The strength of the paper and the dataset is that the modern pollen surface samples are from lake sediments. This is an edge compared to most other modern pollen datasets from China and other regions. The procedure for using this dataset for constructing a pollen-climate calibration model is described in the paper, and it follows a standard routine with WA-PLS-based quantitative climate reconstructions. The data and the results are generally clearly presented, although the paper is very short with hardly any relevant discussion.

As a data description, the paper can be fine as such, but it would interesting to develop the study in the future. The authors stress the importance of having both the modern pollen samples and fossil pollen samples from the same sedimentary environment, lakes in this case, and I agree with this. Consequently, it would be interesting to test whether the current calibration model works better as compared to the calibration models based on samples collected from varying sedimentary environments, such as topsoils, moss polsters etc.

Another interesting angle for the study would be to apply some novel transfer function approach in the study. In Fig. 4 we can see that the calibration model has a quite serious bias towards too low values at the upper end of the precipitation gradient. The authors briefly comment on this underestimation of high values. Such an edge effect is a typical problem with WA-PLS-based transfer functions. An amendment has been suggested with the use of tolerance-weighted WA-PLS (Liu et al., 2020) and it would be really interesting to see whether the use of tolerance-weighted WA-PLS would improve the edge effect problem in the calibration model and influence the precipitation reconstruction shown in Fig. 6.

Some minor remarks

-page 2 line 36 “reconstruction of climate data” remove “data”

-page 2 line 51 remove “desperately”

-page 3 as this is a dataset paper, it would be better to include key data from all 90 sites, such as location, altitude, climate etc.

-page 4 Fig. 1 add an index map

-page 5 lines 103-104. This in unclear. What does “reanalysis datasets” mean? And how were the altitudinal differences handled? Was the windward or leeward side location of the lakes in relation to the mountains considered?

 -page 6 add citations to the selection of RDA as the linear method and add citations to the use of VIF to check for collinearity.

-page 7 line 137 “a novel method” delete “novel”. What was novel in 2011 is not novel any more.

-page 9 It would have been better not to remove this one outlier site from the dataset. While outliers are sometimes deleted this way, it is a questionable thing to do. Firstly, it is an easy trick to improve performance statistics by removing the “dodgy” samples. Secondly, the performance statistics of the current dataset (e.g. R2 values) cannot be compared directly with other datasets in which no samples have been removed.

References

Liu, M. et al. 2020. An improved statistical approach for reconstructing past climates from biotic assemblages. Proc.R.Soc.A476:20200346.https://doi.org/10.1098/rspa.2020.0346