see PDF

Review of Baudouin et al. 2024

Kaustubh Thirumalai University of Arizona

Baudouin et al. employ pseudo-proxy experiments to test the sensitivity of global mean surface temperature (GMST) reconstructions to paleoclimate proxy uncertainty using output from several transient climate simulations over the last glacial cycle. They utilize the sedproxy forward model to distort temperature timeseries from the simulations in quasi-realistic ways to mimic the processes involved in the preservation and extraction of proxy signals from paleoclimate sensors. Focusing on a recent and comprehensive collation of sea-surface temperature (SST) records over the past 130 kyrs (Jonkers et al. 2020—J20), the authors perform several statistical resampling experiments on the simulated SSTs, parsed by various metadata parameters within the J20 dataset (number of records, proxy type, location, etc.) Their main aim is to assess the sensitivity of the uncertainty in reconstructing GMSTs over the past 130 kyr and the role of each of the investigated proxy metadata parameters. Broadly, the authors conclude that whereas orbital-scale variability of GMST is accurately retrievable over this time interval, centennial-to-millennial scale GMST reconstructions are “unreliable,” primarily due to age-model uncertainty and the lack of detailed site-based sedimentological parameters. Overall, I enjoyed reading this manuscript and found their analysis and discussion convincing. I believe that this work presents an advance to our sub-discipline. Below, I detail some major and minor comments/suggestions to improve the scope of the authors’ work and potentially increase the accessibility of their manuscript.

Major Points

1. Bioturbation mixing depth parameterization: The authors conclude that “Our results show also the existence of a trade-off between the inclusion of a large number of records, which overall reduces the uncertainty, and of only the highest quality records (low age uncertainty, high accumulation rate, high resolution), which improves the reconstruction of the short timescale” where the latter refers to reconstructions being relatively free from bioturbation- and age-model-related smoothening to sufficiently preserve short-timescale (or multicentennial-to-millennial) climatic signals. The manuscript (Table 1 and Figure 2 caption) suggests that the authors used a constant (presumably?) sediment mixed-layer depth (Table 1 only indicates ‘bioturbation’ and would benefit from more explicit details about what is being parameterized) of 10 cm. In my opinion, and according to my cursory assessment of global bioturbation rates (I realize that Dolman & Laepple, 2018 argue otherwise—but I’d like to see the numbers) of the data presented in Teal et al. (2010)—10 cm is far too high for average global mixing depths, particularly given J20’s bias towards tropical and near-coastal proxy locations. I would like to see how a value of, for example, 5 cm would perform for the Full PP and related experiments. I understand that the authors have attempted to parse the sensitivity of ‘age uncertainty’ versus ‘bioturbation’ and other associated parameters in their analysis, but this does not address the entire hierarchy of choices with a lower bioturbation rate. If feasible, I’d recommend that the authors perform such an experiment (Full PP with reduced bioturbation rate) and check whether more high-frequency information is retained in the associated spectra.
2. Clarity regarding the use+utility of J20: Unless I am mistaken (which is highly likely), the authors do not actually use or show the GMST calculated (using any algorithm) from the reconstructions collated in J20. They merely use the metadata of proxy parameters within J20 as a framework for their pseudo-proxy experiments. The clarity of the manuscript would be improved if the authors were more upfront about this aspect in their abstract and introduction. On the other hand, this also leads me to question this omission as potential comparisons between reconstructed (from data), simulated, and resampled GMST calculations would be highly interesting. However, I recognize that this may be beyond the scope of this work—accordingly, I feel that the authors should preface this aspect and consider using the term metadata in their abstract and text. I think that the title of the manuscript should include/reflect ‘sensitivity/uncertainty experiments’ and/or ‘pseudo-proxy experiments’ because in its current form, without comparing simulated and actually reconstructed GMST, I do not think the authors can claim to test the ‘reliability the reliability of global surface temperature reconstructions of the last glacial cycle’. Rather, they are testing the reliability of the methodologies used to create global mean reconstructions… hence I feel that a title revision is needed.
3. Inclusion of ‘Full PP at random locations’: It appears that the authors do not show any results from this experiment, yet it plays an important role in their analysis (see, e.g., Lines 335–340: “In addition, location resampling and latitude band configurations, which aim to account for it, are not large enough to cover the bias. Yet, the pseudo-proxy experiments using random proxy locations can reproduce the simulated MSST.”) I recommend that the authors make a plot showing results from this experiment and to be more quantitative/precise regarding the ability or lack thereof of these simulations to reproduce simulated mean SSTs.
4. Attribution of a specific set of J20 locations as a significant bias: Based on the last point, the authors state that “Therefore, the bias is caused by the specific set of locations in the J20 dataset: there is an over-representation of regions with strong LGM cooling”. Whereas this assessment may be accurate, I do not find the regions that the authors identify to be a convincing explanation (e.g., NW Atlantic/Kurushio extension)—instead, it seems to me that this is a bias related to the relative proximity of core locations to continents—where land-based cooling strongly impacts these sites as opposed to open-ocean marine conditions. Is it possible for the authors to combine inferences from the ‘Full PP at random locations’ or another sensitivity experiment to test this possibility?

Minor Suggestions

• Line 81: “…needed to develop our evaluation standards.” Please rephrase.
• Lines 91–93: Are there only 7 (112–105) unspecified datasets? Figure 1 says otherwise—please clarify.
• Lines 162–163: Please consider adding more information to contextualize why the following steps are being performed. This would be a great spot to clarify the involvement/extent of usage (or lack thereof) of the actual records within J20.
• Lines 236–237: 0.26 K and 0.23 K seem to be exceedingly low values for analytical uncertainty. Does this take into account sampling uncertainty (see e.g. Thirumalai et al. 2013) which is the uncertainty that foraminiferal shells (with lifespans of a ~month) would have grown at different points of time within the sampling interval, and thus will have uncertainty in reconstructing the ‘interval mean’? If sedproxy takes this into account, it would be good to mention this aspect.
• Lines 245–246: Have you considered performing a depth-sensitivity test? i.e. instead of the uppermost grid location, what about the integration of the top 50 m—which is a more realistic scenario for the proxy integration of temperature signals for the chosen sensors. Perhaps this also might explain the cool bias?
• General comment on Figures: Where there are many lines depicted in figures, it is very difficult to parse the colors of each line (especially on the spectral plots) and to associate them with the legend. I would strongly consider using a different backdrop color or thicker lines with a different subplot layout to more cleary delineate results from various experiments.
• Discussion and Lines 545–555: The authors should consider discussing the values of scaling utilized in Clark et al. (2024) and how they fit within this portion of the discussion.

References

Teal, L. R., Bulling, M. T., Parker, E. R., & Solan, M. (2010). Global patterns of bioturbation intensity and mixed depth of marine soft sediments. Aquatic Biology, 2(3), 207–218. https://doi.org/10.3354/ab00052

Thirumalai, K., Partin, J. W., Jackson, C. S., & Quinn, T. M. (2013). Statistical constraints on El Niño Southern Oscillation reconstructions using individual foraminifera: A sensitivity analysis. Paleoceanography, 28(3), 401–412. https://doi.org/10.1002/palo.20037