top foraminiferal faunas

Transcription

1 PALEOCEANOGRAPHY, VOL. 12, NO.2, PAGES , APRIL 1997 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 top foraminiferal faunas J. D. Ortiz' and A. C. Mix College of Oceanic and Atmospheric Sciences, Oregon State University, Corvallis Abstract. We evaluate the the reliability of of statistical estimates of sea surface temperature (SST) derived from planktonic foraminiferal faunas using the modem modern analog method and the Imbrie-Kipp method. Global core top faunas provide a calibration data set, while modem modern sedimentrap 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 the Jmbrie-Kipp Imbrie-Kipp transfer function temperature estimates had an intercept 1.3øC 13 C warmer than modern modem analog estimates and 1.7øC 1.7 C warmer than recorded atlas SST. The RMS error for the core top data set using the the modern modem analog method (1.5 C) (1.5øC) was smaller than that of of the the Jmbrie- Imbrie- Kipp method (1.9øC). (1.9 C). SST errors for for the the sedimentrap 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 in the the global Imbrie- Kipp transfer function. Dissolution simulations with the the sedimentrap 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 the faunal fauna] structure of sedimentrap and core top assemblages are similar; (2) both methods estimate SST reliably for modem modern foraminifera] foraminiferal flux assemblages, but the modem modern analog method exhibits less bias; and (3) both methods are relatively robust to to samples with low communality but sensitive to to selective faunal dissolution. Introduction than modern. modem. This 2-3 C 2ø-3øC cooling was indicated by several types of microfossils [Molfino et al. 1982]. Because sea surface temperature (SST) provides an important Broecker [1986] compared these findings with planktonic climate diagnostic on a variety of spatial and temporal scales, foraminiferal 'O and concluded the ' O data was also we assess the reliability of of paleotemperature reconstructions reasonably consistent with a a 2-3 C 2ø-3øC cooling. Stott and and Tang derived from planktonic foraminiferal foraminifera] faunas. The two [1996] reached a similar conclusion in in their study of tropical techniques most commonly used to estimate paleotemperature 8180 measurements. The recent study of Broccoli and Marciniak from these microfossils are the transfer function method of [1996] suggests that point by point comparisons of of climate Imbrie and awl Kipp [1971] and the modern analog method of model output and observations reducesome of the low-latitude Hutson Huison [1980] as modified by Prell [1985]. Do these two SST discrepancy. Unfortunately, they could not make a clear methods provide consistent, unbiased paleotemperature determination as to the validity of one approach over the other. estimates? In contrast, results derived from coral SrjCa St/Ca paleo- A potential problem is most evident in in the the temperature thermometry [e.g., Beck et et al. 1992; Guilderson et al. 1994] dilemma of the low-latitude last glacial maximum (LGM) ocean and a variety of terrestrial based methods [e.g., Rind and Peteet, where microfossil paleotemperature reconstructions differ from 1985] suggesthat low-latitude glacial SST may have been as other methods by as much as 2-3 C 2ø-3øC and are inconsistent with much as 5-6 C 5ø-6øC cooler than modern. modem. Because of this apparent some climate model results [Rind and Peteet, 1985]. Statistical discrepancy, we reevaluate the reliability of statistically based reconstructions based on microfossils [e.g., CLIMAP, 1976, microfossil SST estimates derived from planktonic 1981] suggesthat low - latitude SST was at at most 2ø-3øC 2-3 C cooler foraminiferal faunas using the transfer function method of lmbrie Imbrie and Kipp [1971] and the modern analog method of Hutson [1980] as modified by Prell [1985]. Our analysis differs 'Now 1Now at Lamosn-Doherty Lamont-Doherty Earth Observatosy Observatory of Columbia University, from previous sediment based studies in that we estimate SST SST Palisades, New York. for sedimentrap assemblages caught in the water column as as well as for core tops. This test is is designed to assess whether Copyright 1997 by the American Geophysical Union. smoothing and modification involved with the generation of of the fossil record induces bias in foraminiferal faunal SST Paper number 96PA O883-83O5/976PA-O2878$l2.00 estimates /97/96PA

2 176 ORTIZ AND MIX: SEDIMENT TRAP-CORE TOP COMPARISON Methods Core Top and Sediment Trap Data Sets 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 Imbrie lmbrie and Kipp [1971]. These tests were conducted using core top foraminiferal faunas from the three major ocean basins. Unfortunately, validation of paleotemperature 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 foraminiferal forarniniferal shells from different times, and (4) the possibility that the statistical correlation to to SST observed in in the core top faunas is not present in the living faunas but rather is a secondary artifact, caused, for for example, by by the correlation of SST with other environmental variables. The first three problems may be expressed as 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 assessystematic 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 42 N 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 PrelI Prell [1985] and and selected samples from Parker and and Berger [1971], Coulbourn Ct et al. [1980], and Thompson [1981]. Samples in these data data sets sets were were screened to to exclude duplicates and samples with erroneous locations (A. E. E. Morey and A. C. C. Mix, personal communication, 1996). To account account for differences in in the taxonomy used by various workers, we employed a subset of 27 taxonomic species and 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 and arid white varieties; (2) G. sacculifer (total), which includes G. trilobus and and G. G. sacculifer; (3) G. menardii (total), which includes G. menardii, G. menardii fiexuosa flexuosa (=neoflexuosa), (=neofiexuosa), and G. tumida; 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 tow and and sedimentrap studies [Ortiz and Mix, 1992; Ortiz et et al. 1995]. The sediment trap data are of of known modem age age and 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 quantitativestimates of how differenthe sediment trap faunas are from the sedimentary faunas. Unlike the core top data set, the sedimentrap faunal counts are not all based on the >150 g.rn im size size fraction. fraction. Ideally, the sedimentrap sampleshould should be processed in a manner which is is o O O w 60 6O ,", 8 i,i + + C3 10 :::) 0 + 9' ' , 60 i 60 I I I 90 I I I I I I I I LONGITUDE 1U DE FIgure Figure Locations of the sedimentrap faunas used in 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

3 ORTIZ AND MIX: SEDIMENT TRAP-CORE TOP COMPARISON O O 90 9O 60 6O 6O 30 3O w a D 0 3O 30 3O 6O O LONGITUDE Figure FIgure Locations for for the the core core top top samples in in the the global core top data set. identical to that of the core top sediments. In our studies we we have measured both the and >150 [lxn p.m size fractions. This allows us to easily compare our samples to >150 pm [lxn sediments or >125 m pm sediment trap samples. We We recommend that future sediment trap and plankton tow studies follow this procedure or or focus on on the >150 pm [lxn size fraction. While variations in sieve size introduce some uncertainty in in the sedimentrap to to core top comparison, the the residual errors we 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 of error will be evaluated further in in the discussion to follow. Q-Mode Factor Analysis and Imbrle-Kipp 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 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. sets. This is accomplished by decomposing the core top data matrix (Uct) into a factor score matrix (Fat). (Fct), a factor loading matrix (Bct), and an error matrix (Ect): (Eci: Uct = Bt Fct Ect (1) Uct = Bet Fct + Ect (1) Uct, the core top percent abundance data, is is based on counts of n=27 planktonic foraminiferal taxa in N=1 N= coretops. Each core top fauna in in Uct Uct is is row row normalized so so that its sum of squares is unity. The elements of the factor score matrix, Fct, F describe m varimax rotated assemblages composed of of weighted contributions from each of the 27 taxa (see Table 3). A varimax rotation is applied to Fct so so that that the the resulting assemblages remain close to to the centroid of of the sample data. This rotation has the advantage of producing orthogonal assemblages with generally positive coefficients that are are more easily interpreted than an unrotated solution. The elements of Bet Bct describe the relative contribution of each variinax varimax assemblage to each core top sample. The model fit is described by its communality (the sum of squares of Bct Bet or the normalized vector length), which is a linear measure of of the fraction of information retained from each sample [Imbrie and Kipp, 1971]. A perfect model fit fit is achieved when communality goes to 1, 1, and Ect=O. Ect=0. We use the transpose of the core top factor score matrix (F?ct) (FTct) to evaluate the structure of the sediment trap data matrix (Ust) by determining a factor loading matrix (Bst) for the sedimentrap data set: U5 Ust FTct = Bst + Est Et (2) This assumes that the structure of the core top assemblages apply to the the sedimentrap faunas. if ff this assumption is not correct, the resulting sample communality will be low, and the elements of Est will be be nonzero. If If the distribution of the core top assemblages are controlled by their environment, then Bst t should produce distinct patterns when plotted against a controlling environmental variable such as SST [Imbrie and Kipp, 1971]. However, if if the fundamental structure of the sedimentrap faunas differs from that of of the core top faunas, then the distribution pattern of B5t Bst with respecto SST will not match that of Bet. Bct. The core top factor loadings in Bct were regressed against SST to develop a stepwise least least squares squares transfer function.

4 178 ORTIZ AND MIX: SEDIMENT TRAP-CORE TOP COMPARISON 0\ 00 0 ju.,., j";l. 'no 00 In In 00 CV) In N In in In N N C C N In N N AAAAAAAAAAAAA o o N 0 In N.'Cl 00 m N N N N N N N cfl N 0%'n N %00%N00 IO ON 00 NO in N c %D ifl C N In 0 In N N Ifl in in vi N 00 N N O 0' 0% In 00 In N In ON ' C In C Cfl In C C C) 0% u c,cn--. u U v-i '0 N 000% 0N m I 6 I J Following Imbrie lmbrie and Kipp [1971], we we include squared and and cross producterms tenns for each factor in the stepwise regression. These terms tms are are included included in in the the final regression if their partial F value exceeds the critical 5% significance threshold. Squared and and cross product terms are included so so that our results can be be compared with equations of the type type developed for for CLIMAP [1976, 1981] and because each factor may exhibit nonlinear, parabolic responses to to temperature and/or interactive effects [Imbrie [lmbrie and Kipp, 1971]. The regression coefficients from the global Imbrie-Kipp transfer function are used with the sediment trap factor loadings in B5t Bst to determine Imbrie-Kipp temperaturestimates estimates for each of the sedimentrap 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 detennined from its duration and deployment season. This procedure is is necessary because some of of the sediment trap deployments did not sample the entire annual cycle. Assigning an annual average temperature to a a subannual forarniniferal foraminiferal fauna would introduce an an unrealistic temperature bias to our comparisons. The Modern Analog Method The Modern Analog Method We calculate modern analog SST from the sediment trap faunas using the the core tops tops as as the the calibration data set. set. We then compare these SST estimates with measured SST values from Levitus [1982] at the sedimentrap locations. As a final test of the modem modern analog method, we estimated modem modern analog SST for each core top in the the database by comparison against every other core cop top in the the database. database. This global calculation is similar to the separate ocean basin calculations of of Prell 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 top data set. set. The modem analog method compares planktonic foraminiferal assemblages in samples with unknown environmental conditions (SST in 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 produced l:noduced by by similar suites of environmental conditions. The second assumption is is that SST is the environmental variable which determines variation in foraminiferal assemblages or is is correlated with environmental variables which determine foraminiferal variation. The first farst asswnpcion assumption holds well for most of 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. N. pachyderma. However, high southern latitudes are are cooler by by 1ø-2øC l -2 C than high northern latitudes. Because foraminiferal faunas attain essentially monospecific status well before extremely cold southern ocean SST values are reached, use use of of a global data set to to predict predict modem analog SST at at extremely high high latitudes can give erroneous results due ckn to to the averaging of SST from both hemispheres. To avoid this problem, we we follow the the standard

5 ORTIZ AND MIX: SEDIMENT TRAP-CORE TOP COMPARISON Table 2. Flux-Weighted Percent Abundance for the Sediment Trap Foraminiferal Faunas as Numbered in Table 1.g Map Locations wziversa G. conqlobatus G. ruber (total)" G. tenelus G. sacculifer (total)" S. dehiscens G. aequilaierahs G. calida G. builoides G. faclonensis G.digilata G. rubescer,.s oo G. quinqueloba Left-coiling N. pachyderma Right-coiling N. pachyderma N. dutertrei G. conglo,nerata G. hexagona o.o P. obliquiloculata 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 G. inflata Left-coiling G. truncatulinoides Right-coiling G. truncatu1iroides G. crassaformis G. hi,yuta G. scitiila G. menardii (total)" G. glutinata 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 asome rare species that were reported in the original sources are omitted here from the flux weighted percents because these taxa were not unifornily reported by all workers. These categories were essentialy treated as unidentified. bg. ruber (total) = G. ruber (white) and G. ruber (pink); G. saccu1fer (total) = G. sacculjfer and G. trilobus; G. nwnardii (total) = G. menardii, G. tumida, and G. menardii neoflexuosa.

6 180 ORTIZ AND MIX: SEDIMENT TRAP-CORE TOP COMPARESON COMPARISON Table Factor Scores for for the the Seven-Factor, Varimax Rotated,Q-Mode Factor Model Based on 1121 Core Tops Factor 1, Factor 6, G.rub.T.I T./ Factor 2. 2, Factor 3. 3, Factor 4. 4, Factor 5, G. bulij bull./ Factor 7. 7, Taxon Gsac. G vac. T. G. men. T. N.pac.L L. G.inf. N.dul. dut. G.glut. P.obliq. O. 0. w&iversa universa G. conglobatuso conglobahts G. ruber rube, (total) 0.90' G. tenelus tenalus G. sacculiœer sacculifer (total) ø 0.31' S. dehiscens dehircens G. aequilateralis aequiloierallc G. calma caliihi G. bulloides ' G.facloneans faclonensis G. di #ata digitata G. rubescerss rubescerts G. ½uincueloba quinqueloba Left-coiling N. pachyderma ' Right-coiling Ri at-coiling N. pachwlerma pachwie,,na N. dutertrei duterirci ' G. conglomerala conl lomerata G. hexagona P. obli uiloculata obliquioculata ' G.inflata i.n//ata ' Left-coilingG. truncatulinoides Right-coilingG. truncalulinoides truncatulinoides G. crassaformis G.hirsuta G.sciwia scitula G.,nanardii menardii (total) ø ' G. lut/nata glutinata ' 0.27 Infonnation Information per Factoff Factot 27.4% 12.1% % 8.4% 12.5% 14.6% 11.5% 5.1% Dominant Dnninant species. G. rube, ruber (total) = G. G. rube, ruber (white) and and G. G. rube, tuber (pink); (pink); G. G. saccu1fer sacculifer (total) (total) = = G. G. saculjfer saculifer and and G. G. trilobms; trilobus; G. menardii (Total) = G. menardii,,nenardii, G. twnida, turnida, and G. menardii neoflexuosa. 1ota1 Fotal information explained: 91.5%. procedure of of Prell Preli [1985] and compare high-latitude (northern) angle [Prentice, 1980; Overpeck et et al. al. 1985]. The use of the the southern hemisphere trap samples against (northern) southern squared chord distance tends to increase the importance of of rare hemisphere core top samples only. species and decrease the importance of of abundant ones so that Assessment of of the second assumption is more problematic. differences in in the abundant species alone do not dominate the For relatively small spatial scales and short timescales, dissimilarity estimate. Prell [1985] demonstrated that the absolute abundance assemblages of planktonic foraminifera squared chord distance worked well with foraminiferal faunas. from plankton tow and and seasonal sediment traps traps demonstrate Accordingly, we use the squared chord distance between the greater control by biological factors (food availability and target sample and each core top in the database as a measure of light) than temperature [Fairbanks [Fairbanlcs and Wiebe, 1980; Watkins intersample dissimilarity: et al. 1996; Ortiz and Mix, 1992; Ortiz et al. 1995]. Here we mrn focus on the relative abundance of of foraminiferal assemblages at 1t2 1/22 d=[p it -pit I large spatial scales, integrated over time scales of generally dij = [Pill/2 - PjI _ 1/2 2 1 (3) k=l k=1 longer than the annual cycle. Under such conditions, d11 foraminiferal core top assemblages exhibit relatively strong In this notation, dij is the squared chord distance between the ith Imbrie and PrelI, while p. p.1 the fractional relationships to SST [e.g., lmbrie and Kipp, 1971; Prell, and jth samples, while P ik and P jk represent 1985]. We wish to determine whether the correlation to SST is percentages of the kth species in samples i and j. j. The 27 also present in the global data data set set of of temporally averaged taxonomic categories used to to calculate the squared chord sediment trap faunas. distance are the same as as those used in in the the Q-mode factor Temperaturestimates estimates based on the modem modern analog method analysis described above (Table 3). Just as the the communality depend on averages of sample-by-sample comparisons. Hutson provides a quantitative estimate of how well a foraminiferal [1980] used the cosine theta angle as a measure of the fauna is explained by a factor factor model, the the average average sample sample multivariate distance (i.e., dissimilarity) between two dissimilarity provides a quantitativestimate of the similarity foraminiferal faunas. Empirical pollen studies suggesthat the between a target fauna and its its closest analogs. Empirical squared chord distance provides more reliable results than studiesuggest suggest that values of dij d, < 0.25 yield reliable analogs several other dissimilarity measures including the cosine theta (W. Prell, personal communication, 1996). We tested this rule role the fractional

7 ß ß. 0khZ ORTIZ AND M1X MIX: SEDIMENT TRAP-CORE TOP COMPARISON 181 Factor 1 (G. rub. T., G. sac: sac' 27.4%) Factor 2 (G. men. T.' T.: 12.1%) bo C - 0.E 0.75 r2 cl : ' o :c. I I '-'-... '..:':...'i¾:' rj ri I I I I I I Ii Ii Ii I I I I SST ( C) (øc) SST ( C) (øc) Factor 3 (N. pac. Left: Left' 8.4%) Factor4(G. inf.: 12.5%) Factor 4 (G. inf.' 12.5 %) bfj bo C I >:. -:-, t" ß I-,.' <'..,.:: ' g.!? ;::.:.:'q-..': ":.?' *::"::::"'":: i 'i , I I I I I I I ,,,, SST ( C) (øc) SST ( C) (øc) Factor 5 (N. dut.: dut.' 14.6%) Factor 6 (G. bull., G. glut.: glut.' 11.5%) oo '.; gy'-')2' '"."?.':,'::'i bf C C - f 0.75 e._..'.. 7:: '* 5 :-' "'-:"" ;'- "... " '. " ' ' '-.:' :..'.i:::::....i.:i - I ,...'. <'- ;-..! :::,,,,. i.:' : :.:.... :.:...'.'.'"..,,:.!:.;:;.-.,. :gg.;5.':.q C_._" : -")--'r -::...?... ': :' -..::;. : '{ "-.:... '...:. -',.: '..-.:'. i:: :,-.-..::::?., ::... ß ::.:,.!....,:::...!:' '_:. i?: iii,,,,i:" ::::.':; iiiii!' ß :.,y,....., _.:...:::::::::--:.: ',:. :'.::: :..:.: ::.:.:.'...,,.: ,,,,,, , -5 I I I I I I I SST ( C) (øc) t Factor 7 (P. obliq.: obliq.' 5.1%) E C g h 0o ::-?;.f".. ;.. I E ß - y,. ¾,:/. %... :'...::::'.'5,. -.i-..iff: i:::;:i::... :'::i:-:i:"' E L) '" $::. :..:. :::::::::::::::::::::::::::::::::::::::::::::::::::::::::.::::... -: ':5:i! i ; :: +,,.:.i:i: i: i!?:.:.::... - "--o:','-3'" O.25 - I I I I I I I 0.00-, I I i i i I* l I SST ( C) (øc) Communality :%. = ½ ::...,:. I I I I ;..:j..' :!!.!...,::. ':i SST ( C) (øc) SST ( C) (øc) I Sediment Traps C Core Tops Figure Factor Factor loadings loadings and and communalities communalities versus versus sea sea surface surface temperature temperature for for the the global global Q-mode Q-mode factor factor model described in in the text. Core tops are small open squares, while sedimentraps 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 thumby calculating d o for each sample in the core top data set in comparison with all others. While While values of of du d o are not normal in distribution, naturalog transformed values of d o ß are approximately log normal. We note that 97% of the faunas in tho the core top data set have an average d 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. W. L. L. Prell 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.

8 182 ORTIZ AND MDC: MIX: SEDIMENT TRAP-CORE TOP COMPARISON Using the values of dij as the the selection ctiteria, criteria, the SST SST Table Coefficients for the Global Imbrie-Kipp values from the five core tops least dissimilar to the target Transfer Function sample are averaged to estimate the modern analog SST for that sample. We We use use arithmetic, rather than weighted averages, Term Coefficient when calculating analog SST estimates. Tests we have conducted demonstrate that that SST averages weighted by the Intercept 27.6 individual dissimilarity estimates as proposed by Hiason Hutson Factor [1980] are not significantly different from arithmetic averages. Factor Factor 3 squared 21.2 We did not experiment with more sophisticated geographic Pflaumann Factor 4 squared 18.6 weighting methods [e.g., Pfiaumann et al. 1996]. Factor 5 squared -1.1 Factor 6 squared -3.3 Factor 7 squared 3.0 Results Factor 1 I x factor Factor 1 I x factor Q-Mode Factor Analysis Factor 1 I x factor Factor 2 x factor Seven factor assemblages together account for 91.5% of the Factor 2 x factor information in in the global core top data set. The addition of an Factor 2 x factor have Factor 2 x factor eighth factor assemblage would have increased the total Factor 3 x factor information explained by only 2.4% to 93.9%. Models with Factor 3 x factor few assemblages generated groupings that were ecologically Factor 3 x factor less distinct. Table 3 lists the factor scores for the seven Factor 4 x factor Factor 4 x factor assemblages we chose to retain. The seven factors group Factor 5 c factor 6 similar distributions -6.0 species with similar distributions into assemblages which individually account for 5-27% of of the the total information. These assemblages correspond roughly to oceanographic environments. For example, factor 1, which is dominated by G. rube, ruber and G. sacculifer, is important in warm, oligotrophic Imbrie-KIpp Global Imbrie-Kipp Transfer Function regions, while factor 3, composed of left-coiling N. pachyderma, pachyderina, is is indicative of of cold, high-latitudextremes. A statistically significant SST transfer function was To compare the sedimentrap and core top foraminiferal developed using the core top data set. The terms for for this global faunas, we apply the core top factor scores to the sediment trap Imbrie-Kipp lmbrie-kipp transfer function are listed in Table 4. This data and extract factor loadings for each sample. Plots of factor regression, based on 1121 samples, is significant at p<0.ol, p<<0.01, loading against S F SST from Levitus [1982] serve as a useful has an r: r2 of 0.93, and an RMS error of 1.9øC. 1.9 C. The core top SST indication of sample environments (Figure 3). 3). In general, estimates have a slope slope of of 0.93± and an intercept of factor loadings from the two data sets have similar trends with 1.7± at the 95% confidence limit with respect to to Levitus respect to SST. This is is most clear for factors 1, 3, 5, and 6 [1982] SST (Figure 4a). The slope of of the core top regression is (Figure 3). significantly different from one (with 95% confidence), given Potential differences between the two data sets do exist. The the sample size of 1121 core tops. amplitude of of the factor loadings for for factors 2, 2, 4, 4, and 7 are are To compare the core top and sediment trap assemblages, we diminished in the sedimentrap assemblages relative to the applied the core top transfer function to the sediment trap factor core top assemblages. The dominant species in these factors loadings (Figure 4, Table 5). When the transfer function is are relatively rare in the sediment traps and abundant in some of applied to the sedimentrap samples, the SST estimates follow the core top sediments. This could be be a sampling problem or a slope of 0.92± and and an an intercept of 4.2± at the 95% may reflect a bias in the sediments. All of the species involved confidence limit with respecto the Levitus [1982] SST values. have heavily calcified shells, suggesting that dissolution may The RMS error for the sedimentrap SST estimates with this enrich them in the sediments relative to their abundance in the method was 2.6øC. 2.6 C. The slope of the sediment trap relationship overlying water column. (0.92±0.16) ( ) is not significantly different from that of the core The model communality measures how much of the tops, or from one. However, as was the case for the core tops, information content of of the foraminiferal faunas is is explained by the 95% 95% confidence confidence interval interval on the regression intercept the factor model (Figure 3h). For both core top and sediment indicates a warm bias in the sediment trap SST estimates (Figure trap faunas the factor model explains adequate amounts of 4b). A two-sided t-test of of the the 3.0 C 3.0øC average Imbrie Kipp-SST sample information at latitudinal extremes (SST <9 C <9øC or >20 C) 20øC) sediment trap residual demonstrates that this bias is significant but explains little information in the the midlatitude faunas at at SST at pso.ol p<<0.01 (degrees of freedom (4t) (dr)=12, RMSE=2.6 C, RMSE=2.6øC, t- from 9ø-15øC C. Note, however, that this pattern is seen in the value value=4.18 > t-crit@0.01=3.055). t-crit 0.01=3.055). The sedimentrap SST SST communalities of both the core top and sediment trap faunas, estimates also appear to exhibit a residual trend as a function of suggesting that a similar process contributes to to the effect in the factor model communality. However, it is unlikely that this both data sets. We explore how low sample communality may feature is is statistically significant: The trend is is not present in affect the SST estimates derived from the the Imbrie-Kipp lmbrie-kipp and and the core top top SST SST residuals residuals and and occurs occurs only only at very low modem modern analog methods in the discussion section. communality in the sediment trap residuals (Figure 4c).

9 ß -. ;.'".',.. (¾.,...:...?:..-:}..,,.% f"::: =5- <"=" ;;.,...,:.,., 'ø:'"" ' """'"' ""...,:0 ''- ':-. "' '.' ß.d.., L".O o I ORTIZ AND MIX: SEDIMENT SEDIMElff ThAP-CORE TRAP-CORE TOP COMPARISON 183 I I I I I I I Levitus Levims SST ( C) (øc) &....,.:, 5? :"'%,.., '-"-:. : '" ', :... -,..:-._'. -., : <'.. :5.....,:.,.,...-' -10 "o'o o o a Modem 26- ß. :: ' " ':...-.ff:'..{ ß... 1D 22-,--,.,.,,. :...,.,,,' r.'.'.=.,$ '.., - 7-., ":'"" -C 18- '"..L,: ;a.4 '?" : -v ,, I, I, I, I I,, I, I I Levims Levitus SST ( C) (øc) , I I I I I I I I I a Imbrie-Kipp,.:., , -- ;-, -" -.,7.[? :i:?.'... - ß '...-' ß.'.." &-2 -%' , e I I I I I I I I Levitus Levims SST ( C) (øc) [' 10- E Levitus Levims SST ( C) (øc) , Anal.,g...,...,_,... I I I I Communality Sediment Traps o Core Tops Dissimilarity Sediment Traps Core Tops Dissimilarity U Sediment.i Figure 4. Results of the global Imbrie-Kipp transfer I Sediment Traps. : Core Tops function. (a)lev/tus 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. 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 defined in Figure 4. (a) Levitus (solid line) and sedimentrap (short dashed line) samples. (b) [1982] SST versus predicted SST. (b) Levitus Levims SST versus versus Levims Levitus SST versus residual SST. (e) (c) Residual SST versus factor residual SST. (c) Residual SST versus versus average sample model communality. Lower eommunalities communalities denote weaker dissimilarity. Greater dissimilarity denotes less similar factor model fit. Note reversed communality axis. samples.

10 184 ORTIZ AND MIX: SEDIMENT TRAP-CORE TOP COMPARISON Table 5. $. Global Imbrie-Kipp lmbrie-kipp and and Modern Modem Analog Temperature Estimates for the Sediment Trap Faunas Modem Modem Average Sample Actual Analog Analog hnbrie-kipp Imbrie-Kii Imbrie-Kipp Site Dissimilarity Communality Temperature Temperature Residual Temperature Residual Gulf of Alaska Nearshore Midway Gyre Gvre San Pedm Pedro Basin Sargasso Sar asso Sea &6 3.6 Central Pacific Troica1 Tropical Atlantic Panama Basin MANOP Site C King George Basin North Weddell Sea Mend Maud Rise Modern Analog Results We assessed potential bias in the the modern modern analog SST estimates in the same manner as with the Imbrie-Kipp transfer function. The modem modern analog SST estimates for the sediment trap samples are fisted listed in Table We determined modem analog SST for every sample in the core core top top database database by comparing each sample against all other samples in the data set (Figure 5a). These core top top MAT SST estimates have an an RMS error of 1.5 C. 1.5øC. The modem modern analog core top SST estimates have a slope of 0.98± and an intercept of 0.4± relative to to Levitus [1982] SST. When the core tops are used to estimate MAT SgI' SST values for each of the sedimentrap faunas, the resulting estimates have an RMS error of of 2.2øC 2.2 C and follow a slope of ±0.15 and an intercept of 1.9± _2.5 relative to to Levitus [1982]. The modem analog method produced slopes and intercepts with no statistically significant difference between the the core top and and sedimentrap data sets. The The intercepts for for the the modem modern analog SST estimates were smaller than those for the global Imbrie- Kipp transfer function (Fable (Table 6). Unlike the result using the Imbrie-Kipp method, a two-sided t-test of the 1.2 C 1.2øC average MAT SST sedimentrap residual demonstrates that this offset is not statistically different from zero (df-12, (df=12, RMSE = 2.2øC, 2.2 C, t- value=1.89 value=l.89 < t-orit 0.05=2.179). t-critt)0.05=2.179). Using the core top data set, the slope for the modem modern analog method was closer to unity than the slope for for the the global Imbrie-Kipp transfer function. Modem analog SST residuals displayed no no significant trends as a function of Levims Levitus SST (Figure 5b) Sb) or or average modem modern analog dissimilarity (Figure 5c). This l'his 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 over the the past two decades. However, not all of these studies have reached the same conclusions regarding the the relative applicability of of the two methods. PreIl 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 sedimentrap samples. Anderson at 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 Working with a calibration data set composed of 499 Pacific Table Statistical Comparison of the Two Methods Core top Faunal Faun Comparison Comp iso (N=1121) Sediment Trap Faunal Comparison (N=13) Mean RMS Mean RMS R Analysis Type Residual Error Slope Intercept Residual Error Slope Intercept Global Jmb,ie-Kipp Imbrie-Kipp ±0A2' 0.93_+0.02' 1.7± _+0.4' 3.0" ± ±2.8k ' Transfer function Modem analog method ± _ ± _ ± ± Temperature estimate 'Signifrant 'Significant difference from 1 for slopes and from zero for intercepts at p < <0.05. "Significant bsignificant difference from fmm zero at atp p <<0.01.

11 ORTIZ AND MIX: SEDIMENr SEDIMENT TRAP-CORE TOP COMPARISON 185 core tops, Le [1992] compared these methods at at two sites in in the intercept of 4.2_-t:2.8, 4.2±2.8, and an RMS error of of 2.6øC. 2.6 C. The statistics western Pacific. The primary conclusion of Le I.e [1992] was that for the modern modem analog method using the sedimentrap data set the Imbrie-Kipp method provided consistent, reliable SST $ST were a slope of , 0.95±0.15, an an intercept of 1.9± , and and an an estimates and that the modem analog method did not. RMS RidS error of 2.2 C 2.2øC (Table 6). The confidence intervals Methodological differences between the study of Le [[1992] and associated with the much smaller sediment trap data set (N=13) those of PrelI Prell [1985] and and our our study must must be be addressed. addressed. Le are wider than those of the larger, coretop data data set set (N=1 (N=I 121). [1992] compared the the two methods using 18 and species for As a result, using the sedimentraps, both methods produced modem analog calculations and and a subset of of 24 species for SST estimates with slopes that were not significantly different transfer function calculations. This introduces a second from unity relative to to observed SST SST (with 95% 95% confidence). variable into the comparison, making it difficult to to determine if However, using the small sample two-sided f-test, t-test, we the obtained results are fundamental or or potentially related to demonstrate that the the mean residual value of the the lmbrie-kipp Imbrie-Kipp differences in the species lists. Le [1992] [ made use of a smaller method was significantly different from zero (with 99% calibration data set than Prell [1985] [ or or this this study and and evaluated confidence), while the mean residual value produced by the the methods by comparing downcore SST estimates from two modern modem analog method was not significantly different from cores in a relatively small, low-latitude region. The first of zero. This was the same pattern observed for the regression these factors decr es decreases the range of hydrographic variation intercepts with the much larger core top data set. included in the calibration data set, while the second decreases The greatest potential systematic bias we we observed using the range of temperature variation over which the two methods either the core top or or sediment trap data was related to the were assessed. While evaluating the temporal response of the differences in the intercepts of the actual versus estimated SST two methods provides indications of of their precision, it cannot regressions for the two methods. Using the core top data set, assess their accuracy at at reconstructing true SS1' SST variation the Imbrie-Kipp method generated an intercept of 1.7±0.4 C øC because of of the the lack of a priori knowledge of the true relative to Levitus [1982] SST. This value is is 1.3 C 1.3øC warmer than paleotemperatures. For these reasons, we employ global data the 0.4øC C±0.3 intercept of the modern modem analog method for the sets of coretop and sedimentrap faunas as a means of assessing coretop data set (Table 6). both the accuracy and precision of of the two methods over the Because Imbrie-Kipp Imbcie-Kipp transfer functions are often developed observed global SST range. for specific oceanic regions, we generated two additional factor models and and Imbrie-Kipp transfer functions functions using (1) only samples from sites >8 C >8øC and (2) only samples from sites Comparisons of Both Methods >20 C. >20øC. The RMS error in the the >8 C >8øC equation was 1.9 C, 1.9øC, while To evaluate the the temperature biases associated with the that for the >20 C >20øC equation was was 1.3 C. 1.3øC. Despite RMS errors modem analog and global hnbrie-kipp Imbrie-Kipp SST estimates, we we smaller or equal to the global Imbrie-Kipp equation, both calculated simple linear regressions of actual SST against regional equations had more significant, systematic estimated SST for for both the coretop and sediment trap Irap data sets temperature bias bias than than the the global relationship (Figure 6). 6). This (Figures 4 and 5). Four regression statistics (slope, intercept, finding fmding suggests the need for great caution when regional RMS error, and mean residual value) are summarized in Table 6. lmbrie-kipp Imbrie-Kipp transfer functions are are employed. Extreme SST A slope significantly different from 1 indicates residual trends errors can occur if a downcore foraminiferal fauna was generated in the SST estimates. A regression intercept or a mean residual when the true paleotemperature was outside the calibration value significantly different from zero indicates a constant range of the data set set (i.e., during "no-analog" situations). offset in the SST estimates relative to the actual SST if the Temperature estimates within the thermal bounds of the data set slope of the regression is is not not significantly different from may also be questionable due to to the the systematic residual bias. unity. Larger RMS errors indicate greater random error in in the The results in Figure 5, Table 5, and Table 6 indicate that the SST estimates. Thus minimal temperature bias is displayed modern modem analog method provides relatively unbiased estimates when slopes are close to 1, intercepts are close to zero, and of SST over a range of of almost 30øC. 30 C. This result is particularly RMS errors are small. impressive when one considers the the relatively poor quality of Temperature estimates derived from the the coretop data set the core top analogs which were identified for for many of of the (N=1121) 121) using the the modern modem analog method came closest to to this sedimentrap assemblages. In 7 of the 13 cases, the average ideal, with a slope of of , 0.98±0.01, an intercept of , 0.4±0.3, and dissimilarity coefficient was >0.20, the critical threshold we an RMS error of 1.5 C 1.5øC (Table 6). The variability associated discuss in the methods section. While we do not recommend with the slope and intercept of the regression are within the relaxing this constraint in downcore studies, the robustness of 95% confidence interval of the statistic. Using the same core the method is demonstrated by the fact that even when pushed top data set but estimating SST with the global Imbrie-Kipp well beyond the scope of its geologic application parameters, transfer function method resulted in greater temperature bias. it still provided relatively accurate SSI' SST values for most of the This can be seen by comparing the coretop statistics down each sediment trap localities. column in Table 6. The The statistics for for the the coretop global Imbrie- The two most severe severe SST SST errors errors in the sediment trap Kipp transfer function yield a slope of of , 0.93±0.02, which is comparison occurred for traps in the San Pedro Basin (-3.8 C) (-3.8øC) significantly different from one, an intercept of 1.7±0.4, 1.7_+0.4, which and at the Multitracers Gyre site (+4.2øC). (-i-4.2 C). Both of of these sites is significantly different from zero, and an RMS error of of 1.9øC, 1.9 C, are located in in the California Current region where coretops are almost 0.5øC 0.5 C larger than the RMS error for the modern modem analog relatively scarce and calcite dissolution is heavy. At least some method (Table 6). of the temperaturerror error associated with these two traps must Temperature bias in the Imbrie-Kipp method for the sediment arise from inadequacy of the coretop calibration data set. While trap faunas was expressed as as a slope of of , 0.92±0.16, a nonzero both of the errors described above are serious, similar

12 186 ORTIZ AND MIX: SEDIM SEDIMENT TRAP-CORE ThAP-CORE TOP COMPARISON J._ b h inihiie-kipp, Imbrie-Kipp, >20 C Cl) 26- S >20øC. :1.l) 26- ri ".,g'3 TM u q a) 10- q 14...""." T,,' / , I I I I I I I - _I I U U I I I U o o lo o Levi s Levitus SST ( C) (øc) Levitus SST (øc) ( C) 15- C,, Cl) -u C) C o E- lo-c i 30- F-' Cl) c20-,, -10-,,, , Levitus SST ( C) (øc) Levitus SST ( C) (øc) f 10- lo F- i:.' ca : ::, -2, 2 6, 2..-*.-,o ß ':,.,,,, ':'-b, :'"",.-4'-. '½',,!, "-',*.'.{!4:' 5",,. ß 4- '?,":-;? '" - " "' "' -'.., ß '"'?-? ' '::'"' '? 4'"*"' ß -t-' ',4. I I I I I I I I ::.-'t.,"", : : ' Communality Communality o -..:,.-. Sediment Traps (in calibration range) Sediment Traps (outside calibration range) + Core Tops (in calibration range) Figure Results of of the the culled Imbrie-Kipp transfer functions. Dashed and solid lines are as defmed defined Figure 4. Figures 6a, 6c, 6c, and and 6e 6e show transfer function based on all core tops >8øC. >8 C. Figures 6b, 6d. 6d, and 6f show transfer function based on all core tops >20 C. >20øC. Note reversed communality axis d 'S o o o e S S I S 0

13 [- < a b I I I I I I Trap deployment duration (years) Slope of-0.4øc/year -0.4 C/year explains 45% of the >125 jim!xm error variance. I I I I I Trap deployment duration (years) gm m sieve 125 gm pmsieve o 150 j.tm gm sieve Figure 7. Absolute magnitude of of the sediment trap SST SST residuals for (a) the hnbrie-kipp Imbrie-Kipp method and (b) the the modem analog method as a function of the sedimentrap deployment duration and seive size. Slope in Figure 7b lb is based on a least squares regression of the >125 jim!xm samples only. situations are avoidable in downcore applications by by careful attention to to the the quality of the the modem analogs as as indicated by the magnitude of the dissimilarity coefficient. In a sediment application, erroneous SST estimates such as those at at Gyre and the San Pedro Basin would not not be predicted if if a critical dissimilarity threshold of 0.20 were used as a cutoff criteria. ORTIZ AND MIX: SEDIMENT TRAP-CORE TOP COMPARISON 187 Assessing Potential Bias in in the Sediment Trap Data Set We observed that the basic structure of the sedimentrap and core top faunas relative to SST were comparable by applying the coretop factor model to the the sediment trap data data set. set. The greatest differences between the two data sets sets occurred in midlatitude midlathude faunas which had relatively poor coznmunalities communalities and large modem modern analog dissimilarities relative to coretops (Table 5). These differences between the two data sets may occur occur because (1) the global factor model does not adequately resolve foraminiferal forazniniferal faunal variability in the midlatitudes, (2) delicate, soluble individuals in the sediment trap faunas are unlikely to be weli well preserved in the geologic record, or (3) the difference between the sedimentrap and coretop data set is is due to 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 sedimentrap residuals would increase as sieve size moved farther from the standard >150!lm pm sieve size and would decrease with increasing trap duration. Perhaps surprisingly, the absolute SST residuals from both the Imbrie- Kipp and modern modem analog methods do not indicate any systematic dependence on sieve size (Figure 7). Indeed the >100 jim pxn sieved samples that were deployed for the shortest overall duration exhibit smaller error than most of the >125 > jim gm and >150 Ilm pm 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 in the >125 > pm lira integrated sediment trap samples from the Gulf of Alaska and the San Pedro basin. This accounts in in large part for the low communality and high dissimilarities of these two integrated sediment trap samples (Fable (Table 5). While the extreme dissimilarity and low communality for 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 sedimentrap 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 faunas that determine the the terms in 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 teployments (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 in the seven >125 pm sieved samples. Interestingly, errors that are not statistically

14 188 ORTIZ AND MIX: SEDIMENT TRAP-CORE TOP COMPARISON different from zero are achieved by the two sedimentrap faunas 21 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 of their differences in in integration time (months to 18 years in sediment traps versus decades to to millennia in core tops). One way to explain this result is that variance at an '... '... ß,. '... annual cycle dominates living foraminiferal assemblages, ß , ß while variance at interannual time scales is significantly 15 ß.. % smaller. If so, so, a a few few years years of of sediment trap trap deployment would capture the long-term mean assemblage with sufficient '**,,o precision for calibration with temperature data collected over the last few decades, or or for comparison with geologic samples 12 that accumulated in the last few thousand years. To summarize, sieve size does not appear to to play a dominant factor in determining the magnitude of the the SST errors recorded by these sedimentrap faunas. Sedimentrap duration does not I appear to contribute significantly to the Imbrie-Kipp SST 9 i I I I errors but could account for for up up to to 45% 45% of of the variance in the temperature 0- modem analog estimates. Sediment trap records of Longitude ( W) (øw) more than 4 years duration appear sufficient to provide average to O ( C) assemblages analogous to those in geologic samples. Because Levitus SST (øc) significant sources of of error between the sediment trap and core Predicted SST for 0% simulated dissolution top sediments remain, we explore the two remaining alternative hypotheses: (1) the the global factor factor model does does not not adequately... &... Predicted SST for 50% simulated dissolution resolve foraminiferal faunal variability in the the midlatitudes and (2) delicate, soluble individuals in the sediment trap faunas are... 0O... Predicted SST for 75% simulated dissolution unlikely to be well preserved in the the geologic record. The first possibility seems unlikely. For example, several Predicted SST for 90% simulated dissolution of the factors exhibit high factor loadings in in the the midlatitude temperature range (Figure 3), suggesting model terms of Figure 8. Effects of of numerical dissolution scenarios on significance to these regions have been been isolated. Dismissing modem analog temperature estimates from the Multitracers the first possibility leads us to conclude that the second sedimentraps. Error bars of +1.5øC ±1.5 C apply in all cases but for for and possibility, dissolution, may play a role in the sediment trap clarity are shown only for 0% and 75% simulations. See text for further details. versus core top differences. Although the results were considerably noisy, the sediments appear to to be enriched in robust species which remain after the dissolution of more fragile species from the sediment trap trap faunas. faunas. Despite these expectations for its relative abundance in in the underlying recent differences, the coretop calibration data set estimated accurate of the trap In sediments by at least 32%. Moreover, the the 37% abundance for SST for most of the sediment trap faunas. In the final section, at in as 0. O. universa exceeds the maximum abundance for this species we explore the error at Gyre in closer detail as a means of bias the recorded in in the the entire global coretop data data base base by by 19%. The evaluating this potential dissolution bias in the sediment record. 13% abundance for G. ruber exceeds expectations for its relative abundance in the underlying sediments by by at least 8%. Because these two species are indicative of warmer waters than the Assessing Potential Dissolution Bias in the Core remainder of the species species in the Gyre fauna, and both are Top Data Set relatively sensitive to dissolution, it seems likely they The Gyre sedimentrap is is located in the California Current, contribute heavily to the 4.2 C 4.2øC warm temperature bias at the 650 km off the southern Oregon coast. On a seasonal basis, Gyre site. this site shifts from a summer/fall subtropical fauna dominated To evaluate the contribution of these two species to this by O. 0. wziversa universa and G. ruber to a diverse winter/spring fauna rich problem, we recalculated modem modern analog SST for for the three in right-coiling N. pachyderrna, pachyderma, N. N. dutertrei, G. G. glutinata, gltainaa, G. sedimentraps in the Multitracers transect after after "numerically quinqueloba, G. bulloides, and G. falconensis. The remainder of dissolving" the sediment trap faunas (Figure 8). We the species in this fauna include G. calida, T. humilis, and G. accomplished this by making very simple assumptions of how scitula [Ortiz and Mix, 1992]. The resulting flux-weighted dissolution might affect these samples. These assumptions are annual average fauna is composed of of 37% O. 0. universa and 13% that O. 0. universa and G. ruber would be removed at equal rates at G. rube,'. ruber. all three sites and that no dissolution of other species species would In the underlying core top sediments of the northeast occur. This provides a very simple scenario for evaluating the Pacific, O. 0. universa and G. ruber accounts for 0-5% of the impact of these two species on estimated SST while holding all foraminiferal assemblage [Coulbourn et et al. al. 1980]. An annually other variables constant. Sensitivity studies were carried carded out averaged abundance of 37% for 0. O. universa exceeds on faunas with 50%, 75%, and 90% of the individuals of these

15 ORTIZ AND MIX: SEDIMENT TRAP-CORE TOP COMPARISON 189 species removed from the faunal list. The various species OSU was provided by a grant from the NSF. This is is Lamont-Doherty percentages were recalculated following each "dissolution" step Earth Observatosy Observatory contribution to preserve percent abundance closure, and then modern analog SST estimates were generated for the new fauna. References Removal of up to 50% of the individuals of these two species results in SST estimates which overlap within errors with the Anderson, D.M., W.L Prell, and NJ. N.J. Barratt, Estimates of sea surface modem the initial modern analog SST estimates for these three samples temperature in the coral sea at the last glacial maximum, Paleoceanography, modem Paleoceaaography, 4, , (Figure 8). Thus the modern analog method is quite robust to Beck, LW., J.W., R.L Edwards, E. Ito, F.W. Taylor, I. $. Recy, F. F. Rougerie, P. P. moderate dissolution. Removal of 75% of the individuals of Joanno $oannot, and and C. C. Heinin, Sea-surface temperature from coral skeletal these two species from the Gyre fauna resulted in in SST estimates stronsiwn/calcium strontium/calcium ratios, Science, 257, , which were similar to the historically recorded SST at that site. Bradshaw, L, I., Ecology of of living planktonic foraminifera in in the north and Contrib. Foraminfera1 In the more coastal sites, removing greater than 50% of these equatorial Pacific Oceans, Coatrib. Cushman Found., Foraminiferal Rat., Res., 10,25-64, two species resulted in SST SST underestimates of l -3 C. 1ø-3øC. This Broccoli, A.J., AJ., and E.P. Marciniak, Comparing simulated glacial climate result suggests the Gyre site thermal bias derives from the high and paleodata: palcodata: A reexamination, re. examination, Paleoceanography, 11, 11, 3-14, relative abundance of O. 0. isniversa universa and G. ruber in the sediment Broecker, W.S., Oxygen isotope isotope constraints on surface ocean in Rat., traps at the Gyre site, a no-analog condition in the traps temperatures, Quat. Res., 26, , the to Climate: Long-Range Investigation, Prediction, and Mapping (CLIMAP) relative to the regional coretops due to the removal by Project Members, The surface of the ice age age Earth, Earth, Science, 191, dissolution of these fragile species from the fossil assemblage , Unfortunately, simply culling these species from the list used used CLIMAP Project Members, Seasonal reconstructions of the earth's of susface the last Ser.. in the calculation of sample dissimilarities does not solve this surface at the last glacial maximum, Geol. Soc. Am. Map Chart Ser., preservation problem. We conducted additional experiments in MC-36, 1-18, Coulboum, Coulbown, W.T., F.L., FL, Parker, and W.H. Berger, Faunal and solution which we excluded these two species from the taxonomic list patterns pauems of planktcnic planktonic foraminifera in in surface sediments of of the the North North and then recalculated modern analog SST. In the absence of Pacific, Mar. Micropaleontol.. Micropaleontol., 5, , these species from the dissimilarity function matrix, analogs I)euser, Deuser, W.G., Seasonal variation in isotopic composition and deep-water ci perennially for slightly more northern latitudes are selected, and the fluxes of the tests of Perennially abundant planktonic foraminifera of resulting modern modem analog SST underestimates the true the Sargasso Sea: Results from sediment-trap collections and their SST. paleoceanographic significance, I. Foraminferal Res., 17, 14-27, paleoceanographic significance, J. Foraminiferal Res., 17, 14-27, Deuser, W.G., and E.H. Ross, Ross, C., Seasonally abundant planktomc planktonic Conclusions foraminifera of the Sargasso Sea: succession, deep-water fluxes, the and isotopic compositions, and paleoceanographic implications, I. J. We reassessed the utility of the Imbrie-Kipp and modem Foranwsjferal Foraminiferal Ram., Res., 19, , analog methods of estimating paleotemperature palcotemperature from Donner, B., and G. Wefer, Flux and and stable stable isotopic composition of foraminiferal faunas. Om Our approach differs from previous!'leogloboquad.rina Neogloboquadrina pachyderma and other planktonic foraminifers in in (Atlantic Pan 1, sediment based calibration studies because we use global core the Southern Ocean (Ariantic sector), Deep Sea Res., Part I, 41, , top faunas for calibration and sedimentrap faunas for Fairbanks R., and P.H. Wiebe, Fomaminiferal Foraminiferal and chlorophyll maximum: validation. Our results can be summarized as follows. Vertical distribution, seasonal succession, and paleoceanographic 1. The basic structure of the sediment trap and coretop faunal significance, Science, 209, , assemblages are comparable. The greatest difference occurs in Guilderson, T., R.G. Fairbanks, and J.L. Rubenstone, Tropical midlatitude faunas where the sediment traps have poor temperature variations since 20,000 years ago: Modulating interhemnispheric comxnunalities interhemispheric climate change, change. Science, 263, , communalities relative to coretops. These differences may Hutaon, Hutson, W.H., The Aguihas Agulhas Current during the Late Pleistocene: Analysis arise from the scarcity of coretops in in midlatitude regions as of modem faunal analogs, Science, 207, 64-66, well as the presence of delicate, soluble forms in the sediment Imbrie, 1., J., and N.G. Kipp, A new new micmpaleontological micropaleontological method for be traps which are unlikely to be well preserved in sediments. quantitative paleoclimatology: Application to a Late Pleistocene set Caribbean core, in The Late Cenozoic Glacial Ages, edited by K. Despite these differences, the coretop calibration data set Turekian, pp , Yale Univ. Press, New Haven, Conn., effectively estimated SST for most of the sediment trap faunas. Kipp, N.G., New transfer function for estimating past seasurface 2. The modern modem analog method exhibited less systematic and conditions from sea level distribution of of planktonic foraminifera in the random bias than the Imbrie-Kipp method over the full range of North Atlantic, Geol. Soc. Am. Me,n.,v. Mem.,v. 18, 3-41, Le, J., J., Palaeotemperaturestimation estimation methods: Sensitivity test on two global SST. 11, Imbne-Kipp western equatorial pacific cores, Quaternary Science Reviews, 3. Regional Imbrie-Kipp transfer functions exhibited greater , systematic bias and equal or smaller random bias than the Levitus, S., Climatological atlas arias of the world ocean. ocean, NOAA Prof. Pap. 13, global Imbrie-Kipp transfer function we developed. 173 Pp. pp. US. U.S. Ckwi. Govt. Print. Off., Washington, D.C Molfino, B., N.G. Kipp, and J. J. Morley, Comparison of foraminiferal, We Coccolithophorid, Coocotithophorid, and Radiolarian paleotemperature equations: Acknowledgments. We thank the captain and crew of R/V Wecoma Assemblage coherency and estimate concordancy, Quat. Qua:. Res., 17, 279- and the Muhutracers Multitracer sediment trap group for their efforts during the 313, Multitracersediment Muhitracers trap croises. cruises. Early versions of the manuscript Ortiz, LD., J.D., and A.C. Mix, The spatial distribution and seasonal succession were improved by comments from N. N. Pisias, P. P. Wheeler, M. Abbott, and of planktonic foraminifera in the California Current off Oregon, W. D. and two 1988, Systemr: D. Birkes. We thank W. Prell, D. Andreasen, and two anonymous September , in Upwelling Systems: Evolution Since the Early W.L reviewers for their helpful reviews. Funding for this project was Miocene, edited by C.P. Summerhayes, W.L. Prell, and K.C. Emeis, was , Geol. Soc. Spec. Publ., PubI.. 64, provided by a NASA Graduate student fellowship to the first author and Ortiz, LD., J.D., A.C. Mix, and R.W. Collier, Environmental control of living by NSF funding to the Multitracers project. Curation of the Multitracers symbiotic and asymbiotic planktonic foraminifera in the California Repositosy 10, sediment trap samples at the NORCOR Marine Geological Repository at Current, Paleoceanography, 10, , 1995.

16 190 ORTIZ AND MIX: SEDIMENT SEDIM TRAP-CORE TOP COMPARISON Overpeck, Overpeck. J.T., T. Webb, and I.C. Prentice, Quantitative interpretation of fossil pollen spectra: Dissimilarity coefficients and and the method of modem analogs, Qual. Quat. Res., Re:., 23, , Parker, F.L, F.L, and W.L. Berger, Faunal and solution patterns pauerns of planktonic foraminifera in surface sediments of of the South Pacific, Deep Sea., Sea.. 18, , Pflaumann, U., J. 1. Duprat, C. Pujol, and L.D. LD. Labeyne, Labeyrie, SINMAX: A modem modern analog technique to deduce Atlantic sea surface temperatures from planktosuc planktonic foraminifera in deep-sea sediments, Paleoceanography, 11, 15-36, Pitil, Prell, W.L, W.L., The The stability stability of 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, Prentice, I.C., Multidimensional scaling sealing as as a a research tool in in Quaternary Quatemary palynology: A review of theory and methods, Rev. Paleobot. and and Pa1,,iol., Palynol., 31,71-104, Rind, D., and D. Peteet, Terrestrial conditions at the Last Glacial Maximum and CLIMAP sea-surface temperaturestimates: estimates: Are they consistent?, Quat. Qwjt. Res., Re:., 24, 1-22, Sauter, L, L, and and It P Thunell, Thunell, Seasonal Seasonal succession succession of of the the planktonic forasninifera: foraminifera: Results from from a four year time-series sediment trap experiment in the northern Pacific, J. J. Foraminiferal Fora,ninjferoj Res., Re:., 19, , Sauter, L, and Thunel, Planktornc Planktonic foraminiferal response to upwelling and seasonal hydrographic conditions: Sedimentrap results from San San Pedro basin, southem California, J. I. Foraminiferal Foramintferal Res., 21, , Stott, Stott. L., L, and CM. C.M. Tang, Reassessment of foraminiferal-based foraminfferal-based tropical sea surface 5 O 8"O paleutemperatures, paleotemperatures, Paleooceariog., Paleooceanog., 11, 37-56, Thompson, P.R., Planktonic foraminifera in in the western North Pacific during the past 150,000 years: Comparison of modern and fossil fossil assemblages, assemblages. Palaeogeogr. Palaeoclimator. Palaeoecol., 35, , 1981 Thunnell, R C., ItC., and S. S. Honjo, Plankionic Planktonic foraminiferal foraminifera! flux to the deep ocean: Sediment trap results from the Tropical Atlantic and the Central Pacific, Mar. Geo., 40, , 253, Thunnell, R.C., and LA. L.A. Reynolds, Reynolds, Sedimentation of planktocnc planktonic fotaminifera: foraminifera: Seasonal changes in species flux in the panama Basin, Micropaleo., 30, , Watkins, J., I., A.C. Mix, and I. J. Wilson, Living planktonic foraminifera: Tracers of of circulation and and productivity regimes in in the central equatorial pacific, Deep Sea Res., Re:., Part II,, in press, 19% A.C. Mix, College of Oceanic and Atmospheric Sciences, Oregon State University, Corvallis, OR ( amix@oce.orsat.edu) J.D. Ostiz, Ortiz, Lamont-Doheity Lamont-Doherty Earth Observatory of of Columbia University, Palisades, NY ( jortiz ldeo.columbia.edu) joniz)ldeo.columbia.edu) (Received November 29, 1995; revised September 16, 1996; accepted September 23, 1996.)