Review on:

Antarctic ice sheet paleo-constraint database

Benoit S. Lecavalier, Lev Tarasov, Greg Balco, Perry Spector, Claus-Dieter Hillenbrand, Christo Buizert, Catherine Ritz, Marion Leduc-Leballeur, Robert Mulvaney, Pippa L. Whitehouse, Michael J. Bentley and Jonathan Bamber

Earth System Science Data https://doi.org/10.5194/essd-2022-398

general comments:

Lecavalier et al. present a compilation of 7 different observational data types for the Antarctic Ice Sheet covering a period from present day, Holocene and the last deglaciation with a sparse coverage of the last glacial cycle. The database is made available as an Excel spreadsheet (via the open data repository Ghub and/or as Supplement) and can be used to constrain Antarctic Ice Sheet reconstructions, or in general to calibrate paleo model spin-ups, as basis for stability analysis or future sea level projections. The database also encompasses two present-day data sets for the ice sheet and bed topography (BedMachine Antarctica v.2, Morlighem et al., 2020) and surface velocities (MeaSUREs, Mouginot et al., 2019), both available at NSIDC (free NASA Earthdata Login account is required to access these data), and several related key metrics are named.

The AntICE2 database is an extension of the AntICEdat, published 10 years ago by Briggs and Tarasov (2013), and comes with specified observational uncertainties. Briggs et al., (2014) proposed a methodology by which the database can be applied to evaluate (score) an ensemble of modeled Antarctic Ice Sheet reconstructions by using observational error models and data-weighting to address the heterogeneous distribution of the data in space and time. To my knowledge, a few ice sheet model ensemble studies used a subset of the first iteration database (in combination with GPS and RAISED data, e.g. Pollard et al., 2016, Albrecht et al., 2020b) to evaluate model-data misfit. But I very much support the need for a general integration of such a constraint database and the systematic exploration of structural model uncertainties in a proper uncertainty assessment (Tarasov and Goldstein, 2021).

The updated dataset has been improvement in (at least) three ways: Two new data types (GPS and borehole temperature profiles) have been integrated, newly available data have been merged (more than 1000 direct observational constraints), and the whole dataset has been consistently re-calibrated and curated (quality-assessed). Some of the data types are based on various source publications, for instance the relative sea level indicators (paleoRSL), with valuable updates in locations like Dronning Maud Land and the Amundsen Sea sector. Other data types build on other open databases, for instance exposure histories (paleoH), which are based on the informal and inclusive ICE-D database, which still suffers from inconsistencies. Grounding line retreat proxies (paleoExt) are based on the ICE-D marine database and now include the RAISED consortium compilation (Bentley et al., 2014a).

The data description paper explains the selection and quality criteria adequately. It is really helpful for modelers, in particular as is hard to find, sort and evaluate the quality and consistency of the data quality, when not trained in geology. The corresponding article describes the evaluation processes in a well-structured form. The figures are informative. I would have some minor practical recommendations on the structure and format of the dataset (see below).

AntICEx may become an ongoing project open for new data to come. The article addresses potential future data types and reasons for rejecting data types. It also suggests standards of experts quality control on new proxy data, which would simplify the process of updating datasets. I highly support the publication of the AntICE2 database, this is extremely valuable work!

specific comments:

The paleo data is categorized by site (four digit identifier for data type, basin sector, site within each sector) and include location data (latitude, longitude), age data and an indication of the relationship between the proxy observation and the constrained characteristic.

From a modelers perspective there are quite some steps to take from the pure dataset to the evaluation of model simulations or even model calibration. And I wonder what could simplify this processing already on the data level. For instance, the most relevant data columns (e.g. SITID, SSID, DBCD, LON, LAT, VAL, VALU, ACAL, ACALL1U, ACALU1U, CTY, TIERS) could be re-ordered in a consistent way for each data type.

The format of an Excel Sheet (with size of < 1 MB) is easy to access for a wider community, but there might by other formats (e.g. the human-readable JSON), which are easier to access with data processing tools in python etc. There are also quantitative tools like the Automated Timing Accordance Tool (ATAT version 1.1, Ely et al., 2018), which use gridded geochronological data to evaluate model-data misfit (e.g. with a GIS software package). There are obvious shortcomings in the accuracy of this approach, but such an intermediate step may help to bridge first obstacles for potential data users. The focus of this paper is on the data selection, but such practical applications may be worth to mention.

The databases also encompasses two present-day data sets for the ice sheet and bed topography (BedMachine Antarctica v.2, Morlighem et al., 2020) and surface velocities (MeaSUREs, Mouginot et al., 2019). Both datasets are available at NSIDC (free NASA Earthdata Login account is required to access these data), a link should be added to the article, or at least to the citation. From the article it is not immediately clear, that the two datasets are just selected to be part of AntICE2 with assigned tiers and several key metrics named (maybe add the reference to Briggs et al., 2014 here), but are not actually provided for download (as part of this data publication). The BedMachine data are already available as Version 3 (https://nsidc.org/data/nsidc-0756/versions/3, https://nsidc.org/data/nsidc-0484/versions/2).

The article describes three different quality levels (tiers). What is lacking is some kind of recommendation, which minimum classification to use in model evaluation. Should one avoid tier 0 (and -1) in any case? For instance, can we trust the 6 tier-2 exposure ages at site 1701, spanning 300m of ice thickness change within 2kyr time? Or should only the highest quality (tier 1) be considered?

technical corrections:

l.71: “… the PD ice sheet geometry (surface elevation, ice thickness, basal topography) and surface ice velocities are provided.” → Better indicate, that those datasets are available elsewhere and have been selected in the quality assessment.

Fig. 3B: How is this uncertainty range (grey shading) constructed?

l.150: Provide a URL to ICE-D for consistency.

l.172: When referring to certain data locations in the text (e.g. the correctd paleoH in the Weddell Sea sector), please provide the site ID?

l.223: Please reorder to formulate a sentence?

l.280: It could help readers without a background in GIA modeling to mention that RSL comprises both the bed rock uplift but also the near-field and far-field sea level changes in response to global ice sheet changes, and that after the deglacial RSL rise, data are only available in the following period of delayed bedrock adjustment.

l.304: I recommend to add some clarifications for non data scientists on the radiocarbon age re-calibration and the marine reservoir correction.

l.355: Here, you could also provide an example of the second case, WDC and Byrd?

l.445: Always good to define the Holocene age period.

l.461/Fig.8: A median may provide a more intuitive metric of this long tail statistical distribution, although statistical mean and standard deviation may be consistently used in the error model.

Sect. 3.1.: It could be helpful to follow the quality assessment in an example figure, or refer to already existing Figures (e.g. Fig. 5). Or maybe add a check table of criteria and sites for a better overview.

Table 1: Where does the factor 3 come from in the boreTemp data?

l.660 I very much support to add age structures as soon as possible (AntICE3), such an accurate ice age tracing module has been implemented in PISM recently.

l.728: I understand, this is not a complete list of online data repositories, but Pangaea may also provide some data to be added (https://www.pangaea.de/?t=Cryosphere&f.location%5B%5D=Antarctica)

l.775: Update the to the final publication (part 2?): The Cryosphere, 14, 633–656, 2020, https://doi.org/10.5194/tc-14-633-2020

data:

It could be helpful to have a column (AntICEdat) that provides a quick check on which data have been updated with respect to Version 1 (as done for PaleoRSL).

I would wish a consistent reordering of the most relevant data columns, as indicated above, for easier data processing. The first 12 columns should include the calibrated age, while the raw data (e.g. uncorrected and corrected age) can be put further to the right of each sheet. The age uncertainty could be consistently defined as lower and upper bound (as for PaleoRSL), even when the uncertainty range is symmetric around the mean. Not all data types have a sample identifier SSID, which would be helpful to add..

The PaleoRSL entry 9602 has a wrong sign in lon and lat.

Hat does a Tier of -1 would mean (for the consideration in model-data evaluation), as for instance paleoH site 1103?

References not already mentioned in the article:

Ely, Jeremy C., Chris D. Clark, David Small, and Richard CA Hindmarsh. "ATAT 1.1, the Automated Timing Accordance Tool for comparing ice-sheet model output with geochronological data." Geoscientific Mode Development 12, no. 3 (2019, 933-953)

General comments

The manuscript presents the AntICE2 database, an updated database of observational constraints on the historical evolution of the Antarctic Ice Sheet (AIS). The new database improves upon the previous version not only by including a much greater number of records, but also by classifying each record and explicitly quantifying the uncertainties.

Observational constraints are critical to constrain our understanding of the dynamics of the AIS, including the ice sheet models used to predict its future evolution. The authors correctly note that such constraints are currently lacking, particularly in the interior of the continent. The AntICE2 database is therefore an extremely valuable resource to the community, not just in order to improve our understanding of the past evolution of the AIS but also to improve future projections. The database is therefore of considerable scientific importance, and arguably even of broader socio-economic importance.

The authors are therefore to be commended for their efforts in constructing the AntICE2 database and in making it available to the community. I consider that the manuscript is clear, comprehensive and well written. I recommend it for publication, subject to consideration of the minor comments provided below.

Specific comments

Lines 29-31 and 737-740: Stronger statements could be made here, particularly in the Abstract. From the perspective of ice sheet modelling, databases of this nature provide much more than just accurate spin-ups. Reliable reconstructions of past states of the AIS are also an essential part of the process of calibrating and evaluating ice sheet models. The authors could state this, as well as stating that the implementation of the AntICE2 database therefore has the potential to generate more accurate predictions of the future evolution of the AIS and, equally importantly, more accurate quantification of the uncertainties in those predictions.

Line 59: The text could say “to evaluate and calibrate ice sheet models” (or similar) rather than just “for ice sheet models”.

Line 150: Is there no reference or link of any description for the ICE-D database? Could the authors elaborate at least a little more on factors such as the contents, location and maintenance of this database, and whether/how members of the community might be able to access it?

Lines 165, 302, 303 and 364: Are the authors able to provide references or other justifications for the assumed default uncertainties of 10m, 1m, 1m and 0.1 degrees Celsius respectively? Otherwise they seem to be excessively arbitrary.

Section 3.2: The depth-age relationships derived from individual ice cores still provide a valuable constraint for ice sheet models, even in the absence of radiostratigraphy or any other form of spatial extrapolation. Hence I would strongly encourage the authors to consider including this data in future versions of the database.

Technical corrections

    • Line 184: Add a comma after “local”.

    • Line 223: Replace “Sometimes resulting” with “Sometimes this results”.

    • Line 234: Add a comma after “rich”.

    • Line 400: The word “associated” might be better than “affiliated”.

    • Line 438: Replace “On the modeller side” with “From the perspective of modellers” (or similar).

    • Line 495: Replace “most constraining power on” with “greatest power to constrain”.

    • Line 631: Remove the word “internally”.

    • Line 645: Replace “being” with “represent”.