Background & Summary

The stable isotopic ratio of carbon and oxygen of benthic foraminifera, commonly expressed in delta notations (δ13C and δ18O) when compared with the ratio of established standards, are often used as tracers of ocean circulation, climate and carbon cycle processes. δ18O values from CaCO3 tests of epibenthic to shallow infaunal foramifera have been linked to bottom water temperatures and sea level1,2, sea water densities3, transport rates4,5,6 as well as the transport in the deep ocean7. The δ13C values from CaCO3 tests traces the δ13C values of bottom water dissolved inorganic carbon (DIC) and is used to infer carbon cycling and the distribution of deep ocean water masses8,9,10,11.

Despite the relatively large amounts of existing data, the use of stable isotope compilations in paleoclimate research is hindered by the following issues:

  • Heterogeneous, dispersed data: Data from sediment cores are typically processed, analyzed, and archived separately in data repositories or personal computers. The format and content of the data files varies across cores and operators, and often different data files for a single core exist. Thus, paleoceanographic data in existing repositories are highly heterogeneous. This makes compiling data difficult and time-consuming, complicating their reusability.

  • Age models: Interpretations of paleoceanographic data require age-depth models to associate the depths in core with calendar ages. Different types of age constraints exist, for instance 14C dates12, ash layers13, alignment to benthic or planktonic foraminiferal δ18O variations14, surface temperatures, magnetic properties15 or 14C features16. Additionally, multiple age models can be produced from the same underlying age data depending on the software package used, adjustable parameters within the software package, the atmospheric radiocarbon calibration curve used, and the radiocarbon reservoir ages assumed for the core site. The diversity of methodologies makes it difficult to compare stable isotope time series from cores provided by different sources, especially for climate change events such as during the last deglaciation (~20-10 thousand years before present (ky BP)).

  • Species offsets: Because of its epifaunal (i.e., on and slightly above the sea floor) habitat, δ13C determined from tests of the genus Cibicidoides, in particular Cibicidoides wuellerstorfi, has the lowest offsets with respect to δ13C of DIC11, making it the preferred analyzed species for δ13C values of seawater DIC reconstructions. However, numerous sites include δ13C values determined from other species or even genera, including infaunal Uvigerina, which yield higher offsets. Benthic foraminiferal δ18O values are also affected by species offsets17, and some publications include species-specific corrections to obtain equilibrium or seawater δ18O18.

The Ocean Circulation and Carbon Cycling (OC3) working group of the Past Global Changes (PAGES) program seeks to understand global ocean carbon cycling, ocean circulation and climate during the last deglaciation. One major goal is to create a global database of δ13C and δ18O data from benthic foraminifera that would overcome the shortcomings outlined above. OC3 members have developed specific targets, criteria for inclusion of data, a quality control procedure, and a database structure. One of the specific goals is that the new database should be easy to update in the future and extendable to other variables. Specifically, the OC3 database is an ever-evolving database that can be used for many different purposes beyond the specific scientific goals of OC3. Its first version, which is presented here, consists of a compilation of high-resolution benthic foraminiferal δ13C and δ18O time series from the global ocean. Stable isotopes of oxygen and carbon of benthic foraminifera as well as data used for the calculation of age models are compiled, including different age models for each site, when available. All components undergo a quality control to standardize the database, and we only include sites that can resolve millennial-scale changes associated with the last deglaciation.

One important goal of OC3 is to quantify uncertainty. This includes chronostratigraphic uncertainties. For this purpose, we included different age models for sediment cores, if multiple age model approaches are available. The OC3 database archives both stable isotope data and age model information, yet separately. In other words, isotope data are kept separate from age model information, but a connection of both is provided by the OC3 database. This facilitates future updates of age models without information loss. When available, the database includes all relevant data necessary to construct the age model, such as radiocarbon dates, reservoir age corrections, and tie points to reference records.

The purpose of this paper is to describe the first version of the OC3 database. We describe its structure and list the sites and age models included. We then describe several programming tools used to facilitate analysis of the database. Finally, we illustrate the utility of the database by comparing different age models across the last deglaciation.


Data acquisition

Benthic foraminiferal δ13C and δ18O data from global marine sediment core sites were collected from on-line repositories, original publications, personal communications, and recent data compilations (Tables 16). Species included in the database are displayed in Table 7. We include benthic foraminiferal species from the genus Cibicidoides, especially Cibicidoides wuellerstorfi. Some Uvigerina stable isotope data are also included, in particular for the sake of documentation of previously-unpublished sites. We define a data quality control protocol to identify “good data”, of sufficient quality and resolution according to the following criteria:

  • The temporal resolution of the benthic foraminiferal δ13C and/or δ18O data is 1 ky or better for the Last Glacial Maximum (LGM, 23-19 ky BP) and/or early deglaciation (ED, 19-15 ky BP).

  • The original publication, as well as the source of the isotope data and age models, were checked for differences with the values presented in the database. When possible, a quality control was performed by the original author or compiler of the data. Data sources labeled as personal communications were provided directly from the original owner of the data to the authors of this work.

  • We identified whether species-specific corrections were applied to the raw stable isotope data. Both uncorrected and corrected data are reported in the database.

  • Outliers and hiatuses, when reported in the original publications, were checked for and marked.

  • Species names were checked and standardized within the database.

Table 1 Sites from the OC3 deglacial compilation.
Table 2 Continuation of Table 1.
Table 3 Continuation of Table 2.
Table 4 Continuation of Table 3.
Table 5 Continuation of Table 4.
Table 6 Continuation of Table 5.
Table 7 Taxon flags associated with the different benthic foraminifera species included in our data base.

For most sites, the depth-in-core scale is a quantity directly measured in the core. However, some records are based on spliced sections (mainly Ocean Drilling Program (ODP) and Integrated Ocean Drilling Program/International Ocean Discovery Program (IODP) sites) of several nearby cores to generate a composite with a corresponding composite depth to define the seafloor referenced depth scale for the site. When available, these depth models are documented in the database, accompanied by archival depths that correspond to the original depth within each cored interval.

To have a measure of the uncertainty in the timing of deglacial shifts in isotope time series, we include as many published age models associated with the data series as attainable. Only those age models that include information about how they were calculated are included. Age models were either obtained from original publications and recent syntheses, or generated for this work. We include age models from three published compilations, which focus mostly on Atlantic sites:

  • From Peterson et al.19 we include age models for 48 sites, calculated using benthic foraminiferal δ18O values combined with radiocarbon-based age models14. These age models are referred to as P hereafter.

  • From Waelbroeck et al.12 we include Undatable software age models20. They were calculated from planktic foraminiferal calibrated accelerator mass spectrometry (AMS) radiocarbon dates in low- and mid-latitude sites. In areas of large changes in surface reservoir ages, they were calculated using a combination of radiocarbon dates and alignment tie points between sea surface temperature or magnetic property records to ice core records. We include age models for 44 sites from the original publication, with radiocarbon data calibrated to the IntCal1321 curve, and age models for 48 sites from an update using the IntCal2022 calibration curve. These age models are referred to as W13 and W20, respectively, hereafter.

  • From compilations by Jonkers et al.23 and Repschläger et al.18 we include age models from 151 sites (referred to as J + R hereafter). We combine these two compilations because they share Atlantic sites and methodologies. Most age models are based on AMS radiocarbon dates on planktic foraminifera using the software BACON24 version within the data management toolbox PaleoDataView25 and calibrated to the IntCal1321 curve. Some additional age models in Repschläger et al.18 were calculated using benthic foraminiferal δ18O stratigraphy or using automated alignment with a stacking method described in Lee et al.26.

The database includes several sets of age models calculated for this publication:

  • 41 new age models for Pacific sites calculated based on benthic foraminiferal δ18O stratigraphy aligned to the LR04 stack27 between the LGM and the early Holocene.

  • 17 new age models calculated from AMS radiocarbon dates on planktic foraminifera calibrated to the IntCal1321 curve with the software BACON24 version All parameters are recorded in the database as age model text files. These age models were calculated before the release of the IntCal2022 calibration curve.

  • 211 new age models calculated using the software BACON24 version within the data management toolbox PaleoDataView25. Radiocarbon data were calibrated using the IntCal20 calibration curve22. Prior to calibration and BACON age modeling, a local reservoir age simulated with the Large Scale Geostrophic ocean general circulation model28 over the last 55 ky29 was subtracted. To produce local time series of the total radiocarbon age versus reservoir age, we added the modelled reservoir ages to the IntCal20 radiocarbon ages (by associating the modeled and IntCal20 calendar ages). For each measured radiocarbon age we then selected the corresponding local reservoir age. Specifically, the surface (0–50 m) reservoir age range corresponding to the measured radiocarbon age range from the nearest gridbox in the simulated data were extracted. The downcore age model and its uncertainties is based on 1000 BACON age-depth realizations. All parameters are recorded in the database as age model text files. The sites in this age model ensemble include the 17 sites for which we calculated age models with IntCal13 calibration as described above.

Data Records

Data Availability

The database was developed by the OC3 community, following the FAIR (Findability, Accessibility, Interoparability, Reusability) guiding principles for scientific data management and stewardship30. Conforming to the accessibility principle (the “A”) of the FAIR data standard, the database has been stored in the public repository Zenodo31. This repository allows updates on the database after publication. Future additions of new sites and age models will be uploaded by the OC3 members.

Database description

Sites included in Version 1.0 of the OC3 database are listed in Tables 16, with citations for isotope data and age models. They come from the global ocean and a water depth range between 200 and 5000 m (Fig. 1, top). 98% of sites report stable isotope data from Cibicidoides spp., and 74% correspond to Cibicidoides wuellerstorfi (Fig. 1, middle). We include some sites that report unpublished data obtained from other species, mostly Uvigerina spp. The number of isotope measurements at each site (Fig. 1, bottom) for 23-15 ky BP has a mean of 16 and a median of 12 data points available per record. 84% of sites have a time resolution of at least 1 ky for either the 23-19 or 19-15 ky BP time slices. The remaining sites were included because they either have 1 ky or higher resolution for the subsequent 15-11 ky BP time slice, or because they present new, unpublished data (see Tables 16). We include in Zenodo a table with the number of data points for the 23-19, 19-15, and 15-11 ky BP time slices at each site31. Users may use that tables or software tools that accompany this publication31 to discern, based on temporal resolution and region, which sites to include in their analyses. Binning the data into 500-year time slices between 23 and 15 ky BP, yields 130 to 200 coring sites per time slice (Fig. 2), with a higher number in the ED. Geographically, 63% of sites correspond to the Atlantic, 28% are from the Pacific, and 9% correspond to the Indian Ocean. 12% sites lie in the Southern Ocean (south of 35 °S).

Fig. 1
figure 1

(Top) Positions and depths (in m) of all sites included in our database. (Middle) Isotope data species codes: (orange) Cibicidoides wuellerstorfi, (blue) other Cibicidoides, (black) other benthic foraminifera. (Bottom) Number of data points at each site in the 23-15 ky BP time interval.

Fig. 2
figure 2

Number of sites per 500 y time slice in our data base.

Database structure

The database is organized in different folders, each named after and corresponding to a specific coring site. The folders contain comma separated value (csv) files (Fig. 3). The file format choice makes the files easily machine-readable on computers with different operating systems, conforming to the interoperability principle (the “I”) of the FAIR data standard. It also makes them human-readable, which facilitates access and editing. Each site folder contains at least one of each of the following file types:

  • A metadata file, with ocean basin, site name, latitude, longitude, and seafloor depth.

  • A depth model file with depth scale information.

  • An age data file, with measured age constraints (e.g., radiocarbon) and/or tie points information, including type of age constraints and references.

  • Isotope data files, with δ13C and/or δ18O data on a depth scale, and measurement methodology, taxon, and reference. There can be more than one isotope data file, each corresponding to different taxa, or as new data is added to the site. The different isotope files are identified in their names with dates of addition to the database in year-month-day (yyyymmdd) format, author name, and/or taxon name.

  • Age model files, with depth scale and age determinations, and information on age model type and source. There can be more than one age model file, each corresponding to a different age model. The different age model files are identified in their names with dates of addition to the database in year-month-day (yyyymmdd) format and/or author name.

Fig. 3
figure 3

Diagram of a general OC3 site folder, with its file structure as described in the text.

The csv files are accompanied by unformatted text files where additional information is documented. All files are identified with the same site name as in the database, to conform the findability principle (the “F”) of the FAIR data standard.

In addition to the raw data and age models, we include the reference and when available, name of the laboratory and methodology followed for analysis. For radiocarbon-based age models calculated with the software BACON, we include all parameters used in the calculation in separate age model text files included within each of the site folders. This aims to fulfill the reusability principle (the “R”) of the FAIR data standard. Columns are left blank when the information is not available, but they could be filled in with new version releases and new contributions. The data type and format of each column in the csv files is specified as follows. Missing data are indicated with a blank column. Columns with the “Notes” label in their name are to be used by operators to add unformatted information that they consider relevant. For stable isotopes the units used are permil, in terms of Vienna PDB (VPDB).

  • site_metadata.csv

    Ocean: Pacific, Indian, Atlantic (includes Arctic and Mediterranean).

    Sea: A more specific region, if it corresponds, e.g., South China Sea

    Site: Site name. Corresponding to the name that appears in the files and folder names. For Deep Sea Drilling Project (DSDP)/ODP/IODP sites we use DSDP/ODP/IODP-leg/expedition-site as name convention.

    Latitude (degN): Latitude, with the highest precision possible. Between −90 and 90 °N)

    Longitude (degE): Longitude, with the highest precision possible. Between −180 and 180 °E

    Site Depth (m): Depth of the sea floor below modern mean sea level, with the highest precision possible, in negative numbers.

  • site_depth_model.csv

    Site: Site as in metadata file.

    sample_label: Label of individual sample from original publication, if available.

    hole_label: Label for holes in the site, for sites that include more than one hole.

    section_label: Label of section in the core.

    published_archival_depth (m): In cases where only one core is sampled at each site, this usually coincides with the reported depth in core of the original publication. For sites with more than one core (e.g., IODP sites), it is defined as the value assigned by the estimated depth of the bottom of the drill string below the sea floor, plus the sum of the depths in sections in the cores shallower than the section being analyzed.

    current_depth_model (m): It coincides with the archival depth in sites where only one core is sampled. For sites with more than one core (e.g., IODP sites), the depth model transforms archival depths into true sample depths, considering processes such as compression/expansion during the coring process.

    current_depth_model_note: Any important information on the depth model.

    DEPTH(mid) (m): As defined for IODP cores32.

    MBSF(mid) (m): Meters below sea floor, as defined for IODP cores32.

    MCD(mid) (m): Meters composite depth, as defined for IODP cores32.

    CCSF(mid) (m): Core composite depth below sea floor, as defined for IODP cores32.

    depth_model_1 (m): Spaces to include older depth models. This column is usually filled with a copy of the published_archival_depth (m) column.

    depth_model_note_1: Any important information on depth_model_1.

    older_depth_model_2 (m): Spaces to include older depth models. More columns of this kind may be added if needed.

    older_depth_model_note_2: Any important information on older_depth_model_2.

  • site_isotope_data_yyyymmdd.csv

    Site: Site as in metadata file.

    Sample Label: Label of individual sample.

    archival_depth (m): Archival depth at which data were taken.

    d13C (permil): Benthic foraminiferal δ13C values without any vital effect corrections.

    d18O (permil): Benthic foraminiferal δ18O values without any vital effect corrections.

    d13C_corrected (permil): Benthic foraminiferal δ13C values with vital effect corrections.

    d18O_corrected (permil): Benthic foraminiferal δ18O values with vital effect corrections.

    Number of shells: Number of shells measured.

    Minimum mesh size (um): Minimum mesh size used for (dry) sample sieving prior to picking.

    Maximum mesh size (um): Maximum mesh size used for (dry) sample sieving prior to picking.

    Taxon: Taxon of sample, e.g., Cibicidoides wuellerstorfi.

    Taxon_flag: A number that identifies the species. See Table 7 for the list of taxon flags.

    Taxon_note: A note on the taxon.

    Taxon_note2: Space for notes on taxon or methodology.

    Taxon_note3: Space for notes on taxon or methodology.

    Additional_note: Note on methodology.

    Publication source: Publication from where data were obtained.

    Original reference: Original publication associated with the data.

    File name: File name in original repository.

    Data source: Publication where data is found. Usually a Digital Object Identifier (DOI).

    Quality control: 1 means that the data has been quality controlled as described in the data acquisition section. 0 means that the data were defined as an outlier or bad data in the quality control process.

  • site_age_data.csv

    Site: Site as in metadata file.

    Sample label: Label of individual sample.

    sample_depth: Depth in core (meters below the sea floor) for the sample, in meters.

    technique: Method used to calibrate age data into calendar age.

    lab. code: Identifying code of the laboratory where the age data were taken.

    species/material: Species or type of material used for age measurements.

    radiocarbon_age (y): Measured conventional radiocarbon ages (using Libby’s half-life).

    radiocarbon_age_error_plus (y): Uncertainty of the radiocarbon dates in the positive direction.

    radiocarbon_age_error_minus (y): Uncertainty of the radiocarbon dates in the negative direction.

    reservoir_age (y): Estimated reservoir age used to calculate the calendar age

    reservoir_age_error_plus (y): Uncertainty of the estimated reservoir age in the positive direction.

    reservoir_age_error_minus (y): Uncertainty of the estimated reservoir age in the negative direction.

    calendar_age (y BP): Calibrated age.

    calendar_age_error_plus (y BP): Uncertainty of the calibration in the positive direction.

    calendar_age_error_minus (y BP): Uncertainty of the calibration in the negative direction.

    calibration curve: Calibration curve used to calculate calendar ages (e.g., IntCal13; IntCal20).

    note1: Unformatted information considered relevant.

    note2: Unformatted information considered relevant.

    original reference: Reference on the age data and/or the calibrated age.

    data doi: age data DOI and/or reference.

  • site_age_model_yyyymmdd.csv

Site: Site as in metadata file.

Sample Label: Label of individual sample.

age_model_depth (m): Depths at which the age model is calculated.

age_model (y BP): Modeled calendar age.

age_model_sigma_plus (y BP): Uncertainty of modeled age in the positive direction.

age_model_sigma_minus (y BP): Uncertainty of modeled age in the negative direction.

upper_95_percent (y BP): 95% confidence level of modeled age in the positive direction.

lower_95_percent (y BP): 95% confidence level of modeled age in the negative direction.

age_flag: Number flag indicating age model method. See Table 8.

Table 8 Age model flags associated with different methodologies included in the data base.

age_model_note: Any note on the age model.


quality control: 1 means that the data has been quality controlled as described in the Data acquisition section.

All file names begin with a string referring to the core site that matches the site name in the metadata files. Isotope data and age model files also include a date in their names, which corresponds to the date at which the information was added to the database, and it is written in yyyymmdd (year-month-day) format. If more than one isotope data and/or age model is available for a particular site, separate files with different dates are created for each one. For sites that include isotope data and/or age models from other syntheses, additional isotope data, age model, and depth model files are included in the corresponding folders, with a distinctive string added to their names. In cases where more than one species was reported for a site, we keep the isotope data and age model associated with each species in separate files, with the species specified in the file names. The name structure and use of csv files in the database allows the user to make specific updates. New isotope data and age models can be easily added, using the date format described above.

Technical Validation

Time slice comparison

Despite its sparsity, the coverage of the database resolves the general structure of deep water masses in depth-latitude plots (Fig. 4). During the LGM, the North Atlantic shows high benthic foraminiferal δ13C values in the North Atlantic above 2500 m, associated to the glacial equivalent North Atlantic Deep Water9 (NADW). Deeper Atlantic waters exhibit lower δ13C values related with a mixture of glacial NADW and Antarctic Bottom Water. In the Pacific, δ13C-depleted Pacific Deep Water can be distinguished, as well as shallower, δ13C-enriched waters in the Southern Ocean associated with the transport of Antarctic Intermediate Water.

Fig. 4
figure 4

(Left) Zonally-collapsed Cibicidoides δ13 C values from our database for an LGM time slice (21-19 ky BP). (Right) Cibicidoides δ13 C difference between a deglacial time slice (17-15 ky BP) and the LGM. In order to calculate differences between the time slices, all data were binned into a latitude-depth grid of 5°×200 m resolution. The most recent age model available at each site was used to make this plot. The lengths of the time slices were chosen such that both were 2000 y long.

In the Atlantic, compared with the LGM, deglacial benthic foraminiferal δ13C values from the 17-15 ky time slice (Fig. 4, right) is lower in northern-component waters (above 2500 m) and higher in most sites in regions of southern-component waters. This is in agreement with previous reconstructions19,33,34, and consistent with Atlantic Meridional Overturning Circulation shallowing and accumulation of respired carbon in deep waters35. Benthic foraminiferal δ13C is also higher in the Pacific and Indian Oceans in the 17-15 ky time slice time slice compared with the LGM.

Concerning benthic foraminiferal δ18O values, inter-laboratory calibration offsets of several tenths of a per mil complicate the analysis of anomalies36,37, proving it difficult to have a quantitative measure of LGM-deglacial changes. However, a decrease is observed in most regions between the 17-15 ky time interval and the LGM (Fig. 5). This decrease reflects deglacial changes in temperature and δ18O values of deep waters38.

Fig. 5
figure 5

As Fig. 4, but for Cibicidoides δ18 O data.

Age model comparisons

The OC3 database includes sites with more than one age model (Tables 16), allowing an evaluation of the sensitivity of the reconstructed time evolution of benthic foraminiferal δ13C and δ18O values with respect to different age models. Such analysis gives insights into the bias associated with age model uncertainties and enables us to investigate the robustness of leads and lags between deglacial stable isotope records.

We include in the Zenodo repository31 plots of benthic foraminiferal δ13C and δ18O values versus age of all sites. The lags between age models are not constant through the LGM and ED (e.g., South Atlantic site MD07-3076Q in Fig. 6) with lags generally comprised between 0 and 1 ky. Even for sites where lags of the order of 2 ky exist (e.g., North Atlanitc site SO82-5-2, Fig. 6), there is overlap among the uncertainty intervals of the age models, meaning that differences in timing are likely smaller than the uncertainties of the respective age estimates.

Fig. 6
figure 6

(Left) Time series of Cibicidoides δ13 C values calculated with age models from different compilations, as indicated. Age uncertainty bars are included. (Right) Age models as a function of depth in core. Shadings correspond to the reported age uncertainties, based on 95 % confidence intervals. Two example sites from the database are displayed: (top) MD07-3076Q and (bottom) SO82-5-2. Plots for the rest of the sites are included in the Zenodo repository31. Age models labeled as “calculated by Stefan Mulitza” were calculated for this work from radiocaron dates with the IntCal20 calibration curve, as explained in the Methods section (OC320 age models in Tables 16).

To further assess the impacts of age model on the data assessment, we calculated the correlation coefficient R and root mean square error RMSE at each site, between the benthic foraminiferal (Cibicidoides) δ13C time series generated for this work from 14C-calibrated age models (labeled as OC320 in Tables 16) and with other age models, namely the J + R, W20, and P (previous compilations). The time window chosen for this analysis is 23-15 ky BP, and mostly Atlantic Ocean sites are used, since most sites with multiple age models are situated there (Fig. 7). To allow the calculation of correlations and RMSE, all data were linearly interpolated to a regular age grid with a 500 y time step. Other time steps were trialed (100 and 1000 y), yielding no different results. Correlation coefficients have values higher than 0.60 in 73% and 54% of the sites for the comparison of OC320 with the W20 or P age models, respectively. The comparison of Cibicidoides δ13C time series generated with the OC320 and J + R age models yields correlation coefficients higher than 0.60 for 75% of the sites, highlighting the high compatibility of 14C age models that use the same methodology. Discrepancies in several North Atlantic sites, that lead to low and even negative correlations between time series (Fig. 7, left), are due to surface reservoir age differences among age model approaches. The comparison among time series calculated with either of the age models yields RMSE values lower than 0.3 permil in 90% of the cases (red circles in Fig. 7, right panels). The discrepancies among time series of Cibicidoides δ13C values associated with the use of different age model approaches are thus generally lower than estimates of LGM-Holocene changes in benthic foraminiferal δ13C values (0.38 permil39).

Fig. 7
figure 7

Map distribution of correlation coefficient R and RMSE of Cibicidoides δ13 C from OC3 sites between 23 and 15 ky BP calculated with the age models from this work (OC320 in Tables 16) and (top) J + R; (middle) W20; (bottom) P age models.

Another approach to assess age model uncertainty is to compare time slices generated with the same data, but with different age model approaches. We compare sites with radiocarbon age models calculated for this publication (OC320 in Tables 16) and other age model compilations. We calculated at each site the Cibicidoides δ13C difference between the 21-19 and 17-15 ky BP time slices (Fig. 8). Due to the scarcity of records in other basins, the analysis is limited to the Atlantic Ocean. The Cibicidoides δ13C time slice difference calculated using OC320 age models is similar in spatial structure to the time slice differences calculated using J + R, W20, and P age models (comparison of left- and right-side plots in Fig. 8). Correlation coefficients are 0.83, 0.75, and 0.90, respectively. This reflects a high agreement in the direction of deglacial changes in δ13C values, irrespective of which age model is used. The corresponding RMSE′s are 0.20, 0.19, and 0.13 permil, which is of the same order of magnitude as the differences in δ13C values between the two time slices at each individual site (Fig. 8). This indicates that the resulting magnitude of Cibicidoides δ13C changes between time slices may differ considerably when using different age model approaches. We repeated the analysis for the single 17-15 ky BP time slice, without calculating a time slice difference (Fig. 9). In that case we get correlation coefficients higher than 0.9 for the three Cibicidoides δ13C time slice comparisons, and RMSE′s lower than 0.20 permil. The result reflects that Cibicidoides δ13C in single time slices may be less dependent on the age model approach than the difference between Cibicidoides δ13C values from different time slices.

Fig. 8
figure 8

Comparison of latitude-depth Atlantic sections calculated for the difference between the 21-19 and 17-15 ky BP time slices. First row of plots: Cibicidoides δ13 C time slice calculated with OC320 (as in Tables 16) age models at sites where both OC320 and J + R age models are available. Left(right) plot shows the time slice calculated with OC320(J + R) age models. Second(third) row of plots: Same as top plots but for OC320 and W20(P) age models. Data were binned to the same grid than in Fig. 3. Correlation coefficients and RMSE are (OC320 and J + R comparison) 0.83, 0.20 permil; (OC320 and W20 comparison) 0.75, 0.19 permil; (OC320 and P comparisons) 0.90, 0.13 permil, respectively.

Fig. 9
figure 9

Comparison of latitude-depth Atlantic sections calculated for 17-15 ky BP time slice. First row of plots: Cibicidoides δ13 C time slice calculated with OC320 (as in Tables 16) age models at sites where both OC320 and J + R age models are available. Left(right) plot shows the time slice calculated with OC320(J + R) age models. Second(third) row of plots: Same as top plots but for OC320 and W20(P) age models. Data were binned to the same grid than in Fig. 8. Correlation coefficients and normalized RMSE are (OC320 and J + R comparison) 0.93, 0.18 permil; (OC320 and W20 comparison) 0.93, 0.19 permil; (OC320 and P comparisons) 0.97, 0.11 permil, respectively.

The above analyses illustrate that the OC3 database coverage is sufficient to resolve deep ocean water mass features through time. The number of sites in the Pacific and Indian Oceans is still considerably lower than in the Atlantic Ocean, and future versions of the database will focus on improving the coverage for those basins. An analysis of stable isotope distributions through the LGM and ED, whose time dimension were calculated from different age model approaches, shows that the direction of changes may be captured, irrespective of the age model approach used, but the magnitude of those changes differs among age model approaches. The database features allow to construct a four-dimensional picture of stable carbon and oxygen isotopes through the LGM and last deglacial periods. The included software tools31 allow quick calculations and the selection of sites for data analysis or model-data comparisons.

Usage Notes

The choice of csv format for the OC3 database allows accessibility from a wide variety of computer software, and very light computational needs. In order to facilitate analysis, we have created a number of python programming language scripts that perform tasks for users. Because the scripts are equipped with simple user interfaces, no knowledge of python is required.

The python scripts are included in the repository Zenodo, in the same location of the dataset31. They are simultaneously compiled and run by entering, in the command line (Windows systems) or terminal (UNIX systems), “python”, where scriptname refers to the name of the chosen python script. The minimum python version required is 3.6. The scripts run locally. In order to retrieve OC3 data, the entire or parts of the OC3 database needs to be downloaded to the local system. In order to run, the scripts need a number of python packages to be installed. The packages needed for each script are listed in the repository31.

The scripts provided for analyzing the OC3 database are as follows:

  • This script retrieves the position and site name metadata of a region of interest (defined by longitude, latitude and depth ranges) and lists them in a single csv file. This allows users to quickly visualize the position and basin information of all sites in a chosen region.

  • and These scripts retrieve the data and age models from the OC3 database location and create time series plots (encapsulated postscript (eps) files) of benthic foraminiferal δ13C and δ18O values, respectively, with all age models available for each of the sites. The name of the site and the benthic foraminifera species are displayed in the time series images. Age model uncertainties are displayed as error bars when available.

  • This script grabs the isotope data from the OC3 location, and lets the user choose one of the available age models to linearly interpolate to the isotope data’s depth-in-core scale. Once the age model is chosen, the script generates a folder of merged csv files with position, age, isotope data, and taxon information for each site. The number of rows of all columns in each generated file is the same, in order to facilitate access with any data analysis software. The following python scripts included with the database make use of the merged csv files generated with this scirpt:

    • This script lists the number of data points at each site inside a predefined time slice. The result is saved in a csv file.

    • This script lets the user define a taxon group (Cibicidoides wuellerstorfi, any Cibicidoides, or all taxa), a time interval, and a region of interest (defined by longitude, latitude and depth ranges), and calculates the time mean of the benthic foraminiferal δ13C and δ18O data for all sites that include data in the defined time interval and region. The result is saved in a csv file, and plotted in longitude-latitude, latitude-depth, and longitude-depth two dimensional scatter plots. The images are saved as eps files.

    • This script lets the user define a taxon group as in the previous script and two time intervals. It plots, in latitude-depth sections for each basin, the benthic foraminiferal δ13C or δ18O data from the first time slice (left panels), and the benthic foraminiferal δ13C or δ18O difference between the second and first time slices (right panels). The images are saved as eps files. In order to calculate the differences and visualize, the scripts bins the data positions into a regular 5°×200 m grid.

For authors who are not familiar with running python scripts, we also include in Zenodo31 merged files (in csv format) that contain metadata, depth, age model, and isotope data for all sites. We include one merged file for each of the age model groups available.