This is my second review of Herzschu et al’s manuscript “LegacyClimate 1.0: A dataset of pollen-based climate reconstructions from 2594 Northern Hemisphere sites covering the last 30 ka and beyond”.
In general the authors have done a good job incorporating the comments raised in the previous round. However, I would recommend that more of the response to the review is included in the main text as I think this would make the reasoning for several analyses steps easier to follow.
There are also some points that require additional attention. Most of these I already raised in the previous round of review and I think the manuscript would still benefit from some more discussion and guidance for the user of these data.
The choice of reconstruction methods remains a bit vague and could go beyond stating that these are the most commonly used methods. Surely the authors have good scientific reasons to use these. The argument that with providing the raw pollen data, every one can make their own reconstruction seems to either downplay the importance of carefully evaluating such reconstructions or make the current manuscript superfluous. As such, I find that a bit of a slippery slope and I encourage the authors to clearly state why they chose these methods.
Towards the end of the revised manuscript the authors mention other reconstruction methods and that they could help to “explore a larger fraction of the “method uncertainty” space”. Exactly how this could be done remains unclear and the authors should more clearly outline an approach on how the difference between the methods can be used to evaluate the reconstructions (see e.g. Kucera et al. 2005).
The authors have carried out several analyses which can be used to evaluate the reconstructions and they should be complimented for this rigorous approach. They provide three different reconstructions, information on the transfer function performance (from the CCA), information about the analogue quality, on the significance of the reconstruction etc and indicate that all this information can be used to assess the reconstructions. However, I miss clear guidelines of how this should be done, or what the authors think is the best approach. All these quality measures are provided but in the end not used (e.g. figure 10 contains all time series from a single reconstruction method), so what is the point exactly? I suggest that the authors provide instructions on how to use the data, e.g. indicate which labda ratios should be omitted, at which analogue distance researchers should ignore the reconstruction, how to use the results significance test, how to interpret the difference between the methods, what to do with sites that show marked human influence, etc. Alternatively, they should provide a single reconstruction for each site (with uncertainty) that they think is best (with of course an explanation of the reason why).
The authors explicitly state that the purpose of these reconstructions is to evaluate climate models, but without clear instructions it is unclear how these data can be used for this purpose. The data certainly cannot be used "out of the box" but require processing. This is fine of course and almost always the case with complex data like palaeoclimate data, but it means that users need to be provided with clear guidelines and examples.
Finally, the authors assess the influence of human land use on their reconstructions by looking at the proportion of certain pollen types in the time series. This seems a reasonable approach, as far as I can judge, but the influence of humans not only affects the time series, but also the pollen data set and the climatic variables used for calibration. Perhaps the calibration even more? This is alluded to on page 34, but not really explored. Is it not possible to remove sites with high human influence from the training set? Or are there other approaches, such as (perhaps an outrageous suggestion) calibrating using deeper/older samples assuming relatively stable late Holocene conditions?
And a related question, to what degree are the climate variables averaged over the period between 1970 and 2000 representative of the conditions during deposition of the pollen?
I also imagine that human influence on the core top data also affects the identification of analogues and wonder if it could be the reason for the low analogue quality in Western Europe and the British Isles (page 35)?
These issues should be addressed in a revision.
Minor comments (page numbers refer to version with tracked changes)
Title: …”and beyond.” Why not state the exact duration of the time covered by the synthesis?
Page 4: Holocene conundrum. Is there still a conundrum, or is there mechanistic value in comparing global mean temperatures. The debate has progressed since Liu et al. (Osman et al. 2021; Cartapanis et al. 2022; Kaufman et al. 2020).
Page 4: “Pollen data are the only land-derived proxy … Quaternary period”. I suggest to be a bit more specific and tone this down. There is no a priori reason why one could not evaluate a model based on a single or a few observations. High spatio-temporal coverage allows one to investigate different aspects, but I don’t think there is a reason to be offensive to other proxy types.
Page 4-5: “MAT and WA-PLS rely on extensive collections of modern training data.” Does that not hold for any transfer-function based reconstruction?
Page 6: PANGAEA is a data publisher, or repository, not a data base.
Page 6: “We restricted the analyses to the 70 most common taxa to reduce computational power…” this seems odd wording, why would you like to reduce computational power. Is demand meant? More importantly, how was this tested? Please provide details.
Page 8: “... to measure how well the target environmental variable is strongly related…” delete strongly.
Page 8: please provide details of what the minDC function actually does. What are these probability thresholds and what are they based on?
Page 9: significance test. I asked before, how were the random environmental fields generated? Was it by simple permutation or was spatial autocorrelation taken into account. If the environment is spatially auto- correlated, which I imagine is likely the case, a red-noise null should be used instead of the default white noise null.
Page 9: “In addition, we calculated the correlation between WA-PLS reconstruction of …” please provide a rationale for this analysis that helps to understand why this has been done and how the results should be interpreted.
Page 9: “To ease data handling, the dataset files are separated into…” I disagree that splitting a single file into more than one eases data handling. I would argue that it is easier to filter a data set than combine different ones as I don’t need to load multiple files.
Page 11: “Minimum dissimilarities between modern pollen assemblages and fossil pollen assemblages for each site for MAT” Why not provide this information for each sample, rather than for each site? Like this it does not allow for meaningful filtering of the data since part of a time series may have poor analogues.
Page 11: Located in instead of located from.
Page 12/Fig 1: some of the fossil samples seem to come from marine sites (e.g. the Bay of Biscay or the Caspian Sea). Is that correct and if so, what do these time series tell us about vegetation and climate at that location?
Page 14/Fig 3: and what do values below 1 mean, after all the scale goes down to 0, implying that for some sites none of the variance is constrained?
Page 21/Fig 6: better to show the full range of p values. Or somehow indicate where the sites are that did not pass the test as we can assess if there are any spatial patterns.
Page 21/Table 2: there are no values for MAT.
Page 22: “only in single records” please rephrase. Single means one.
Page 22: “High Plantaginaceae correlate with low TJuly in Central Europe indicating potential biases” please explain why only negative correlations are a problem, the reasoning is unclear to the non-pollen specialist.
Page 23/Fig 7: the colour scale is not colourblind friendly.
Page 28/Fig 10: which method was used for the reconstructions shown here?
Page 34: “We a priori selected … Tann and Tjuly” it would be good to move this section to earlier in the discussion.
Page 36: “Climate reconstruction data sets like LegacyClimate 1.0 thus…” this is not a logical conclusion, it is only valid for reconstructions with a similar coverage as LegacyPollen 1.0. Reword.
Page 37: “Temperature reconstructions from proxy data indicate peak temperatures during the Holocene Thermal Maximum around 6000 years BP followed by a pronounced cooling trend toward the late Holocene (Kaufman et al., 2020b)” I think this is an oversimplification of the Kaufman reconstruction. They also highlight spatial variability in the Holocene temperature trends. As do Osman et al and Cartapanis et al.
Page 37: “Temperature reconstructions are often derived from sea-surface temperatures as either mean annual temperatures (Birks, 2019; Bova et al., 2021) or global mean surface temperature (Marcott et al., 2013; Marsicek et al., 2018; Kaufman et al., 2020a and 2020b).” That seems an oversimplification of the cited literature. Some of the cited studies included seawater temperature estimates, but not all are exclusively based on SST.
Page 37: “In this respect, it might help…” how confident are the authors about the independence of the Tann and Tjuly reconstructions? In other words, can we really reconstruct seasonality?
Page 37: “So far … hemispheric scale” is that statement still true after publication of https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2022GL099730 by the same lead author.
Page 37: The last two paragraphs of section 5 seem to stand on their own and should be better integrated with the remaining text.
Cartapanis, Olivier, Lukas Jonkers, Paola Moffa-Sanchez, Samuel L. Jaccard, and Anne de Vernal. 2022. “Complex Spatio-Temporal Structure of the Holocene Thermal Maximum.” Nature Communications 13 (1): 1–11.
Kaufman, Darrell, Nicholas McKay, Cody Routson, Michael Erb, Christoph Dätwyler, Philipp S. Sommer, Oliver Heiri, and Basil Davis. 2020. “Holocene Global Mean Surface Temperature, a Multi-Method Reconstruction Approach.” Scientific Data 7 (1): 201.
Kucera, Michal, Mara Weinelt, Thorsten Kiefer, Uwe Pflaumann, Angela Hayes, Martin Weinelt, Min-Te Chen, et al. 2005. “Reconstruction of Sea-Surface Temperatures from Assemblages of Planktonic Foraminifera: Multi-Technique Approach Based on Geographically Constrained Calibration Data Sets and Its Application to Glacial Atlantic and Pacific Oceans.” Quaternary Science Reviews 24 (7-9): 951–98.
Osman, Matthew B., Jessica E. Tierney, Jiang Zhu, Robert Tardif, Gregory J. Hakim, Jonathan King, and Christopher J. Poulsen. 2021. “Globally Resolved Surface Temperatures since the Last Glacial Maximum.” Nature 599 (7884): 239–44. |
