Reply on RC2

The paper is interesting in that it (i) gives a synthesis of weather and climate changes in the Czech Republic in the period 1501–2020 based on documentary evidence and instrumental observations, (ii) tries to describe the main causes of climate change in this time using statistical attribution analysis (regression and wavelet techniques), and finally (iii) investigates spatiotemporal relationships with gridded European climate reconstructions. All three of these topics are very important for scientists interested in historical climate reconstructions, and especially in those based on documentary evidence.

The paper is interesting in that it (i) gives a synthesis of weather and climate changes in the Czech Republic in the period 1501-2020 based on documentary evidence and instrumental observations, (ii) tries to describe the main causes of climate change in this time using statistical attribution analysis (regression and wavelet techniques), and finally (iii) investigates spatiotemporal relationships with gridded European climate reconstructions. All three of these topics are very important for scientists interested in historical climate reconstructions, and especially in those based on documentary evidence.
To be published in the journal, however, the paper needs some substantial improvements and corrections, propositions for which are listed below: REPONSE: We thank the reviewer for careful evaluation of our paper and rising critical comments we are trying respond below.
Generally, all four drought indices are well correlated (Table 1), and I therefore suggest limiting their number to two indices. The text describing the results will be more concise and readable. The best choice in my view is to use SPI and SPEI. SPEI is the index best correlated with temperature and precipitation in all seasons, and, moreover, only this index was independently reconstructed for the Czech Republic using phenological data.
RESPONSE: The four drought indices belong to those used most frequently in papers analysing droughts. Each of them shows different aspect of drought both in terms of considered drivers as well as time scale. SPI reflects particularly to the deficit of precipitation compared to normal patterns, SPEI combines effects of precipitation and temperatures including evapotranspiration, Z-index and PDSI reflect particularly soil drought, calculated without memory in monthly step (Z-index) or taking memory of drought into account (PDSI). Because PDSI is the most complex and broadly used index for drought evaluation (for example, PDSI is used in dendroclimatological reconstructions), we would like to preserve both PDSI (including drought memory) and Z-index, expressing drought without such drought memory (similarly as SPI and SPEI). Furthermore, despite correlations calculated between climate variables for the whole series being high in some cases, their partial components may behave very different (for example, the trend correlated with GHGRF in DJF is different for SPEI and for Z-index, including differences in statistical significance -see Fig. 11). SPEI calculated from phenological data we count less representative than SPEI calculated from temperature and precipitation indices.
In the Discussion section a comparison of the obtained results against other similar climate reconstructions of local and regional character available for the central and other parts of Europe should be also presented.
RESPONSE: Accepted. To follow the reviewer comments, we created a new section 5.1, in which the following paragraphs are particularly relevant to addressing this comment (please check in the context of the whole section): "With respect to these facts, mutual comparison of different climate reconstructions is an important tool to highlight strengths and weaknesses of individual reconstructions and outline possible reasons for some peculiarities in their variability. In this study, the comparison was based on the correlation analysis as well as on the direct comparison of smoothed series to highlight common variability on decadal and multidecadal scales (see Figs. 2,8,and 14). The following text summarizes the main features of such comparison that have been explained in detail in the original "reconstruction" papers. Moreover, we are trying to explain possible reasons that may be responsible for the loss of common signals in some periods.  2004) shows positive differences before the mid-18th century and negative afterward. This shift is responsible for a sharp decrease in running correlations. In order to evaluate this inconsistency, differences of these two series with regard to completely independent JJA multiproxy temperature reconstruction for the Alps by Trachsel et al. (2012) were calculated. For better comparison, the series were first transformed to have a mean of zero and a standard deviation of one. While the differences with the series by Dobrovolný et al. (2010) were distributed more or less randomly around zero, the differences with the Luterbacher et al. (2004) series showed the same patterns as described above: positive differences before the 1750s (i.e., higher temperatures by Trachsel et al., 2012) and negative differences afterward. This indicates that the problem of lost coherence around the 1750s in Fig. 14a cannot be attributed to Dobrovolný et al. (2010) reconstruction.
As for series derived from phenological data, MAMJ temperatures reconstructed from winter wheat harvest dates were compared with 11 late spring and summer temperature series in central Europe (see Fig. 6 in Možný et al., 2012). Better coherence was found with documentary-based and biophysically-based reconstructions (harvest dates) than those based on tree-rings. A significant drop in correlations appeared particularly in the second half of the 17th century and around the 1750s. This may be partly related to the problem in the data quality of the winter wheat harvest dates. These dates had to be recalculated from the harvest dates of other available cereals in periods when the winter wheat dates were not available. The distinct role may be attributed to the "social bias" in data related to the complicated social and political situation in the country (see discussion related to those periods in Možný et al., 2012, and also Fig. 8a in the current study).
Similarly, AMJJ temperatures reconstructed from grape harvest dates were compared with 17 European temperature reconstructions based on temperature indices derived from documentary data, grape harvest dates, tree-rings, and multiproxies (see Fig. 9 in Možný et al., 2016a). Possible inconsistencies were found in the first half of the 16th century, around 1650, 1750, and 1900. Four periods with potential "social bias" were identified in the last decades of the 16th century and then in the 1640s-1670s, 1750s-1780s, and 1850s-1910s.
The comparison seems to be more problematic in the case of precipitation, characterised by high spatiotemporal variability. For example, less spatially homogeneous Czech JJA precipitation totals were plotted against six similar European precipitation reconstructions (see Fig. 9 in Dobrovolný et al., 2015). Periods of quite similar precipitation fluctuations were revealed particularly in the first half of the 16th century, in the 1630s and 1710s (dry decades), and approximately in the 1590s, 1690s, 1730s and 1810s (wet decades).
Documentary-based reconstructions of drought indices in the Czech Lands were correlated against six different European drought series (see Fig. 6 in Brázdil et al., 2016). The overall patterns were the same as in Figure 14c in this study. While there was a good agreement especially in the first half of the 16th and the 17th centuries, a drop in common variance appeared in the second half of the 16th century, in the 1650s-1750s and after the 1950s.
Differences between reconstructions and loss of coherence between them may also result from a natural climate variability. This applies especially for those covering a slightly different spatial domain or those reconstructing climate variables characterized by high spatial variability. As discussed in more detail in Možný et al. (2016a), some periods (e.g., Maunder minimum in 1675-1715-Frenzel et al., 1994 can be characterized with a higher frequency of meteorological extremes of the regional extent. Their more frequent occurrence in some regions may be conditioned dynamically (i.e. by different circulation patterns -see e.g. Wanner et al., 1995) and thus may be responsible for higher spatial climate variability and subsequently for lower correlations in comparison to related series on a central European scale." The attribution analysis must be done separately -for pre-instrumental (reconstructed series) and instrumental periods at least. For example, for the periods 1501-1800(50) and 1801(51)-2020. It is obvious that until about the mid-19th century climate changes were caused mainly by naturals factors (volcanic and solar forcing). Anthropogenic factors (mainly greenhouse gases) are important only for the industrial period and therefore should be limited to this period.
RESPONSE: Please note that such application of regression analysis to shorter data segments was already carried out in a prior paper, Mikšovský et al. (2019), where subperiods 1501-1850 and 1851-2006 were considered separately in addition to the full length of the series. We did not deem it useful to repeat these partial tests in the current paper, as the conclusion would likely be near-identical to those in Mikšovský et al. (2019). Furthermore, using shorter data segments (and thus fewer data points) increases the uncertainty of the regression coefficients (i.e., the size of the respective confidence intervals), making the attribution analysis less sensitive. This even applies to the analysis of long-term trends such as those related to greenhouse gases forcing -even when the predictor only exhibits noteworthy variability in a part of the analysis period, using the entire length of available data allows the regression mapping to better quantify the link to target variable(s), and to more reliably distinguish between different sources of trend-like changes.
Minor weaknesses: 5, line 39 -please explain the reason for such a big change in correlation coefficients (from about Ë0.7 to 0.0-0.2, Fig. 2a) around 1900 between all studied series. What happened at the end of the 19th century and the beginning of the 20th century that the correlation between temperature and other variables was lost? Is this a problem of loss of homogeneity of temperature or precipitations series? RESPONSE: Accepted. Response to this comments is included in the following paragraph in the newly created section 5.1 (please check in the context of the whole section): "An interesting aspect of lost common signal manifested by a decrease in running correlations below the 0.05 significance level can also appear in the "instrumental part" of the reconstructed series as documented in Fig. 2a. Running correlations of annual temperatures with other five climate variables are highly significant from the 16th century up to the early 19th century. These negative correlations are physically consistent as they show that higher temperatures usually correspond to low precipitation and vice versa. Approximately from the mid-19th to the mid-20th centuries correlations among all compared series are not significant. Despite the fact, that annual means express some mixture of different seasonal patterns, this gradual loss of common signal may be interpreted as follows. The fact, that before the 19th century the series are reconstructed from dependent (and thus less variable) temperature and precipitation indices, can be reflected in significant correlations. The instrumental parts of series (target data) are mutually less dependent and more variable than indices. The same patterns as in annual values (Fig. 2a) are well expressed also in SON series and partly in MAM and JJA series, while they do not occur in DJF series (non-significant correlations over the whole period) (not shown). The stronger common signal (significant negative correlation) occurring during the last decades can be attributed to a clearly expressed opposite tendency of rising temperatures and decreasing drought indices. The same pattern does not change even when correlating the detrended series or when changing the length of the window, for which running correlations were calculated." 8a -a similar problem to that mentioned in point 1: please explain the reasons for the loss of correlations between the two reconstructed temperature series only just after the mid-17 th century and mid-18 th century for two-three decades.
RESPONSE: Accepted. We tried to explain this problem and general loss of coherence among different reconstructions in the newly created section 5.1, where we reported also weaknesses in both "phenologically-based" reconstructions (please check it in the context of the whole new section). Particularly the following paragraphs concern of the above problem: "As for series derived from phenological data, MAMJ temperatures reconstructed from winter wheat harvest dates were compared with 11 late spring and summer temperature series in central Europe (see Fig. 6 in Možný et al., 2012). Better coherence was found with documentary-based and biophysically-based reconstructions (harvest dates) than those based on tree-rings. A significant drop in correlations appeared particularly in the second half of the 17th century and around the 1750s. This may be partly related to the problem in the data quality of the winter wheat harvest dates. These dates had to be recalculated from the harvest dates of other available cereals in periods when the winter wheat dates were not available. The distinct role may be attributed to the "social bias" in data related to the complicated social and political situation in the country (see discussion related to those periods in Možný et al., 2012, and also Fig. 8a in the current study).
Similarly, AMJJ temperatures reconstructed from grape harvest dates were compared with 17 European temperature reconstructions based on temperature indices derived from documentary data, grape harvest dates, tree-rings, and multiproxies (see Fig. 9 in Možný et al., 2016a). Possible inconsistencies were found in the first half of the 16th century, around 1650, 1750, and 1900. Four periods with potential "social bias" were identified in the last decades of the 16th century and then in the 1640s-1670s, 1750s-1780s, and 1850s-1910s." Could you also inform the reader which of the temperature reconstructions presented in Fig. 8a is better and more reliable (based on temperature indices or on wheat harvest dates). Differences in absolute values of temperature are sometimes very large. This is very well seen particularly in the aforementioned times when the correlation is lost.
RESPONSE: We understand the reviewer comment, but the answer will very much depend on the chosen criteria. Each of these reconstructions is based on different type of data with some advantages and disadvantages. For example, if we will take into account the explained variance in the calibration/verification period, both reconstructions are comparable. The wheat harvest day (WHD) reconstruction explains 0.70 of the MAMJ temperatures and it is 0.69 in case of the central European temperature (CEUT) reconstruction (mean value for the corresponding months). From direct comparison in Figure 8a (bottom) it follows that the WHD captures the low frequency signal better than the CEUT. However, this is with a high probability related to the quality of data used for the WHD chronology compilation. The periods that show the largest differences in the two compared reconstructions in Fig. 8a well correspond to a significant drop in correlations. As can be verified from the Figure 6 of Možný et al. (2012) these suspicious periods, especially the second half of the 17th century and the period centred in 1750s, can be well identified when one compares the WHD with several other proxy reconstructions in central European context. This indicates that the problem probably lies in the quality of the data used to compile the WHD chronology that is changing over time. This explanation may be supported by the fact that also the variability of the WHD-based temperatures is clearly changing over time (see Figure 7a, top).
8 -the same scale should be used in Figures 8a and 8b  RESPONSE: Trying to follows this comment, we asked for corresponding data the first author of the paper, Prof. Juerg Luterbacher (WMO, Geneva), but he replied that he no longer has any such data. On his recommendation we contacted also one of Polish coauthors, Prof. Rajmund Przybylak (UMK, Torun), but with the same negative result.
I suggest reducing the number of figures and presenting more possible explanations for peculiarities in the course of climate change in the Czech Republic in the study period.
RESPONSE: Accepted. To reduce the number of figures in the main manuscript, the wavelet coherence plots (originally in Fig. 13) have been moved to the Supplement, as Fig. S2. Furthermore, in response to a suggestion by reviewer 1, Fig. 10 has been simplified and the correlation matrix (originally Fig. 10b) moved to the Supplement as Fig.  S1. Concerning of other figures in the manuscript, we consider every of them as important and we would like to preserve them in the manuscript. We extended manuscript in the parts, where it was requested by both referees (see the new section 5.1 and our responses above), and we believe that we have explained basic peculiarities in the course of climate change in the Czech Republic.
I can recommend acceptance of the manuscript for publication in the Climate of the Past only on the condition that the remarks and suggestions listed above are satisfactorily taken into account.