the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 18 Jun 2025
A daily gridded high-resolution meteorological data set for historical impact studies in Switzerland since 1763
Abstract. High-resolution gridded daily data is needed to study historical climate and weather impacts. Current daily gridded data sets in Switzerland extend to 1961 or 1971 for variables such as minimum and maximum temperature and sunshine duration. However, studying historical weather and climate events, such as the year-without-a-summer requires much longer time periods. For Switzerland, high-resolution gridded reconstructions of daily mean temperature and daily precipitation sums have recently been developed based on a large amount of early instrumental data for a period from 1763 to 1960. Here, we present an extension of these daily reconstructions to six more variables, namely, relative sunshine duration, relative humidity, minimum and maximum temperature at 2 m, and u- and v-wind at 10 m with a 1x1 km resolution. These additional reconstructions are based on the same method as the previous reconstructions by combining the analogue resampling method and data assimilation. Cross-validation results using a network representative of early 19th-century observations show a mean squared error skill score ranging from 0.70 to 0.80 for wind speed, depending on the season. For maximum and minimum temperature, values average between 0.48 to 0.82, depending on the seasons. These results indicate reasonable skill of the reconstructions and show that the wind and temperature fields outperform climatology despite the data scarcity in the historical period. However, for relative humidity and relative sunshine duration, the values of the mean squared skill score are significantly lower, ranging between -0.31 to 0.48. Furthermore, we explored the potential of the extended reconstructions by evaluating historical and contemporary wildfire events in Switzerland using the widely used Canadian Forest Fire Weather Index (FWI). The two historical fires were associated with a notably high fire danger in the reconstruction. For the contemporary winter fire, the reconstruction agrees well with the index calculated from the COSMO-1 weather forecast model, though neither indicates exceptionally high fire danger.
Overall, this is the first data set that enables impact studies of weather and climate in Switzerland, reaching as far back as 1763.
- Preprint
(4740 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on essd-2025-249', Anonymous Referee #1, 08 Jul 2025
The authors present a new reconstruction of daily sunshine duration, relative humidity, minimum and maximum temperature, and u- and v-wind with a 1x1 km resolution for Switzerland. The reconstruction covers the whole period since 1763, proposing thus a wide range of conditions that could be used as a baseline to analyse recent events or to estimate the impact of past changes on agriculture and fire development for instance. The methodology is similar to the one applied in a previous study. The skill of the reconstruction is evaluated in details to identify the interest of the approach but also the limitations. This is a very interesting product and it is well presented here. However, I would be happy to have a deeper justification of some of the choices and a longer discussion of the implications of those choices before the publication in the journal.
General comment.
My only general concern is about the consistency between the different variables. The authors mention (lines 469-471) that the ‘an advantage of our reconstruction is that the variables are largely physically consistent with each other since they stem from the same or similar analogue days’. However, this consistency is ensured between some variables like relative humidy and wind speed but not for others. The daily mean temperature and daily precipitation comes from a previous study with a slightly different methodology. Wind and relative humidity come from a different pools of analogues compared to minimum and maximum temperature and relative sunshine duration. The possible issues are discussed for daily minimum and daily mean temperatures as inconsistencies can be obvious for those two variables but a wider discussion is needed. It is also not clear why different fields comes from different pools of analogs (for instance I guess COSMO is providing all the required variables ?) or why an updated reconstruction of daily mean temperature and daily precipitation was not produced here to be have a more consistent product instead of using the existing reconstruction.
Specific comments
Line 16. I would mention that the two historical fires occurred in summer (to make clearer the difference with the winter contemporary wildfire).
Line 36. It is mentioned ‘mean and minimum relative humidity’. Is this correct ?
Line 44. You should justify why you do not apply the data assimilation for those variables.
Line 89. I do not follow to what corresponds the 361 values and later the 190 and 38 values. Which variables are selected and from which dataset ?
Lines 121-122. Is it possible to add a reference where this realistic representation is shown ?
Lines 126-128. Could you comment on the lower quality of the reconstruction using the data from COSMO-1 with ERA-5? I would have expected better results as a larger pool of analogs is available from the longer series. Additionally, it is mentioned ‘using ERA-5 variables as predictors’, I guess it is as boundary conditions for the COSMO-1 model.
Line 135. What is the impact of this choice ? Does it introduce inconsistencies compared to the previous reconstructions ?
Section 3.2 Could you explain how the number of analogs is selected (see for example lines 206-207)? You should also explain why two analogue pools were chosen and the potential consequences (see the general point above).
Line 248. Fig 1b and 1c correspond to year 1864 if I am right. I would write it explicitly for comparison with the two other dates (1767 and 1819).
Figure 2. It took me some time to see that panels a-d were for wind speed and panel e for wind direction. Is it possible to add the information directly on the plot ?
Line 284-285. This sentence is hard to follow. Does it make a reference to the previous study? In that case, I would expand the discussion to be more explicit.
Figure 3. Are the I-l panels showing percentile (and in that case why only numbers between 0 and 9). Is it deciles instead ? (same for Figure 6).
Figure 5. As for Figure 2, would it be possible to put on the figure the variables shown ?
Figure 8. Typo in the caption ‘fg’ instead of ‘g)’.
Line 430. Is a date missing between the parentheses ?
Lines 448-450. The same sentence is repeated twice.
Citation: https://doi.org/10.5194/essd-2025-249-RC1 -
RC2: 'Comment on essd-2025-249', Anonymous Referee #2, 31 Jul 2025
The manuscript essentially presents an extension of long-term daily reconstruction data over Switzerland by six more variables, against previous work. Overall, I find the manuscript to be written clearly and a valuable add-on and dataset extension to the related previous studies. I therefore clearly rate it worth of publication in ESSD, upon revisions as suggested below. In particular, some parts are really in need of more detailed explanation, to make it easier for the reader to properly understand the study, and the aspects of quality and weaknesses of the data. While I rate all suggestions together to constitute a major revision, I don’t think they are particularly difficult to implement, however. They mostly intend to aid improved clarity and readability of what basically is considered a decent paper. See the comments below.
Major Comments
#1: The different networks (NW 1, 3, and 5) used in the study are in need of a clearer and more consistent description and/or referencing. For example, an overview table including the NW’s abbreviation, the number of stations, and the period a NW refers to, could be very helpful. In the text, I suggest to stick on using the NW abbreviations. The switching between the numbers and the descriptions makes it unduly (and without good cause) hard to keep in mind which network is which. Possibly also add a Fig similar to Fig. 1b and 1c for the other two NWs. Also, better add to the Fig captions on which NW the results shown are based on (see also the minor comment 1 related to the Fig captions).#2: The description of the ARM appears a bit too minimalistic, while it should at least sufficiently convey the basic idea behind this method. While I do well understand that the description is kept short on purpose, since it is described in more detail in Imfeld et al. (2023), summarizing briefly also here how the analogue days are selected etc. would make it easier to understand how this very important step in the study works. In principle, this could also be explained aided by a related overview Figure.
#3: Please state more clearly, which periods are compared to which in the evaluation. For example, I found the description in Imfeld et al. (2023) quite easier to understand, because it clearly stated that the reference period got compared to the historical period. In a way, the issue with the ‘obscurities’ re the different periods is similar to the one mentioned in major comment #1 above related to the NWs; it sometimes remains unclear to which periods a piece of discussion is referring. Hence please revise adequately and make sure that you use consistent descriptions/names of the periods.
#4: By always showing the results for Tmax and Tmin together, the Figs for these results get very crowded and it becomes hard to keep in mind which panel refers to which parameter. Please consider separating the results of Tmax from the ones of Tmin. If you want to keep them together (to show a direct comparison, which I don’t see strongly needed for the science discussed), consider improving these Figs by adding subtitles that clearly state, where the results for Tmax are shown and where the ones for Tmin. Also, Figs 6 and 8 could benefit from reducing the number of colors in use. For example, consider sticking just to the 3 colors used for the 3 different NWs. Tmax and Tmin could then be destinguished by different color saturations, for example, or by getting the respective boxplots dashed. See also the minor comments 2 to 4 related to the Figs.
#5: Regarding the results of relative sunshine duration and humidity, the authors clearly state that their results are not convincing and the reconstruction performs (at best) only as good as the climatology. I understand that the reconstruction focused on the wind variables and the setup was chosen accordingly, which is fine. However, if the reconstruction of relative humidity and relative sunshine duration are known to be of poor accuracy, how can they still be “important for certain types of studies”? Could these studies not simply use the climatologies instead? What is the added value of the reconstructed fields? Please either strengthen the scientific arguments or reconsider/tone down the way of inclusion of these two “lowest-quality” variables.
#6: Though the authors mention that “early data may be subject to uncertainty”, the uncertainties are not explicitly quantified in the study. Could the authors provide at least a rough quantitative estimation of the associated uncertainties and how they develop over time? Or, at least, I strongly suggest to add a paragraph, where the uncertainties and their possible development over time are (semi-quantitatively) discussed.
Minor comments
1: On the Fig. captions: some of the captions are done a bit ‘lazy’, missing important information (e.g., which NW used) and are sometimes not correct (e.g., “[…] for five different networks […]” in Fig. 8, or saying “[…] shown as black asterisks […]” in Fig. 10, though dots are used). Please correct, and concisely complete the description of what is actually visible (where needed).2: Specifically on the Figs with boxplots: These plots are quite ‘crowded’ and hence hard to grasp. Please think, for example, to somehow increase the horizontal space available for each plot (maybe by flipping the rows and columns). And/or consider to restrict to the most important boxplots really discussed in the text.
3: Specifically on Fig. 1: The maps are somewhat hard to read. Consider, for example, the following changes:
- Rearrange the individual panels. Panel a could be on top and the maps below. And if you stretch panel a slightly, the horizontal space available for the maps increases, enhancing their readability.
- Use colors that can be distinguished more easily. In particular, the Tmean and Tmax/Tmin colors do appear not suitable in the maps because they look very similar.
- The asterisks are quite hard to see. Consider using a symbol that is more clearly distinguishable from the dots. Also, the black color appears too similar to the purple used for pressure.
4: Specifically on Fig. 4: The maps in panels g and h appear really too small, and the wind arrows are hardly visible in fact. Consider moving the maps to a separate bit-larger Fig (well possible to have 4a-f separate).
5: On the linking of dates/periods in the text to Figs: the text at several places refers to certain dates/periods and discusses what can be seen in the Figs around these dates/periods. Consider, as applicable, marking these dates with vertical (e.g., dashed) lines in the Figs and adding, for periods, a shaded gray area in the Figs. This would help with finding the features referred to in the text more quickly.
6: On occasional split of numbers and units in physical quantities: Better use “unbreakable” spaces between numbers and units; ensures that this is always well legible even if line breaks interfere.
7: In line 1: Typo in this line, “… data is needed to …” should read “… data are needed to …” (data are a plurality of individual elements, hence plural form is right; please check, and rectify as needed, throughout the text).
8: In lines 14-16, and at end of Abstract: please better embed the meaning of "... The two historical fires ..." (it is not clear from text context before, how the case “two fires” suddenly pops up here and which these are). And: at the end of an ESSD paper abstract it is good practice to explicitly note the data availability (including DOI) in a closing sentence.
9: In line 47: Typo in this line, “… to compare it two …” should read “… to compare it to …”
10: In line 430: reference to the cited newspaper seems missing (just a blank “()” visible). Maybe you also might meanwhile have a more scientific reference? (but if the news source is properly cited, ok also in this case).
11: Finally, on data availability (DOI in line 487): looks data are orderly available and carefully prepared, and reasonably simple to download via the .zip files; well in line with what the standards should be for an ESSD paper. These few suggestions, nevertheless:
a. fire weather indices (fwi) would be, in my view, more readily aligned for use if the six files were stratified by index, instead of chunked into subperiods of time. That is, one full _1763-2020.zip file per index, and six files in this way (if keeping with the apparent goal to safeguard max. size to no more than ~15 GB per file).
b. the “readme.txt” is basically fine but it could have simply dropped all those subinfo-points that are anyway unused (but well covered via the paper). For example, under “Methodological Information”, all but the first point can be dropped since the unfilled stuff makes it look ‘unfinished’ etc. And: the umlaut characters (e.g., “ö” in Brönnimann) don’t correctly unfold in an EN-based viewer/at least not in mine (perhaps resort to pure ascii, like “oe” and so).Citation: https://doi.org/10.5194/essd-2025-249-RC2
Data sets
A daily gridded high-resolution meteorological data set for Switzerland since 1763 Noemi Imfeld, Stefan Brönnimann https://doi.org/10.48620/87086
Model code and software
swiss-histmetgrids Noemi Imfeld https://github.com/imfeldn/swiss-histmetgrids
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 232 | 25 | 27 | 284 | 20 | 29 |
- HTML: 232
- PDF: 25
- XML: 27
- Total: 284
- BibTeX: 20
- EndNote: 29
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1