Here we analyzed published alkenone SST datasets from ODP Site 882 and ODP Site 883, located 49 nautical miles to the north of Site 882 (Rea et al., 1993a, b; Fig. 1). Modern sea surface temperatures at the two sites are essentially identical (5.82 °C at ODP Site 882, 5.75 °C at ODP Site 883, Fig. 1).The original ODP Site 882 alkenone SST record was determined by GC-CI-MS (Haug, 1995; Rosell-Mele et al., 1995), and was subsequently published by Haug et al. (2005) and Martínez-Garcia et al. (2010). The potential concentration-dependent nonlinear differences between GC-FID and GC-CI-MS U37K′ determinations identified by Chaler et al. (2000, 2003) and their impacts on absolute SST values were not discussed in these publications, which focused mainly on the analysis of reconstructed SST trends. More recently, alkenone records from Site 882 itself and the nearby Site 883 were determined by GC-FID (Herbert et al., 2016; Novak et al., 2024; Studer et al., 2012; Yamamoto and Kobayashi, 2016), allowing the assessment of potential biases in the original dataset from ODP Site 882 (Haug, 1995). From here on, we base all of our calculations on U37K and U37K′ values calculated from the original dataset measured by Haug (1995).

Chaler et al. (2003) derived a mathematical expression of U37K′ as measured on GC-FID as a function of GC-CI-MS measurements of alkenone concentration: 3U37K′=(c2MC37:22+b2MC37:2+a2M)(c2MC37:22+b2MC37:2+a2M+c3MC37:32+b3MC37:3+a3M) here, a2M, a3M, b2M, b3M, c2M, and c3M are curve fitted constants reported by Chaler et al. (2003). These constants are instrument-specific (Chaler et al., 2003). We utilized the constanst from GC-CI-MS “Instrument A” of Chaler et al. (2003) to calculate theoretical GC-FID-equivalent U37K′ values to facilitate comparison between the GC-FID and GC-CI-MS alkenone SST datasets from ODP Sites 882 and 883 (Table 1). These theoretical GC-FID-equivalent U37K′ values were calculated over the range of U37K′ values in the alkenone SST records from ODP Sites 882 and 883 to understand the expected sign and magnitude of any potential analytical bias in the GC-CI-MS method (Novak et al., 2024; Studer et al., 2012; Yamamoto and Kobayashi, 2016). We varied the mass of each theoretical alkenone analyte between 5 and 30 ng on column to mimic the “trace” quantities of alkenones reported by Haug et al. (2005) because no specific alkenone concentrations were reported alongside their SST determinations; the raw data files generously shared with us from their work report molecule abundances as peak areas rather than concentration. For example, a low U37K′ value was represented as 5 ng of the C37:2Me alkenone and 30 ng of the C37:3Me alkenone. These analyte masses are slightly below to slightly above the lower bound of the range of acceptable analyte mass for U37K′ determinations by GC-FID as documented by Villanueva and Grimalt (1996).

Because we lack instrument-specific information about the mass-dependent ionization efficiency of the C37:3Me and C37:2Me alkenones on the GC-CI-MS used by Haug (1995) to generate their U37K′ data from ODP Site 882, we cannot use the equations of Chaler et al. (2003) or Hefter (2008) to correct their record. Instead, we propose to quantify the uncertainties in the Site 882 alkenone SST record introduced by the GC-CI-MS method based on the correlation between U37K′ values of adjacent samples at ODP Site 882 generated by GC-CI-MS and GC-FID (Haug, 1995; Studer et al., 2012; Yamamoto and Kobayashi, 2016). GC-FID and inferred GC-CI-MS U37K′ values were paired by interpolating the more densely sampled GC-CI-MS record onto the more sparesly sampled GC-FID records.

We explored both linear and quadratic fits of the ODP 882 GC-CI-MS and GC-FID values in addition to multiple linear regression models that incorporated the absolute peak areas measured by Haug (1995) with GC-CI-MS (Figs. S1 and S2 in the Supplement). Ultimately, a linear model was selected to transform the inferred GC-CI-MS U37K′ values to GC-FID equivalents as it provided the optimal fit while ensuring the model terms remained statistically significant (p<0.001, adj. r2=0.56, df=68, Eq. 4). This equation was derived after screening samples based on the absolute peak areas measured by GC-CI-MS (see Sect. 3.2; note that we lack the necessary information about the laboratory standards used by Haug (1995) to calculate concentrations from peak area). 4GC-FIDU37K′=GC-CI-MSU37K′⋅061(±0.07)+0.07(±0.03) The substantial standard error of this relationship, ±0.07 U37K′ units, is equivalent to an analytical uncertainty of 2.05 °C SST in addition to the calibration uncertainties inherent to estimating SSTs from U37K′ values. We note that this estimate of the analytical uncertainty of the ODP Site 882 GC-CI-MS dataset is inherently imperfect because we lack the instrument-and-analysis-specific information required to more precisely quantify drift in the concentration-dependent ionization efficiencies of the C37:2Me and C37:3Me alkenones. Therefore, our quantification of the analytical uncertainty serves as an estimate of the mean analytical uncertainty for the portion of the ODP Site 882 GC-CI-MS alkenone dataset with measured peak areas <2×107 and may underestimate or overestimate the uncertainties associated with specific samples (see Sect. 3.1).

The SST equivalent of the analytical uncertainty of the Site 882 inferred GC-CI-MS U37K′ values was calculated from the slope of the widely-applied BAYSPLINE U37K′ SST core top calibration (Tierney and Tingley, 2018). 5Analytical uncertainty SST equivalent=±0.070.034=+2.05°C This additional analytical uncertainty is greater than the prediction uncertainty of the U37K′ proxy when estimating SST (Conte et al., 2006; Müller et al., 1998; Prahl and Wakeham, 1987; Tierney and Tingley, 2018). We applied Eq. (7) to generate GC-FID equivalent U37K′ values from the U37K′ values measured by GC-CI-MS (Haug, 1995). We then propagated the additional uncertainty to calculate a 95 % confidence interval for the resulting median SST estimates by adding in quadrature the analytical uncertainty sea surface temperature equivalent (2.05 °C) multiplied by 1.96 to the upper and lower range of the 95 % SST confidence interval returned by BAYSPLINE. 695%CI lower bound=BAYSPLINE median SST-(BAYSPLINE median SST-BAYSPLINE SST95%CI lower bound)2+(1.96⋅2.05)2795%CI upper bound=BAYSPLINE median SST+(BAYSPLINE SST95%CI upper bound-BAYSPLINE median SST)2+(1.96⋅2.05)2 Because BAYSPLINE generates a unique 95 % confidence interval for each sample, this calculation was applied to the BAYSPLINE output for each sample individually.

We compared the various alkenone datasets from ODP Sites 882 and 883 with two-sided t-tests to assess the differences in mean values between the records. Because there are substantial differences in the timescale and sampling resolution, the datasets were first resampled to avoid the consideration of data from intervals where one record may have data while another does not. In all cases, the more highly resolved record was resampled onto the less resolved record.

We examined the implications of the corrected ODP Site 882 temperature in part by calculating the implied changes in the density of surface water from the thermodynamical equation of seawater (TEOS-10) as implemented in the Givvs SeaWater Oceanographic Toolbox (Feistel, 2018; McDougall and Barker, 2011).

The differences between U37K′ values measured by GC-CI-MS and GC-FID are both nonlinear and concentration dependent (Chaler et al., 2000, 2003; Hefter, 2008). While we do not have the necessary information to calculate alkenone concentrations from the Haug (1995) alkenone measurements, the relative intensity of the instrument response (i.e., the peak areas of the molecular ion ammonium adducts) provides qualitative information about the amount of analyte introduced into the detector (Fig.  2c). Peak areas of the alkenones quantifed in ODP Site 882 sediments by Haug (1995) varied by several orders of magnitude (105-108; Fig. 2c), implying substantial sample-to-sample differences in the concentration of analyte measured, which is consistent with absolute alkenone concentrations reported by others (Novak et al., 2024; Studer et al., 2012; Yamamoto and Kobayashi, 2016) and the differences we observe between the GC-CI-MS and GC-FID U37K′ values at ODP Site 882 (Fig. 3a).

The GC-CI-MS U37K′ values from ODP Site 882 are systematically larger (warmer) than the U37K′ values determined by GC-FID (Fig. 3a). The mean difference between the inferred GC-CI-MS and GC-FID U37K′ values is 0.113 U37K′ units (p<0.001, two-sided t-test, GC-CI-MS values are larger), which is equivalent to 3.32 °C SST (Eq. 5). This finding is consistent when comparing the inferred GC-CI-MS U37K′ values with U37K′ measurements from the nearby ODP Site 883 (Fig. 3b), which have a mean difference of 0.09 U37K′ units equivalent to 2.65 °C from the ODP Site 882 inferred GC-CI-MS U37K′ values (p<0.001, two-sided t-test, GC-CI-MS values are larger). This is greater than the typical interlaboratory differences in alkenone measurements (±1.3–2.6 °C; Rosell-Melé et al., 2001). By comparison, the U37K′ values determined by GC-FID at ODP Sites 882 and 883 are not systematically offset as a function of increasing U37K′ (Fig. 3c); the mean difference in ODP Site 882 and ODP Site 883 U37K′ is 0.023 U37K′ units or 0.68 °C SST (p=0.06, two-sided t-test, ODP 882 values are larger).

The systematically larger (warmer) inferred GC-CI-MS U37K′ values compared to those determined by GC-FID at ODP Site 882 are consistent with the known concentration-dependent nonlinear differences in the chemical ionization of C37:2Me and C37:3Me alkenones (Chaler et al., 2000, 2003). We explored whether the GC-CI-MS peak area data might improve predictions of GC-FID equivalent U37K′ values from the GC-CI-MS U37K′ values (Fig. S2). We found that 80.6 % of the variance in the ODP Site 882 GC-CI-MS alkenone dataset can be described by two principal components (Fig. S2a). The GC-FID U37K′ values are more strongly correlated with PC1 (Fig. S2b), while the GC-CI-MS U37K′ values are more strongly correlated with PC2 (Fig. S2b). The GC-CI-MS peak areas, whether the sum of peak areas for all molecules or the peak areas of individual molecules, are negatively correlated with PC1 and the GC-FID U37K′ values but positively correlated with PC2 and the GC-CI-MS U37K′ values (Fig. S2b). These statistical relationships lead us to experiment with multiple linear regression models, one of which moderately improved the accuracy of GC-FID equivalent U37K′ values predicted from the GC-CI-MS data (Fig. S2c vs. d). However, application of this model to the GC-CI-MS U37K′ dataset in an effort to produce GC-FID equivalent U37K′ values resulted in nonsensical (negative) estimates. Therefore, we do not think it is possible to use a single model to correct for the complex offsets between the GC-CI-MS and GC-FID alkenone datasets at ODP Site 882, at least with the information available to us.

We then experimented with cutoffs based on the GC-CI-MS instrument response (peak area) in an effort to distinguish between different populations of samples that might have different relationships between GC-CI-MS and GC-FID U37K′ values (Fig. 4). We found that the strength of the regression between GC-CI-MS and GC-FID U37K′ values is related to the magnitude of the GC-CI-MS instrument response (peak area) for a given sample (Fig. 4b). The differences between the ODP Site 882 GC-CI-MS and GC-FID U37K′ values become larger in samples with larger GC-CI-MS peak areas versus those with smaller peak areas (Fig. 4a). While these differences should be systematic (Chaler et al., 2000, 2003), the lack of GC-FID data from intervals with higher alkeneone abundances does not permit us to robustly characterize the correlation between GC-CI-MS and GC-FID U37K′ values in samples with GC-CI-MS-measured peak areas >2×107 (Fig. 4a). These observations lead us to propose a correction to the ODP Site 882 alkenone dataset based upon a linear relationship between GC-CI-MS and GC-FID U37K′ values in samples with measured peak aras <2×107 (Fig. 4a blue line, Eq. (7), p<0.001, adj. r2=0.56, df=68). All samples with measured peak area >2×107 were excluded from further consideration since we lack the necessary information to propose a correction to those data.

The proposed corrections to the ODP Site 882 GC-CI-MS alkenone record result in sea surface temperature estimates that are consistent with the nearby alkenone SST record from ODP Site 883 (Fig. 5). The mean of the SST calculated from the corrected GC-CI-MS dataset is significantly smaller than the original GC-CI-MS ODP Site 882 record (p<0.001, df=806.5, two-sided t-test). The magnitude of the correction is larger at higher U37K′ values (i.e., warmer SSTs; Fig. 5c). The mean difference between the corrected ODP Site 882 GC-CI-MS dataset and the GC-FID record from ODP Site 883 is 0.59 °C (p<0.001, df=903.5, two-sided t-test), which is consistent with the difference between the GC-FID records from Sites 882 and 883 (0.68 °C, Fig. 5d, see also Sect. 3.1). These observations suggest that our proposed correction successfully removes the systematic “warm” analytical bias in the original ODP 882 GC-CI-MS dataset while maintaining the potentially real differences in SST indicated by the GC-FID data (i.e., the slightly more northward Site 883 is somewhat cooler; Fig. 1).

The same methodological differences (GC-FID vs. GC-CI-MS) exist between the late Neogene and Quaternary alkenone SST datasets from ODP Site 887 and IODP Site U1417 in the subpolar northeast Pacific, but we are unable to determine whether these methodological differences lead to the same systematic biases between the datasets because there are very few proximal samples measured by both methods. The alkenone record from IODP Site U1417 was analyzed by GC-CI-MS (Sánchez-Montes et al., 2020), while the record from the nearby ODP Site 887 was analyzed by GC-FID (Dowsett et al., 2017; Herbert et al., 2016). The selectivity of the GC-CI-MS method permits quantification of alkenones in complex mixtures that cannot be resolved by sample preparation (Rosell-Mele et al., 1995). At IODP Site U1417, the GC-CI-MS method enabled the isolation of alkenones from a complex mixture that resulted from the introduction of petrogenic carbon from the nearby St. Elias mountains (Sánchez-Montes et al., 2020), which was not an issue at the more pelagic ODP Site 887 (Dowsett et al., 2017; Herbert et al., 2016). There are only 11 samples from these two records that overlap in time (Fig. 6a). Although the mean U37K′ values from mid-Piacenzian subsets of these data are statistically indistinguishable (Fig. 6b, 0.008 U37K′ or 0.24 °C, p=0.66), the systematic differences between the GC-CI-MS and GC-FID U37K′ values from ODP Site 882 and the well-documented systematic offset between the GC-CI-MS and GC-FID methods give us reason to suspect that the IODP Site U1417 and ODP Site 887 alkenone datasets may not be strictly intercomparable. For this reason, we exclude the IODP Site U1417 alkenone dataset from our discussion of the implications of the proposed correction to the ODP Site 882 GC-CI-MS alkenone record for the paleoceanographic history and paleoclimate data-model evaluation in the subarctic North Pacific.

Correction of the ODP Site 882 alkenone SST record has notable implications for our understanding of long-term patterns of climate change in the subarctic North Pacific and for inferences of regional patterns of warming in the late Pliocene. To date, the GC-CI-MS alkenone dataset from ODP Site 882 is the only continuous subarctic North Pacific SST record that spans the entirety of the Quaternary. Consequently, this dataset is crucial for calculating SST anomalies – that is, the difference between ancient SST estimates and the SST estimate of the core top sample, which is assumed to represent the preindustrial SST – that are often used for testing the ability of Earth System Models to simulate warmth in this oceanographic region (see Sect. 3.5).

The reporting error in the Pangaea data table and the analytical artifacts we address with our proposed correction are of the opposite sign: (1) U37K values are systematically lower than U37K′ and (2) GC-CI-MS U37K′ values are larger than GC-FID U37K′ values (Figs. 2a and 3). Here, we discuss the implications of these corrections for reconstructing regional SST patterns in the subarctic North Pacific during the mid-Piacenzian Warm Period since several studies have unknowingly used mislabelled U37K values to characterise regional SSTs with a U37K′ SST calibration (e.g., Brennan et al., 2022; Tierney et al., 2019, 2025b).

The proposed correction results in a median alkenone core top SST value of 6.4±2.5 °C (±1σ) from ODP Site 882. This is cooler than both the core top SST estimate when BAYSPLINE is applied to the mislabelled U37K values (7.2±1.4 °C) or to the original U37K′ dataset (8.1±1.4 °C). The mid-Piacenzian Warm Period SST anomaly of the corrected dataset is 2.5±3.5 °C compared to 1.8±2.0 °C when BAYSPLINE is applied to the mislabelled U37K values or 2.8±2.0 °C in the case of the original U37K′ dataset. The differences in median SST values are small, but the uncertainties associated with the corrected dataset are substantially larger than previously recognized and are important for the relative weighting of the ODP Site 882 GC-CI-MS dataset in paleoclimate data-model comparisons and data assimilation exercises (Haywood et al., 2016; Tierney et al., 2025a). It should be noted that the mid-Piacenzian Warm Period SST estimates in the corrected dataset are sparse (n=2), so these mean temperature values will likely change as new data are generated.

The proposed correction also impacts calculated SST anomalies from the nearby ODP Site 883 since the core top value from ODP Site 882 can reasonably be used for this calculation. The ODP Site 883 mid-Piacenzian Warm Period SST anomaly shifts to 1.4±2.7 °C from 0.6±2.0 °C (BAYSPLINE to ODP 882 U37K core top value) or -0.3±2.0 °C (BAYSPLINE to original ODP 882 U37K′ core top value). The proposed correction to the ODP Site 882 dataset therefore contributes towards resolving uncertainties regarding the zonal pattern of SST change in the subarctic North Pacific during the mid-Piacenzian Warm Period (Fig. 7a). The mean SST anomalies at ODP Sites 883 and 887 were not significantly different prior to the correction (0.6 or -0.3 vs. 0.05 °C, p=0.08 or p=0.3, two-sided t-test), but are significantly different after the correction (1.4 vs. 0.05 °C, p=0.0014, Fig. 7a). The temperature difference between the east and west subarctic North Pacific is tightly linked to the strength and position of the Aleutian Low and its interaction with regional ocean currents (Zhou, 2019), a pattern which is reflected by the distribution of core top U37K′ values in the subarctic North Pacific (Max et al., 2020). The reduced mid-Piacenzian zonal subarctic North Pacific SST gradient suggests a northwards shift of the Aleutian Low and a concomitant northwards shift of the Kuroshio Extension (Sugimoto and Hanawa, 2010; Wang and Wu, 2019). Consequently, the subpolar gyre was possibly contracted during the mid-Piacenzian relative to the preindustrial period (Park et al., 2012). These inferences are consistent with a growing body of marine and terrestrial paleoclimate data suggestive of a northwards-shifted midlatitude westerly jet during the mid-Piacenzian warm period relative to the preindustrial (Abell et al., 2021; Li et al., 2015; Novak et al., 2026).

The pattern of deep ocean circulation in the mid-Piacenzian warm period is disputed, with controversy surrounding the existence, or lack thereof, of deep water formation in the subarctic northwest Pacific (Burls et al., 2017; Ford et al., 2022; Novak et al., 2024; Tierney et al., 2025b). The corrected ODP Site 882 alkenone SST record clarifies that the source waters of North Pacific Intermediate Water and any potential North Pacific Deep Water were likely warmer in the mid-Piacenzian than during the preindustrial era (cf., Talley, 1993), which informs the sea surface salinity dynamics required for any intermediate, or potentially deep, waters to form in this region (Burls et al., 2017). Warmer SSTs on the order of ∼1.4 °C correspond to a decerase in the density of surface waters by 0.31 kgm-3 (see Sect. 2.5). An increase in sea surface salinity of 0.24 gkg-1 would be required to compensate for the temperature-driven decrease in surface water density (Fig. 7b), which exceeds the current seasonal variability of sea surface salinity in the Kuroshio Extension region (Wakita et al., 2010). The warmth reconstructed in the subarctic northwest Pacific by the corrected ODP Site 882 alkenone record therefore indicates that any density-driven overturning in this region, either intermediate or deep, would have required even greater changes to regional patterns of sea surface salinity to counterbalance weakened temperature-driven buoyancy forcing than previously recognized (cf., Burls et al., 2017).

A major feature of the ODP Site 882 GC-CI-MS alkenone SST record is the warming signal across the 2.7 Ma interval (Fig. 8), which has been interpreted as a signal of the onset of regional surface ocean stratification and enhanced seasonal SST contrast (Haug et al., 2005). This signal persists in the corrected ODP Site 882 GC-CI-MS dataset and is further corroborated by the GC-FID datasets (Fig. 8). While the different datasets disagree on the amplitude of SST change across this interval (Fig. 8), all three alkenone records indicate statistically significant warming of SSTs across the 2.7 Ma interval (ODP Site 883 GC-FID: 1.9±1.4 °C, p=0.01; ODP Site 882 GC-FID: 3.0±1.4 °C, p<0.001; corrected ODP Site 882 GC-CI-MS: 2.7±2.3 °C, p<0.001). This unusual alkenone-based SST observation has also been corroborated by TEX₈₆ measurements (Studer et al., 2012). The agreement on the sign of SST change and the overlapping 95 % confidence intervals of these four independent estimates confirms that the SSTs recorded by alkenones warmed across the 2.7 Ma interval.

The zonally asymmetric warming that we observe in subarctic North Pacific alkenone SST records is reproduced by the PlioMIP2 mutli-model mean (Haywood et al., 2020; Fig. 9). Notably, the warming pattern observed in the PlioMIP2 multi-model mean is consistent with a northwards expansion of the Kurishiro Extension (Sugimoto and Hanawa, 2010; Wang and Wu, 2019). This contrasts with the removal of this zonally asymmetric warming feature from some, but not all, data assimilation and reduced space reconstructions of late Pliocene SST patterns (cf., Annan et al., 2024; Tierney et al., 2019, 2025b). We suspect that this difference is due to the previously unrecognized reporting errors and uncertainties in the ODP Site 882 GC-CI-MS alkenone dataset and hypothesize that zonally asymmetric warming in the subarctic North Pacific will be a feature of future mid-Piacenzian data assimilation products that incorporate the corrected dataset proposed here (uncertainties shown in Fig. 9c vs. d).