Gyre site thermal bias derives from the high relative abundance of O. universa and G. ruber in the sediment

traps at the Gyre site, a no-analog condition in the traps relative to the regional coretops due to the removal by dissolution of these fragile species from the fossil assemblage. Unfortunately, simply culling these species from the list used in the calculation of sample dissimilarities does not solve this preservation problem. We conducted additional experiments in which we excluded these two species from the taxonomic list and then recalculated modern analog SST. In the absence of these species from the dissimilarity function matrix, analogs for slightly more northern latitudes are selected, and the resulting modem analog SST underestimates the true SST.

Conclusions

We reassessed the utility of the Imbrie-Kipp and modem analog methods of estimating palcotemperature from foraminiferal faunas. Om approach differs from previous sediment based calibration studies because we use global core top faunas for calibration and sediment trap faunas for validation. Our results can be summarized as follows.

1. The basic structure of the sediment trap and coretop faunal assemblages are comparable. The greatest difference occurs in midlatitude faunas where the sediment traps have poor communalities relative to coretops. These differences may arise from the scarcity of coretops in midlatitude regions as well as the presence of delicate, soluble forms in the sediment traps which are unlikely to be well preserved in sediments. Despite these differences, the coretop calibration data set effectively estimated SST for most of the sediment trap faunas.

2. The modern analog method exhibited less systematic and random bias than the Imbrie-Kipp method over the full range of global SST.

3. Regional Imbrie-Kipp transfer functions exhibited greater systematic bias and equal or smaller random bias than the global Imbrie-Kipp transfer function we developed.

Acknowledgments. We thank the captain and crew of R/V Wecoma and the Multitracers sediment trap group for their efforts during the Multitracers sediment trap cruises. Early versions of the manuscript were improved by comments from N. Pisias, P. Wheeler, M. Abbott, and D. Birkes. We thank W. Prell, D. Andreasen, and two anonymous

reviewers for their helpful reviews. Funding for this project was provided by a NASA Graduate student fellowship to the first author and by NSF funding to the Multitracers project. Curation of the Multitracers sediment trap samples at the NORCOR Marine Geological Repository at

OSU was provided by a grant from the NSF. This is Lamont-Doherty Earth Observatory contribution 5598.

References

Anderson, D.M., W.L Prell, and N.J. Barratt, Estimates of sea surface temperature in the coral sea at the last glacial maximum, Paleoceaaography, 4, 615-627, 1989.

Beck, LW., R.L Edwards, E. Ito, F.W. Taylor, $. Recy, F. Rougerie, P. $oannot, and C. Heinin, Sea-surface temperature from coral skeletal strontium/calcium ratios, Science, 257, 644-647, 1992.

Bradshaw, L, Ecology of living planktonic foraminifera in the north and equatorial Pacific Oceans, Coatrib. Cushman Found., Foraminiferal Res., 10, 25-64, 1959.

Broccoli, A.J., and E.P. Marciniak, Comparing simulated glacial climate and palcodata: A re. examination, Paleoceanography, 11, 3-14, 1996.

Broecker, W.S., Oxygen isotope constraints on surface ocean temperatures, Quat. Res., 26, 121-134, 1986.

Climate: Long-Range Investigation, Prediction, and Mapping (CLIMAP) Project Members, The surface of the ice age Earth, Science, 191, 1131-1137, 1976.

CLIMAP Project Members, Seasonal reconstructions of the earth's surface at the last glacial maximum, Geol. Soc. Am. Map Chart Ser., MC-36, 1-18, 1981.

Coulboum, W.T., F.L., Parker, and W.H. Berger, Faunal and solution pauems of planktonic foraminifera in surface sediments of the North Pacific, Mar. Micropaleontol., 5, 329-399, 1980.

I)euser, W.G., Seasonal variation in isotopic composition and deep-water fluxes of the tests of Perennially abundant planktonic foraminifera of the Sargasso Sea: Results from sediment-trap collections and their paleoceanographic significance, J. Foraminiferal Res., 17, 14-27, 1987.

Deuser, W.G., and E.H. Ross, C., Seasonally abundant planktonic foraminifera of the Sargasso Sea: succession, deep-water fluxes, isotopic compositions, and paleoceanographic implications, J. Foraminiferal Res., 19, 268-293, 1989.

Donner, B., and G. Wefer, Flux and stable isotopic composition of Neogloboquadrina pachyderma and other planktonic foraminifers in the Southern Ocean (Ariantic sector), Deep Sea Res., Part I, 41, 1733- 1743, 1994.

Fairbanks R., and P.H. Wiebe, Foraminiferal and chlorophyll maximum: Vertical distribution, seasonal succession, and paleoceanographic significance, Science, 209, 1524-1526, 1980.

Guilderson, T., R.G. Fairbanks, and J.L. Rubenstone, Tropical temperature variations since 20,000 years ago: Modulating interhemispheric climate change, Science, 263, 663-665, 1994.

Hutson, W.H., The Agulhas Current during the Late Pleistocene: Analysis of modem faunal analogs, Science, 207, 64-66, 1980.

Imbrie, J., and N.G. Kipp, A new micropaleontological method for quantitative paleoclimatology: Application to a Late Pleistocene Caribbean core, in The Late Cenozoic Glacial Ages, edited by K. Turekian, pp. 71-181, Yale Univ. Press, New Haven, Conn., 1971.

Kipp, N.G., New transfer function for estimating past seasurface conditions from sea level distribution of planktonic foraminifera in the North Atlantic, Geol. Soc. Am. Mem.,v. 18, 3-41, 1976.

Le, J., Palaeotemperature estimation methods: Sensitivity test on two western equatorial pacific cores, Quaternary Science Reviews, 801-820, 1992.

Levitus, S., Climatological arias of the world ocean, NOAA Prof. Pap. 13, 173 Pp. U.S. Govt. Print. Off., Washington, D.C. 1982.

Molfino, B., N.G. Kipp, and J. Morley, Comparison of foraminiferal, Coocotithophorid, and Radiolarian paleotemperature equations: Assemblage coherency and estimate concordancy, Quat. Res., 17, 279- 313, 1982.

Ortiz, LD., and A.C. Mix, The spatial distribution and seasonal succession of planktonic foraminifera in the California Current off Oregon, September 1987-1988, in Upwelling Systems: Evolution Since the Early Miocene, edited by C.P. Summerhayes, W.L. Prell, and K.C. Emeis, 197-213, Geol. Soc. Spec. Publ., 64, 1992.

Ortiz, LD., A.C. Mix, and R.W. Collier, Environmental control of living symbiotic and asymbiotic planktonic foraminifera in the California Current, Paleoceanography, 10, 987-1009, 1995.


Overpeck. J.T., T. Webb, and I.C. Prentice, Quantitative interpretation offossil pollen spectra: Dissimilarity coefficients and the method ofmodem analogs, Qual. Re:., 23, 87-108, 1985.

Parker, F.L, and W.L Berger, Faunal and solution pauerns of planktonicforaminifera in surface sediments of the South Pacific, Deep Sea.. 18,73-107, 1971.

Pflaumann, U., 1. Duprat, C. Pujol, and LD. Labeyne, SINMAX: Amodem analog technique to deduce Atlantic sea surface temperaturesfrom planktosuc foraminifera in deep-sea sediments,Paleoceanography, 11, 15-36, 1996.

Pitil, W.L, The stability of low-latitude sea-surface temperatures: Anevaluation of the CLIMAP reconstruction with emphasis on the positiveSST anomalies, Rep. TR025, 60 P., U.S. Dep. of Energy, WashingtonDC, 1985.

Prentice, I.C., Multidimensional sealing as a research tool in Quatemarypalynology: A review of theory and methods, Rev. Paleobot. andPa1,,iol., 31,71-104, 1980.

Rind, D., and D. Peteet, Terrestrial conditions at the Last GlacialMaximum and CLIMAP sea-surface temperature estimates: Are theyconsistent?, Qwjt. Re:., 24, 1-22, 1985.

Sauter, L, and It Thunell, Seasonal succession of the planktonicforasninifera: Results from a four year time-series sediment trapexperiment in the northern Pacific, J. Fora,ninjferoj Re:., 19, 253-267,1989.

Sauter, L, and Thunel, Planktornc foraminiferal response to upwellingand seasonal hydrographic conditions: Sediment trap results from San

Pedro basin, southem California, I. Foramintferal Res., 21, 347-363,1991.

Stott. L, and CM. Tang, Reassessment of foraminiferal-based tropicalsea surface 8"O paleutemperatures, Paleooceariog., 11, 37-56, 1996.

Thompson, P.R., Planktonic foraminifera in the western North Pacificduring the past 150,000 years: Comparison of modern and fossilassemblages. Palaeogeogr. Palaeoclimator. Palaeoecol., 35, 241-279,1981

Thunnell, ItC., and S. Honjo, Plankionic foraminiferal flux to the deepocean: Sediment trap results from the Tropical Atlantic and the CentralPacific, Mar. Geo., 40,237-253, 1981.

Thunnell, R.C., and LA. Reynolds, Sedimentation of planktocncfotaminifera: Seasonal changes in species flux in the panama Basin,Micropaleo., 30,243-262, 1984.

Watkins, I., A.C. Mix, and I. Wilson, Living planktonic foraminifera:Tracers of circulation and productivity regimes in the centralequatorial pacific, Deep Sea Re:., Part , in press, 19%.

A.C. Mix, College of Oceanic and Atmospheric Sciences, Oregon StateUniversity, Corvallis, OR 97331-5503. (email: [email protected])J.D. Ostiz, Lamont-Doheity Earth Observatory of Columbia University,Palisades, NY 10964. (email: joniz)ldeo.columbia.edu)

(Received November 29, 1995; revised September 16, 1996;accepted September 23, 1996.)

190

Overpeck, J.T., T. Webb, and I.C. Prentice, Quantitative interpretation of fossil pollen spectra: Dissimilarity coefficients and the method of modem analogs, Quat. Res., 23, 87-108, 1985.

Parker, F.L•, and W.L. Berger, Faunal and solution patterns of planktonic foraminifera in surface sediments of the South Pacific, Deep Sea., 18, 73-107, 1971.

Pflaumann, U., J. Duprat, C. Pujol, and L.D. Labeyrie, SINMAX: A modern analog technique to deduce Atlantic sea surface temperatures from planktonic foraminifera in deep-sea sediments, Paleoceanography, 11, 15-36, 1996.

Prell, W.L., The stability of low-latitude sea-surface temperatures: An evaluation of the CLIMAP reconstruction with emphasis on the positive SST anomalies, Rep. TR025, 60 P., U.S. Dep. of Energy, Washington DC, 1985.

Prentice, I.C., Multidimensional scaling as a research tool in Quaternary palynology: A review of theory and methods, Rev. Paleobot. and Palynol., 31, 71-104, 1980.

Rind, D., and D. Peteet, Terrestrial conditions at the Last Glacial Maximum and CLIMAP sea-surface temperature estimates: Are they consistent?, Quat. Res., 24, 1-22, 1985.

Sauter, L•, and P• Thunell, Seasonal succession of the planktonic foraminifera: Results from a four year time-series sediment trap experiment in the northern Pacific, J. Foraminiferal Res., 19, 253-267, 1989.

Sauter, L, and Thunel, Planktonic foraminiferal response to upwelling and seasonal hydrographic conditions: Sediment trap results from San

ORTIZ AND MIX: SEDIM• TRAP-CORE TOP COMPARISON

Pedro basin, southem California, J. Foraminiferal Res., 21, 347-363, 1991.

Stott, L., and C.M. Tang, Reassessment of foraminfferal-based tropical sea surface •5•O paleotemperatures, Paleooceanog., 11, 37-56, 1996.

Thompson, P.R., Planktonic foraminifera in the western North Pacific during the past 150,000 years: Comparison of modern and fossil assemblages, Palaeogeogr. Palaeoclimator. Palaeoecol., 35, 241-279, 1981

Thunnell, R•C., and S. Honjo, Planktonic foraminifera! flux to the deep ocean: Sediment trap results from the Tropical Atlantic and the Central Pacific, Mar. Geo., 40, 237- 253, 1981.

Thunnell, R.C., and L.A. Reynolds, Sedimentation of planktonic foraminifera: Seasonal changes in species flux in the panama Basin, Micropaleo., 30, 243-262, 1984.

Watkins, J., A.C. Mix, and J. Wilson, Living planktonic foraminifera: Tracers of circulation and productivity regimes in the central equatorial pacific, Deep Sea Res., Part II, in press, 1996.

A.C. Mix, College of Oceanic and Atmospheric Sciences, Oregon State University, Corvallis, OR 97331-5503. (email: [email protected]) J.D. Ortiz, Lamont-Doherty Earth Observatory of Columbia University, Palisades, NY 10964. (email: jortiz•ldeo.columbia.edu)

(Received November 29, 1995; revised September 16, 1996; accepted September 23, 1996.)

Comparison of Imbrie&Kipp Transfer Function and Modern ...

Documents

Transcript of Comparison of Imbrie&Kipp Transfer Function and Modern ...