Review Summary

Herzschuh and colleagues present a very nice set of Holocene pollen-based reconstructions of T_ann, T_July, and P_ann from 1676 sites from the Northern Hemisphere extra-tropics in order to characterize the continental, latitudinal, and regional patterns of Holocene temperature and precipitation changes in the Northern Hemisphere extra-tropics. This synthesis study is excellent and allows for the regional heterogeneity of the temperature and precipitation trend to be mapped.

I have four major comments and a few minor questions, as well as some very minor usage suggestions.

I would recommend the paper for publication after correction of these.

Major Comments:

1) The selection of records in the dataset for the Holocene quantitative reconstruction in this paper is unclear. The quality and accuracy of the synthesis studies depends largely on the chronological framework, archive type, sampling resolution of the original fossil pollen records, and so on. I note that those 991 records do cover the full period of 11 to 1 ka, but do not see an evaluation of the age, resolution, and archive type of the selected records. Are there any selection criteria for chronology and archive type in the dataset? For example, how many age control points does the original record contain that will be selected for quantitative reconstruction? And what is the time resolution of each sample? This takes into account that the amplitude of changes in temperature and precipitation reconstructions would vary substantially with the resolution of the proxy record.

In addition, the range and quantity of selected modern sites in the calibration dataset can also affect the accuracy of temperature and precipitation reconstructions, as suggested by the authors. Then how many transfer functions are used to calculate the 991 records in this synthesis study? Does each record need to establish a transfer function, or does it establish by region? Is the spatial range of modern sites in the calibration dataset for establishing transfer function all within a 2000 km radius? Or are there some differences in different continents or regions? Of course, I believe that the trend of paleo-temperature and paleo-precipitation change will not change substantially, but it will affect the comparison of amplitude.

2) I agree with the authors that “Pollen data are one of the few land-derived proxies available that can theoretically contain independent information on both temperature and precipitation in the same record” (Lines 99-101). Therefore, the authors reconstructed the spatio-temporal patterns of temperature and precipitation from a single dataset simultaneously. However, it is a challenge to distinguish the effect and correlation between temperature and precipitation in quantitative analysis. In the section of Methods and Discussions, the author mentions the issue of the impact of precipitation on temperature reconstruction (Lines 143-145, 410-412). Could you give more explanation as to why such an approach would ‘restrict the impact of precipitation on temperature reconstruction and vice versa’? One or two sentences will do.

In addition, how do the effects of temperature and precipitation on each other differ across continents and regions? How is it evaluated in quantitative reconstruction analysis?

3) I do not see the expression of reconstruction uncertainty in Figures 2 and 5. The evaluation of the reconstruction errors is essential for quantitative reconstruction and comparisons of different results. Therefore, it would be appropriate to add each latitudinal reconstruction curve with 1σ uncertainty shaded to the supplementary file.

The number of records (n) for each curve in Figures 2 and 5 also needs to be displayed in the appropriate place. Is the large range of temperature and precipitation variations in North America north of 60°N caused by the number of records (n)?

4) There is still a great controversy regarding the occurrence of Holocene thermal maximum between the proxy temperature reconstructions and climate models, named as “Holocene temperature conundrum”. One of the main controversies for Holocene temperature conundrum is the occurrence of a maximum in mean annual temperature (MAT) during the early to middle Holocene. The term ‘mid-Holocene optimum/late-Holocene optimum’ has been used in this paper, but in some areas, such as East Asia, there is a difference between mid-Holocene optimum and Holocene warm period/Holocene thermal maximum. Mid-Holocene optimum is thought to be a period of high temperature and high precipitation, when vegetation flourishes. The authors should define the mid-Holocene optimum and distinguish it from the Holocene warm period.

In addition, quantitative Holocene temperature records in East Asia (loess, lakes, marine sediments) reveal a clear early to middle Holocene thermal maximum, such as high-resolution Holocene pollen records from Xiaolongwan Maar Lake in northeastern China, Gonghai Lake in northern China, and Huguangyan Maar Lake in southern China. These records show the occurrence of a maximum in MAT during the early to middle Holocene, which does not support the conclusion of this paper that “The concept of a mid-Holocene temperature optimum only applies mainly to the mid and high northern latitudes in the circum-North Atlantic region while records from mid-latitude Asia, Western North America, and all subtropical areas do not fit into this concept but mostly show an overall Holocene increase or other pattern” (Lines 430-434).

Minor Line-by-Line Comments:

Line 202: for “in Europe north of 60°C” consider “in Europe north of 60°N”.

Line 275: for “~0.07K compared to ~0.18K” consider “…°C compared to…”.

Line 336: “…the modern pollen assemblages are not heavily biased by human impact”, please provide relevant literature here.

Line 362: for “from the early to mid-Holocene” consider “from the middle to late-Holocene”.

Figures 3 and 4: The map would be improved by changing some colors and size. Each 2°x2° grid cell was too small to see even zoomed in. Changing sizes of maps and/or colors may resolve this better.

In this manuscript Herzschuh et al present an analysis of the “LegacyClimate 1.0” data product currently in discussion in ESSDD (https://doi.org/10.5194/essd-2022-38, hereafter H2022)). The authors reconstruct hemispheric, continental and regional time series of Holocene annual mean temperature, mean July temperature and annual precipitation using the pollen-based estimates presented in LegacyClimate. They compare the new reconstructions to temperature estimates based on other proxies and investigate spatial variability in the evolution of Holocene temperature and precipitation.

The main conclusion the authors draw is that regional temperature and precipitation trends differ and that there is hence no global Holocene Thermal Optimum around 6 ka and hence no global Holocene Temperature Conundrum. These conclusions echo those based on earlier work (Kaufman, McKay, Routson, Erb, Davis, et al. 2020; Marsicek et al. 2018; Kaufman, McKay, Routson, Erb, Dätwyler, et al. 2020), but are based on a larger number of time series using a single proxy and include precipitation. Whilst these conclusions are intuitively reasonable, questions about the methodology and the uncertainty of the temperature and precipitation reconstructions, prevent me from assessing whether the data actually support the conclusions. I therefore restrict my review to these methodological issues in this review and am happy to discuss the trends and variability in Holocene temperature and precipitation once these issues have been addressed in a revised manuscript.

My first two issues pertain to aspects of the reconstructions that are dealt with in H2022, but ignored here. This renders the two manuscripts to some degree contradictory and needs to be explained:

1. Significance of the reconstructions. Tests in H2022 show that only approximately ⅓ of the reconstructions show a temperature trend that deviates from noise (i.e. where the reconstruction shows a better correlation with the first principle component of the assemblages than 90 % of the reconstruction based on randomised temperatures; Table 2 in H2022). In the absence of any information about this in the method section, I assume that the same proportion holds for the selection of time series analysed here. So, why did the authors not filter out these records, as was for instance done  in previous work (Marsicek et al. 2018)? As it stands, the analysis presented here is based on reconstructions that are for approximately 66 % noise. Thus the authors really need to convince the reader why they ignore their own previous analyses and present the evidence they have that these reconstructions are valid. One obvious way to do so would be using sensitivity tests and to assess to what degree the observed trends are sensitive to the significance of the individual time series. (If on the other hand, the authors argue that these tests are not meaningful for assessing the robustness of the reconstruction, then that needs to be reflected in H2022.)
2. Here and in H2022 the authors discuss the independence of the temperature and precipitation reconstructions. This is an important issue as the second aim of this study is “What are the continental, latitudinal, and regional patterns of Holocene precipitation change and how do these changes co-vary with temperature trends?” (L111-112). In H2022 the authors use a method to reduce the influence of covariance between temperature and precipitation (tailoring). They conclude that “The tailoring successfully reduced the co-variation of temperature and precipitation in the modern dataset as indicated by the distribution of the correlation coefficient in Fig. 8. Nevertheless, the obtained reconstructions are largely consistent between WA-PLS and WA-PLS-tailored: a correlation of r >= 0.9 is found for 59.2% of all records for TJuly, 60.7% for Tann and 56.5% for Pann.” (L292-296 H2022). Notwithstanding whether the r >= 0.9 is a good criterion or not, my conclusion is that the tailored reconstructions are superior because they suffer less from co-variation and that about 40 % of the time series are markedly different from the non-tailored ones. So if independence of the temperature and precipitation reconstructions is a concern, I fail to understand why the authors ignore their own solution to this problem and not simply use the tailored reconstructions. Similarly, how independent are the annual and July temperature estimates and can one really interpret the difference between them?

Furthermore, the authors mention the reconstruction uncertainty in the method section and refer to LegacyAge 1.0 (https://doi.org/10.5194/essd-2021-212) for the chronology (and its uncertainties). It remains nevertheless unclear how these uncertainties are treated or if they are considered at all in the analyses presented here. This is important as the inferred changes in temperature and precipitation are small relative to the stated error and because LegacyAge 1.0 indicates that age uncertainties of the time series have a median uncertainty of about 500 years (but reach to over 1,000 years). So, are the regional reconstructions really different from each other?

I have two more questions about the reconstruction error

1. L156-160: “As it has already been shown in previous comparisons, WA-PLS can have higher RMSEPs than MAT but these do not necessarily reflect a less reliable reconstruction but methodological differences (Cao et al., 2014).” This is an interesting statement and it would be good to repeat some of the reasoning presented in Cao et al here. More importantly, if the estimate of the error is method dependent, how useful then is the error? Would one not get a better, more meaningful, estimate of the reconstruction uncertainty if the difference between various methods is accounted for (see e.g. (Kaufman, McKay, Routson, Erb, Dätwyler, et al. 2020)).
2. L160-161: “Besides, the reconstruction errors are likely much smaller when only the trends and the relative changes are assessed, as in this study.” This may be true to some extent, but it would be good if the authors provided some explanation for this statement.

Finally, the section on methodology to calculate the time series of temperature and precipitation is descriptive, but I have some additional questions and, crucially, miss some explanation of the rationale. Why were the time series 500-year smoothed and resampled at 100 year resolution and spatially averaged (at 2x2 deg) prior to analysis and why is that the best method if one aims to investigate spatial variability? How was the value of 500 years chosen? How close is the 100 year to the actual resolution of the time series? How were gaps in the time series treated (looking at the data at https://doi.pangaea.de/10.1594/PANGAEA.930500 it appears that the sampling was not done continuously in depth and the time series therefore contain gaps). How spatially representative are the averaged time series for the different subsets? I.e. how many time series (or 2x2 grid cells) used for each regional reconstruction? And how does data availability affect the (un)certainty of the reconstructions and the differences among them?

Minor comments

L79-80: “despite the existence of many Holocene pollen records” this seems an odd comment given that some of the syntheses referred to in this sentence post-date the “previous reconstructions”. Moreover, some of the authors of this study were also involved in temperature 12k project, raising the question why they did not include these records at that time.

L154: rather than referring to a map that shows the error for the entire LegacyClimate dataset, it would be helpful to present a map of the reconstruction error for the subset of ~1600-900 time series analysed here.

L176: please define the boundary between Asia and Europe.

L247: “fewer” instead of “less”

L250: “values outside the range” please show (or mention) the entire range and what proportion of the data points falls within the restricted range. This sentence raises suspicion about the reconstructions that can easily be avoided. 

The maps, especially those in Fig. 3, are really small and difficult to read. The individual panels can be made bigger by removing white space and redundant labelling without the need to increase the overall figure size.

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.

Kaufman, Darrell, Nicholas McKay, Cody Routson, Michael Erb, Basil Davis, Oliver Heiri, Samuel Jaccard, et al. 2020. “A Global Database of Holocene Paleotemperature Records.” Scientific Data 7 (1): 115.

Marsicek, Jeremiah, Bryan N. Shuman, Patrick J. Bartlein, Sarah L. Shafer, and Simon Brewer. 2018. “Reconciling Divergent Trends and Millennial Variations in Holocene Temperatures.” Nature 554 (7690): 92–96.