Reassessment of CLIMAP Methods for Estimating Quaternary Sea­ Surface Temperatures: Examination Using Pacific Coretop Data Sets

Quantitative analyses of planktonic foraminifer faunal data have been applied to reconstruct the past conditions of surface oceans and to verify the results of climate modeling. In the present study, planktonic foramini­ fer faunal SST (sea-surface temperature) estimates were evaluated using a calibration set and a test set of newly compiled coretop data from the low­ latitude Pacific. A standard CLIMAP-type transfer function based on the IKM (lmbrie-Kipp Method) was developed in estimating SST. Compari­ sons between the SST function and the depth of thermocline (DOT) trans­ fer function developed on the basis of the same calibration coretop data indicated that the correlation between the planktonic foraminiferal abun­ dance distribution and DOT is more significant than that with SST. This comparison suggests that the DOT effect is a more important environmen­ tal control on faunal distributions and abundances. After evaluating the functions with a test set of coretop data, the residuals of SST estimates (�ST = estimated -observed SST) were compared with two surface ocean modes which were derived statistically using a principle component analysis of seasonal SST and DOT data of the low-latitude Pacific, as well as an index of carbonate preservation (CPI). The analyses of residuals clearly indi­ cated that the patterns of estimation bias are correlated significantly with the two ocean modes, with a tendency to yield colder estimates for high SST values, and warmer estimates for low SST values, and with a maximum uncertainty around 3° to 4°C. These results also revealed that the carbon­ ate preservation effect may not produce systematic biases. This reevalua­ tion raises questions about the accuracy of faunal SST estimates that are based on the commonly used quantitative techniques, and implies that the CLIMAP low-latitude Pacific SST pattern should be reexamined. From these analyses we suggest that reconstructing DOT or surface ocean modes from planktonic foraminifer faunal data would be more appropriate in fu­ ture paleoceanographic studies.

calibration set and a test set of newly compiled coretop data from the low latitude Pacific. A standard CLIMAP-type transfer function based on the IKM (lmbrie-Kipp Method) was developed in estimating SST. Compari sons between the SST function and the depth of thermocline (DOT) trans fer function developed on the basis of the same calibration coretop data indicated that the correlation between the planktonic foraminiferal abun dance distribution and DOT is more significant than that with SST. This comparison suggests that the DOT effect is a more important environmen tal control on faunal distributions and abundances. After evaluating the functions with a test set of coretop data, the residuals of SST estimates (�ST = estimated -observed SST) were compared with two surface ocean modes which were derived statistically using a principle component analysis of seasonal SST and DOT data of the low-latitude Pacific, as well as an index of carbonate preservation (CPI). The analyses of residuals clearly indi cated that the patterns of estimation bias are correlated significantly with the two ocean modes, with a tendency to yield colder estimates for high SST values, and warmer estimates for low SST values, and with a maximum uncertainty around 3° to 4°C. These results also revealed that the carbon ate preservation effect may not produce systematic biases. This reevalua tion raises questions about the accuracy of faunal SST estimates that are based on the commonly used quantitative techniques, and implies that the CLIMAP low-latitude Pacific SST pattern should be reexamined. From these analyses we suggest that reconstructing DOT or surface ocean modes from planktonic foraminifer faunal data would be more appropriate in fu ture paleoceanographic studies.

INTRODUCTION
Retrieval of interpretable climatic signals from marine sedimentary records is important for understanding the causes of past ocean-climate changes. Among the various paleoceanographic proxy indices, the faunal composition of fo ssil planktonic foraminifers is widely used for reconstructing past sea-surface temperature (SST), salinity, and other environ mental variables. Examples of such reconstructions were made by the CLIMAP ( 1981) for the Last Glacial Maximum (LGM) for surface ocean climate and by the SPECMAP (lmbrie et al., 1989) for the latest Quaternary climatic time series in the Atlantic Ocean.
Such reconstructions of past ocean variability rely on the statistical relationships between coretop faunal compositions and upper-layer ocean conditions. Two well-known techniques exist for extracting quantitative estimates of surface ocean conditions from faunal composi tion data: the Imbrie-Kipp Method (IKM) (lmbrie and Kipp, 197 1;Klovan and lmbrie, 1971;Imbrie et al. , 1973;Kipp, 1976) and the Modern Analog Technique (MAT) (Hutson, 1979;Overpeck et al., 1985;Prell, 1985). The IKM uses fac tor assemblages of modern faunas and multiple regression analyses to calibrate observed environmental variables such as sea-surface temperature (SST) to derive a set of paleoecological equations (transfer functions) for paleoestimation. The MAT computes a dissimilarity coefficient between past and modern faunal data in order to identify the most similar subset of modern analogues. Environmental estimates for past faunas are then obtained by averaging data from the observed environments of the modern analogues. Empirical comparisons on the efficiency of a number of dissimilar ity coefficients (Overpeck et al., 1985;Prell, 1985) suggest that the squared chord distance is the most suitable coefficient for paleoecological studies. The validity of these methods of paleoestimation is based on several assumptions (for IKM: see Imbrie and Kipp, 1971;Sachs et al., 1977;lmbrie and Webb, 1981; for MAT: see Overpeck et al., 1985;Prell, 1985) and has been tested in multibiotic compari sons (Molfino et al., 1982) with statistical confirmation (Molfino, 1993).
The climate pattern reconstructed on the basis of quantitative fau na! analyses provides an independent verification for climate models which simulate past climates. LGM climate simu lations using the General Circulation Model (Gates, 1976a;Manabe and Hahn, 1977) or the newly developed Community Climate Model (Kutzbach and Guetter, 1986) using prescribed CLIMAP SSTs have shown that the ice age SST had a substantial impact on global climate. Alternative models (Manabe and Broccoli, 1985a; which used prescribed LGM boundary conditions to predict ice age SST revealed a significant discrepancy between the computed and CLIMAP SST patterns, especially in the low-latitude oceans. This discrepancy, along with the possibility of overestimation of CLIMAP SST based on terrestrial evidence (Webster and Streten, 1978), led Rind and Peteet ( 1985) to lower the CLIMAP SST by 2°C to balance the radiation budget in their ice age climate simulati ons. This example illustrates the need for accurate reconstruction of SST to verify the results from climate models from which the rela tive importance of the various fac tors which contribute to past climate changes can be evalu ated. Although initial assessments (Prell, 1985;Broecker, 1986;Anderson et al., 1989) sup ported the CLIMAP SST estimates, many recent data/model (Lautenschlager et al., 1992;Hoffe rt and Covey, 1992) and data/data compari sons (Beck et al., 1992;Stute et al., 1992;Guilderson et al., 1994) indicate that CLIMAP SSTs, particularly in the low-latitude Pacific, need to be further examined.
Substantial progress has been made by applying the IKM or MAT to estimate Pacific SST using planktonic foraminifers (Luz, 1977;Moore et al., 1980;Thomp son, 1976;Prell, 1985;Le, 1992), but relatively little has been done to evaluate the accuracy of these estimates. In this study, a thorough comparison and evaluation of these models was done with the inten tion of identifying possible sources of estimation bias. This study sheds new light on the interpretation of biotic indices for ocean processes and will help to develop and improve esti mation methods in paleoceanography. The results presented here indicate that the CLIMAP low-latitude Pacific SST pattern may need revision and that a more detailed understanding of the ecology of planktonic foraminifers is needed in order to reduce biases when using paleoestimation techniques.

BACKGROUND AND RESEARCH STRATEGY
The planktonic foraminifers in the Pacific surface sediments are not as well preserved as those in other oceans. Previous studies of the Pacific surface sediment faunas (Parker and Berger, 1971;Coulbourn et al, 1980) revealed the lack of a strong systematic rela tionship between faunal composition and SST. In viewing the complicated distribution pat terns of fauna] assemblages, many investigators have suggested that other environmental fac tors, particul arly the differential carbonate preservation of the assemblages, might be respon sible for the non-systematic relationship between faunal composition and SST. In principle, the local rain rate of organic carbon and the I, C0 2 content of the surrounding bottom waters jointly determine preservation conditions and affect the composition of calcareous faunal as semblages. Many field studies have provided evidence for a systematic relationship between preservation state and faunal composition of planktonic foraminifers (Adelseck and Berger, 1975;Adelseck, 1977;Berger, 1968;Parker and Berger, 1971;Thunell and Honjo, 1981 a;Malmgren, 1983;Cullen and Prell, 1984;Peterson and Prell, 1985), indicating that the state of carbonate preservation is an important environmental control in the distribution of fauna! assemblages.
Planktonic foraminifers grow throughout the photic zone (0 -80 m) in the upper oceans, and on the thermocline-a layer exhibiting a rap id change in temperature. Because the depth of thermocline (DOT) is an important environmental boundary that partitions faunal assem blages into different depth habitats, the changing position of DOT with respect to the photic zone is considered the primary agent by which the relative abundances of shallow-and deep dwelling species in the faunal assemblages are modified. Detailed studies of vertical varia tions in faunal and isotop ic compositions of planktonic foraminifer fluxes Deuser et al., 198 1;Fairbanks et al., 1982;Curry et al., 1983;Thunell et al. , 1983;Thunell and Reynolds, 1984;Be et al., 1985;Deuser and Ross, 1989;Ravelo and Fairbanks, 1992) revealed a vertical stratification signal that should be discernible in sedimentary assemblages. The thermocline-controlled pattern of planktonic fo raminifers on surfac e sediments was revealed by studies in the tropical Pacific (Luz, 1973) and the tropi cal Atlantic (Ravelo et al., 1990), and was considered as one of the most important factors that could introduce a non-systematic relationship between SST and the "gyre-margin assemblages" (Kipp, 1976) or the "low-latitude assemblages" (Molfino et al., 1982) of the faunas.
As discussed above, observational evidence indicates that the distribution of planktonic foraminifers in modern surface sediments, especially in low-latitude oceans, appears to be controlled by several environmental variables. The existence of multiple environmental con trols may introduce biases in SST estimates for past oceans. In this study, the validity of SST shown by small dots (• )) from a global data base (Prell, 1985). estimation for an independent test set of newly compiled coretop data is assessed using a calibration coretop data base previously collected from the Pacific Ocean (Prell, 1985). With the hope of identifying a more reliable approach when using paleoestimation methods to pre dict downcore environmental conditions, the present study proposes to: (1) examine a test set of newly compiled coretop data (Figure la) which contains faunal com positions, carbonate preservation measures (CPI), and seasonal SST and DOT observa tions from the low-latitude Pacific; (2) screen the test data set and a calibration data set from the Pacific (Prell, 1985) (Figure 1 a; 2a) and identify data that are suspected non-Recent or no-analogue samples (Figure 1 for a Pacific data base of 499, and a test set (N = 132) compiled in this study. The 417 calibration coretop set was formed by eliminating 82 coretops from the 499 original data set. (a) A Pacific map presents site locations of a calibration set of coretops (N = 405; shown by "•" indicating well-preserved [RSP% < 30%] and "�" indicating highly dissolved [RSP% > 30%] samples), a subset from a global data base (Prell, 1985). (b) Bathymetric profiles of the RSP% for the 405 coretops in the western, central, and eastern Pacific. The positions of foraminiferal lysoclines (FL) approximately correspond to the changes in slopes of the RSP% profiles. Over the western and cen tral Pacific, samples with RSP% >30 (indicated by arrows on the hori zontal axes) are situated below the FL and are subjected to significant dissolution. Twelve suspected non-Recent coretops that are situated be low the FL, had abnormally low RSP% in the western and central Pa cific.

117
The faunal compositions of these 82 coretop s are dominated by subpolar-polar sp ecies (Globigerina pachyderma (left coiling)) which are rarely found in the low-latitude oceans. The test set coretop data (N = 132) were collected from Thompson (1977 ) as well as from some unpublished counts generated at Brown University (N.G. Kipp, W.L. Prell, M.-T. Chen, and V.S. McKenna) (Chen, 1994a). This test set of coretops was selected from the low latitude Pacific (140°E to 70°W and by 20°N to 30°S), with greater concentrations from the eastern and western equatorial regions (Figure 1 a) . The test set incorp orates planktonic fora minifer faunal assemblages from various ocean environments. This test set also contains counts of foraminifer fragments which were used to quantitatively evaluate the effect of carbonate preservation (Chen, 1994a). The uses of a taxonomic scheme of planktonic foraminifers, indices of carbonate preser vation (CPI= [whole planktonic foraminifers I (whole planktonic foraminifers + planktonic foraminifer fragments)]) and resistant species ratio (RSP%, Cullen and Prell, 1984), and ex traction of environmental data of upper-layer oceans (SST and DOT) from a NOAA compila tion (Levitus, 1982; were described in Chen ( 1994a;.

CORETOP EXAMINATION AND SCREENING
Reconstructions of past SST based on marine microfossils require using a set of coretops that contain the most recently deposited sediments. In paleoestimation, contamination by non-Recent coretops could lead to significant yet undetectable biases. While direct age-con trols for the coretops are lacking in this study, indirect criteria were used to judge whether the coretops are modern sediments. Water depth variations in preservation indices from test set coretops show a distinct change in preservation at the depth of the foraminiferal lysocline (FL), which can be compared to the depth of the previously reported regional lysocline (Figure   1 b ). Pacific glacial sediments that lie above the FL are characterized by good preservation. Thus coretops that are well preserved but are found below the FL are presumed to be non Recent samples. Poorly preserved coretops that are found above the FL could have resulted from an anomalous calcium carbonate saturation state or from increased winnowing. In either case, such samples should be excluded fro m the analyses. Examining 132 test data based on these criteria, five coretops from the test set were identified as having deviating values for whole planktonic foraminifer ratios and were eliminated in fu rther analyses (Figure I b ).
The calibration data set was screened in the same way except only RSP% was used to indicate sample preservation ( Figure 2b). By this method, 12 of the remaining 417 samples were identified as either non-Recent or containing anomalous sediments, and thus were ex cluded from the analyses.

FUNCTIONS
To study how faunal estimates of SST might be biased by specific environmental controls (e.g. upper-layer ocean conditions or preservation), transfer functions were developed by ap plying the standard procedures of the IKM (Imbrie and Kipp, 1971;Kipp, 1976) and the MAT (Prell, 1985). Eight faunal factors (LP1 to LP8) were analyzed from a Q-mode factor analysis (V ARIMAX solution) using 405 calibration coretops (Table 1, Figure 3a-h). The first four of the eight fac tors represent almost 85% of the variance of the original faunal data, and the last four are ecologically meaningful species that are highly relevant in paleoestimation. These eight fac tors explain 97% of the variance of the data and their first-order distribution patterns appear to be associated with Pacific SST or DOT patterns (Chen,l 994a). For example, the abundances of G. ruber (LPl), G. glutinata (LP6), and G. sacculifer (LP7) are at a maximum in both the western Pacific and central gyres, which are regions where the oceanic conditions are characterized by a relatively warm SST and deep DOT with little seasonal variability. The abundances of G. tumida (LP2), N. dutertrei (LP3), £'.. obliquiloculata (LP4), and G. menardii (LP8) , however, dominate the equatorial divergence and coastal upwelling zones, where the SST is relatively cold and the DOT is shallow. Moreover, G. bulloides (LP5) is most abundant in subpolar regions where the SST is cold and shows large seasonal variations.
In computing regression coefficients (Table 2), linear and curvilinear terms for faunal fac tor loadings from a calibration set of 405 coretops were used to generate transfer functions for estimating the SST. The relationships expressed by these transfer functions may be more appropriate for this data set because faunal variations in the Pacific are primarily nonlinear. Observations of SST-cold season were used because they are characterized by a large range of variation and have been conventionally used for downcore estimation in many previous stud ies. The SST transfer function had a correlation coefficient of 0.88 and a standard error of I .92°C (Figure 4; Table 2).
A standard version of the IKM was also used to generate a transfer function for extracting DOT information from planktonic-foraminifer abundance data. It should be noted that this analysis was confined to the annual mean DOT since seasonal DOT changes are not com monly coupled with seasonal surface temperatures. While seasonal variations in DOT are an important component in analyzing upper-layer ocean structure Ravelo et al., 1990), with the variations increasing over high latitudes and in coastal oceans, these variations are not relevant to the present study. The Pacific DOT transfer fu nction had a multiple correlation coefficient of 0.90 with a standard error of about 28.11 m (   Plots comparing estimated and observed SST and DOT (F igure 4) clearly indicate that although the correlation coefficients were approximately the same, the DOT transfer function resu lted in a more systematic relationship when compared to that for SST. The DOT data are more evenly distributed over the range of prediction than the SST data, which display a cluster of points confined within the temperature range of 20° to 30°C. This distribution of SST data can be deceptive, resulting in a positive correlation which appears stronger than the actual Vol. 8, No. 1, March 1997 FACTOR LPl -G. ruber, G. glutinata  , . : � : : r ; � � � � DOT from the 405 coretops reveal them to have linear independent relationships with no major anomalies ( Figure 5). On the other hand, residuals should not be correlated with the dependent variable (observed value), which would imply that the regression model is underspecified and omits relevant predictors. For instance, plots of residual and observed SST and DOT from the 405 coretops revealed linear trends ( Figure 6). In these plots, the transfer functions predicted biased estimates, with tendencies toward higher estimates for low values and lower estimates for high values. This suggests that other environmental variables may play roles in controlling variations in the faunal terms.
The above analyses demonstrate that the correlative relationship of the distribution of planktonic foraminifers with DOT is more significant than that with SST, and indicates that     Table 2 and 3) which relate eight foraminifer faunal factors to the oceanographic observation data from 405 low-latitude Pacific coretop data (Chen, 1994). Fifteen coretops were not used in the analysis because either the DOT could not be defined (SST :::; ; l 8°C) or subsurface temperature data was missing. The dashed lines show the boundaries at the 80% confidence level . the faunal data can be applied to fo ssil records to systematically reconstruct DOT in paleoenvironments. Many previous studies have suggested that changes in faunal abundances in relation to SST are indirect responses to the more direct influence of DOT (Be and Tolderlund, 1971;Williams and Healy-Williams, 1980;Deuser et al., 1981;Fairbanks et al., 1982;Curry et al. , 1983;Thunell and Reynolds, 1984;Deuser, 1987;Deuser and Ross, 1989;Ravelo et al., 1990). When the DOT becomes deep, the relative abundance of the shallow-dwelling species increases as a result of the deep-dwelling species being forced to migrate into aphotic zones where photosynthesis by phytoplanktons is prohibited; and when the DOT becomes shallow, the abundance of faunal species that prefer living in the deeper portion of the upper-layer increases due to higher levels of photosynthesis and nutrient supply in the photic zone. This DOT transfer function could strengthen interpre tations of surface ocean variability based on late Quaternary deep-sea records. If DOT condi-r = 0.44 i:: tions and the correlative relationship between SST and DOT in the low-latitude oceans have changed since the LGM, as has been suggested (Ravelo et al., 1990), then faunal-derived SST estimates may be biased. Such biases need to be assessed and SST values reevaluated, based on the interrelationships that exi st between SST and other environmental factors in the low latitude Pacific.

TEST SET EVALUATION: A SURFACE OCEAN MODE ANALYSIS
To address the issue of whether the effects of other variables can generate biases in SST estimates, it was assumed that these environmental factors , including upper-layer ocean con ditions and carbonate preservation state, are intercorrelated in the low-latitude Pacific. In which case, faunal estimates of SST may be made indirectly through the effects of one or more related environmental controls. The above analyses of surface ocean modes in the modern low-latitude Pacific can be fu rther explained by the fact that DOT and SST are coupled by two climatic processes which are statistically independent. Moreover, since the IKM and MAT are quantitative techniques which predict the average condition based on the entire set of data samples, the collective use of all calibration data from the low-latitude Pacific may also give biased estimates. Conditions in the upper-layer oceans where the 132 test coretops are located were exam ined using a principal component analysis for four variables: SST-cold season, SST-warm season, DOT-cold season, DOT-warm season ( Table 4). The first component represents the surface ocean mode in which the SST and DOT are positively correlated (mode 1), and the second component represents the mode in which the SST and DOT are negatively correlated (mode 2). High scores for the fi rst component (high SST and deep DOT) occur everywhere in the low-latitude Pacific, except in the eastern equatorial region. The scores of the second component (high SST and shallow DOT) are distributed as a dipole along the equatori al zone with the maximum scores at the eastern and western extremes (Figure 7).
These two linearly independent surface ocean modes are probably controlled by an inter action of atmospheric and upper-layer ocean processes. The vertical heat flux through the thermocline is commonly assumed to be important to the heat budget of upper-layer oceans. A shallow thermocline causes cooling in surface oceans which in turn decreases the SST. Thus the predominance of the positively correlated mode 1 implies an obvious connection between SST and DOT through this mechanism, which is probably driven by the convergence and di vergence of the upp er-layer waters in the low-latitude Pacific. The zonal component of the trade winds, termed "Walker Circulation", converges warm surface waters in the western Pa cific, and exposes cold subsurface waters in the eastern Pacific. The negative correlation between SST and DOT in mode 2 implies only a limited influence fr om upper-layer ocean dynamics on SST fluctuations, or a decoupling of SST and DOT. The zonal pattern of this mode suggests a linkage to the trade wind forces of the low-latitude Pacific, which are perhaps dri ven by the seasonal displacement of the Intertropical Convergence Zone (ITCZ) (Philan der, 1990). During the northern hemisphere summer, when the ITCZ is displaced to the north, incoming solar insolati on is minimal south of the ITCZ, which results in a lower SST. Mean while, the southeast trade winds become stronger and increase ai r-sea heat exchange and in turn deepen the DOT in the southern hemisphere. All of these conditions are reversed during the northern hemisphere winter. The southernmost position of the ITCZ accompanied by maximum solar radiation and weaker winds, results in higher SST and shallower DOT in the southern hemisphere. For these reasons, SST and DOT in mode 2 are negatively correlated.
If fauna! abundances of planktonic foraminifers are primarily controlled by DOT changes and fauna! estimates of SST are made indirectly through the effects of DOT, the use of calibra tion data that contain additional correlati ve relationships between SST, DOT, and other vari ables could generate biases. To test this possibility, the residuals of the SST estimates (� SST) of the 132 test coretops were analyzed in scatter plots in which the �SST was plotted against several factors, including the principle components of sea-surface conditions that represent the modes of upper-layer ocean conditions as well as CPI (Figure 8). Because non-Recent and no-analogue coretops would be estimated with large errors, 17 test coretops which were previ ously identified when testing the significance of the relationships between SST and these other  Table 5) is used to analyze the relationship between SST and DOT from 132 low-latitude Pacific coretop data. Mode 1, in which high SST is correlated with deep DOT, reaches a maximum near the equator, with positive scores in the western and negative scores in the eastern Pacific. Mode 2, in which high SST is correlated with shallow DOT, is a dipole with positive scores over the equator and negative scores away from the equator and nodes near the northernmost and southern most extremes of the lntertropical Convergence Zone (ITCZ). The pat terns of the two surface ocean modes are driven by diffe rent climatic mechanisms. The predominance of mode 1 suggests that thermocline dynamics play an important role in SST variations in the low-latitude Pacific .
variables (Chen, 1994a), were eliminated. Simple regression analyses were performed using the remaining 115 test coretops ( 115 = 132 -17) and clearly demonstrated that both the IKM and MAT predicted SST with biases toward lower estimates for coretops with positive scores of the surface ocean modes 1 and 2 and toward higher estimates for coretops with negative scores of the surface ocean modes 1 and 2 ( Figure 8). Statistical tests indicated that these correlations were significant at the 99% level (Table 5) with estimated error of about 3° to 4°C. A similar regression analysis compar ing SST with CPI revealed a significant though weaker relationship with biases toward lower estimates for highly dissolved coretops and higher estimates for well-preserved coretops. Because the surface ocean modes and CPI are correlated (mode 1 and CPI: +0.24; mode 2 and CPI: -0.71) in the low-latitude Pacific, their common effects on �SST have to be taken into account. By calculating partial correlations to remove these common influences in order to obtain adjusted estimates for the real relationships (Table 5), correlations between the �SST and surface ocean modes 1 and 2 were still significant and correlations between the �SST and CPI were no longer significant. These results suggest that (1) faunal estimates of SST are influenced by upper-layer ocean conditions which give systematic biases in the predicted SST values with errors of about 3° to 4°C; and (2) the effect of carbonate preservation on SST estimates seems to be insignificant.
This study suggests that DOT is the most important environmental variable that biases traditional IKM or MAT SST estimates in the low-latitude Pacific. To understand why esti mate biases exist, we first need to consider that the DOT and SST in the low-latitude Pacific are linked through the governance of two independent modes of ocean dynamics. These two surface ocean modes, one representing a zonal and the other a meridional circulation pattern in the low-latitude Pacific, indicate that there may be a causal link between DOT and SST. Esti mates of SST derived fro m faunal data were probably made indirectly through the more im portant f a ctor DOT, and the accuracy of these estimates, to some extent, relies on the degree of correlation between DOT and SST which may vary between the two ocean modes. Therefore, different estimates of SST will be derived if the correlation between DOT and SST changes. For example, if DOT becomes shallow, low SST estimates will be given for samples from the western Pacific (mode 1 dominance area) and high SST estimates will be given for samples from the equatorial Pacific (mode 2 dominance area).
The bias patterns associated with surface ocean modes shown in Figure 8 are apparently a result of "mean condition prediction", an effect inherent in the IKM and MAT paleoestimation methods which predict the statistical average of modem surface ocean conditions based on a chosen set of coretops. In the IKM, the mean conditions of SST are determined by using regression analyses based on an ocean-wide set of calibration coretop data. In the case where faunal variations are confined within a limited SST range, scatter plot data lying above the regression line would be underestimated (negative i'.lSST) and data lying below the regression line would be overestimated (positive i'.lSST). This problem also exists in MAT estimates in which the mean condition of SST is obtained by taking the average of 10 or less of the most similar analogue coretop data. Biases would occur in this analysis if coretops with similar faunal compositions but different SSTs were selected. By averaging the coretop data, the IKM and MAT bias estimates toward the middle range of values, resulting in less variability in estimation values, yet losing accuracy at the high and low ends of the scale. Thus values are underestimated at the higher and overestimated at the lower extremes. This tendency was also revealed in the results from the analyses of the calibration coretop data set ( Figure 6).
This examination poses questions concerning the limitations of faunal estimates of SST in the low-latitude oceans. In contrast to previous paleoceanographic studies in which the faunal composition of planktonic foraminifers is used as a proxy indicator of SST, this study indi cates that in the low-latitude Pacific, DOT may be a more important indicator which can be used to interpret the ecological significance of the faunas. These results suggest that the DOT effect should be considered when estimating SST, and the CLIMAP low-latitude Pacific SST patterns during the LGM (CLIMAP, 1981) must be reevaluated. The CLIMAP estimates using planktonic foraminifers from the western tropical Pacific suggest little change (only I 0 to 2°C) in SST between the LGM and the present. In contrast, continental temperature proxy records suggest that the tropics were 4° to 6°C colder then than today (Webster and Streten, 1978;Rind and Peteet, 1985). This apparent cooling of the LGM tropics has been further supported in studies using geochemical proxy indices such as Sr/Ca ratio (Beck et al., 1992), noble gas thermometer (Stute et al., 1992), and both oxygen isotope and Sr/Ca evidence (Guilderson et al., 1994 ). These discrepancies suggest that the estimated CLIMAP LGM SST value for the tropics was at the low end of the scale and thus was estimated to be warmer than it should have been, by about 3° to 4°C.
These analyses have so f a r been limited to the environmental controls that are important to the composition of planktonic foraminifers. While it appears that the effects of upper-layer ocean conditions may account for the biases revealed in fau nal estimates of SST in the low latitude Pacific, several fac tors could additionally affect SST estimates. Comparisons of Pa-cific SST transfer functions based on different subsets of coretop data revealed discrepancies in downcore estimates (Le, 1992). Testing the accuracy of CLIMAP SST transfer functions against subsets of coretops from the other oceans also shows significant differences (Prell, 1985). These inconsistencies indicate that the accuracy of SST estimates is sensitive to the spatial distribution and range of the coretop data that are chosen for calibration. The IKM equations are derived with different sets of regression coefficients and give different estimates for identical downcore samples because the correlation between each faunal term and SST changes depending upon the subset of coretops used for calibration. The accuracy of MAT SST estimates also relies on a set of coretop data which are evenly distributed over a wide temperature range. When a more strict cutoff value for dissimilarity coefficients was used, 11 test coretops exceeded the cutoff and were not included in the analysis. The failure of the MAT in this case suggests that the coretop data currently available for the Pacific are not fully representative of all sea-surface conditions. Efforts to collect more coretop data in the Pacific will be needed in future studies.

CONCLUSIONS
Evaluation of faunal SST estimates using calibration and test set coretops from the low latitude Pacific revealed biases in the statistically based estimates which are at least in part a result of changes in DOT, an environmental variable that is interrelated with both SST and faunal distribution. Coretop data sets of planktonic foraminifer faunas from modern surface sediments, seasonal observations of SST and DOT, as well as indices of carbonate preserva tion were compiled and analyzed in this study . A standard CLIMAP-type transfer function for estimating SST was developed and was compared to a transfer function for estimating DOT based on the same calibration coretop data set. Comparisons between these two fac tors dem onstrated that the correlative relationship between the abundance of planktonic foraminifers and DOT was more significant, and could be applied to fossil records to accurately reconstruct DOT in paleoenvironments. In future studies, the DOT should be considered as an important indicator in interpreting the ecological significance of planktonic foraminifers in the low latitude Pacific.
An evaluation of test set data analyzed the relationships between the residuals of SST and DOT estimates and two statistically independent low-latitude Pacific surface ocean modes as well as CPI. These residual analyses revealed that the patterns of SST estimation biases were significantly correlated with the two ocean modes through which the DOT and SST are differ entially coupled. There was a bias toward predicting colder estimates for high SST values and warmer estimates for low SST values. The maximum uncertainty in these biased estimates was about 3° to 4°C. Although previous studies suggested that the change in carbonate pres ervation may cause biases in SST estimation, the present analysis indicates that the relation ship between CPI and �SST was not significant and did not systematically cause estimation biases.
Paleoceanographic observations in the Pacific Ocean are complex and difficult to analyze due to limited control data, relatively low preservation of surface sediments, and complicated surface ocean dynamics. The present results indicate that with respect to the distribution and abundance of planktonic foraminifers in the low-latitude Pacific, the DOT effect may be a more important environmental control than SST. The LGM SST patterns that have been pre sented by the CLIMAP ( 1981) need to be reevaluated, taking the DOT effects into account. Furthermore, it is clear that analysis of ocean modes may be critical in future paleoceanographic applications.