Comments for: “Effects of synoptic atmospheric variability on sub-daily precipitation δ18O-air temperature functions in reconstructions of pre-instrumental δ18O chronicles across Europe”

OVERALL COMMENTS

This is an interesting idea and certainly a neat approach, but I am not convinced that the results are reliable in a truly useful way due to the limitations applying a calibration from a single site across continental scales. Essentially, my comments revolve around the reliability of the method and the utility compared to other possible (including previously published) methods. 

ATMOSPHERIC CIRCULATION PATTERNS AND PHYSICAL EFFECTS

More explanation is needed on how atmospheric climate patterns relate to physical effects. Specifically, I do no understand how you view physical effects acting differently during different kinds of atmospheric circulation patterns. I imagine that you are not arguing that, for example, the isotope fractionation occurring during condensation at a particular temperature changes with different circulation patterns (by definition it cannot). But then, I am not clear on what it means for an atmospheric circulation pattern to affect a local physical effect. Can you please clarify how you envision the relationship between synoptic weather patterns and the actual physical processes causing isotope fractionation? Perhaps I am misunderstanding your meaning due to using terminology differently, but if I have this question, others will too.

SPATIAL AND TEMPORAL TIMESCALES OF MODELING

Why is it valid to extrapolate your calibration at LIST to a continental scale? It looks to me like your modeling approach basically does best at sites near the LIST field site, which aligns well with prior research showing that water isotope values exhibit spatial coherence (e.g., Bowen and Revenaugh, 2003). So wouldn’t you expect this pattern to emerge?

In addition, more explanation is needed as to why your model, which I understand to ultimately be based on monthly aggregate data, eventually including monthly GNIP data, can be used to infer changes on weekly timescales. Doesn’t the data show that the modeling does best on seasonal timescales, often under/overestimating the magnitude of variability on shorter timescales?

THE BROADER CONTEXT AND OVERALL UTILITY OF THIS APPROACH

As the authors note, there are prior isoscapes for this region of the world, including approaches that make use of variables other than (or in addition to) temperature – such as precipitation amount. It would be useful to give broader reference to this prior work as well as to incorporate/argue against these prior approaches. Some thoughts:

Why are other environmental variables besides local temperature not used in the modeling? What happens to model performance if they are included?

For the isoscape creation, how different are the results from using the calibration at LIST from using a continental-scale calibration? As in, instead of trying to develop a d18O-T curve for LIST and then applying that curve to the continent through time, what if you made a d18O-T curve for the continent and then applied it through time (essentially, this approach would be to update previously published d18O(weather parameters, lat, lon, elev) functions and apply them through time). Are the results materially different? Why would we prefer one approach over the other? Or would it be better to use LIST and GNIP data together to create d18O functions by aggregate month, region, etc?

How important is it to include all of the Climate Pattern information for the modeling? If including the CP information reduces RSME by 0.2 (2.8 to 2.6) and increases r2 by 0.07 (0.37 to 0.44), is this truly meaningful? Specifically, what investigations does this approach allow that were previously untenable without the CP information.

Finally, what are the uses for time-transgressive isoscapes beyond the “climate normal” versions that already exist? Accounting for the errors in developing isoscapes in each way, how different would the estimates be? What are the benefits of being able to work with estimates of precipitation isotope values from a particular year, rather than an average of many years?

LINE-BY-LINE COMMENTS

10

Qualify this statement a little further. Worldwide predictive maps of d2H and d18O exist and daily, monthly, and yearly data sets exist in many locations.

11

Clarify what you mean by “long term”

12-15

Clarify this further. Is this not how the community as a whole envisions what drives O and H isotope signatures? If not, what are the alternatives?

14

Suggest not using the abbreviation “CP” as it is not common

29

Define the sense in which stable isotopes of oxygen and hydrogen are “near-conservative”

77

I do not quite understand the linkage being made here between atmospheric circulation patterns/climate patterns and physical effects on isotope values in precipitation. From what is written, I understand that you hypothesize that atmospheric circulation affects how the temperature of condensation is expressed in the isotope composition of precipitation. But how exactly does this work? By definition, if you are considering a strict temperature effect (i.e., how the temperature of condensation induces isotope fractionation between vapor and liquid water), the magnitude of the “temperature effect” must be driven by the starting temperature and overall temperature variability of the air mass induced by atmospheric circulation patterns. But the effect of temperature on isotope fractionation exists regardless of where an air mass originates, right? It is simply that different atmospheric circulation patterns may be associated with more/less temperature variability, which may mean that processes other than temperature will be responsible for the variability observed in precipitation isotope values.

100

Were internal standards used to normalize the data? What were their values?

104

Why was a sin wave chosen? Were other fits considered?

106

Check equation and units. Verify is consistent with results presented in tables

Kirchner (2016) is not listed in the references

110

This section would benefit from a figure with a panel showing each of the climate patterns

120-122

Does use of alternative classification schemes affect your results and interpretation?

130

Why was temperature chosen as the sole input variable? Were other variables considered? What were their relationships with d18O?

135

Were the linear regressions modeled using event-scale data or using monthly data? Line 132 says monthly, but line 136 says event-scale

140

If ERA5 interannual average monthly T data is going to be used in the ultimate evaluation, should it not also be used in the initial assessment at LIST for determining coefficients of Eqn. 2? 

141

Suggest: “To include as many records as possible, the geospatial model…”

To emphasize that the included records do have a full year, but are still shorter than the LIST record

145

How was the “climate pattern” determined for each GNIP station? Is it fair to use the Hess and Brezowsky (1952) categorization for sites outside of Germany, where it was originally designed for?

The constants in Eqn. 2 are specifically for the LIST site. Why are these applicable to other sites 100s km away? Would they not have their own set of constants? In general, it would be useful to have more explanation about why you expect a calibration developed at one particular site to be broadly transferrable across continental scales.

154

Where is the DEM grid from?

173

Citation needed for GMWL definition

189-190

Define “considerable”

224

With such a considerable difference between HCE and the other climate patterns, is fitting a curve to HCE useful? What is the physical meaning of phi = -98 months?

I am not clear on how the sinusoids fitted to the data fit into the larger scope of the work. Please expand on how these were used to investigate the relationships between large-scale atmospheric circulation patterns and local-scale meteorological variables

234-235

Is this gross match between the amplitudes of d-excess and d18O meaningful?

263

What is the correlation on a seasonal scale?

265-267

So how different are the two approaches? Is one inherently more useful than the other?

If the input data for Table 3 was at a monthly scale, why is it fair to use the model at a weekly scale?

294

I could argue that Figure 7 mostly shows that your model does best nearest to the LIST site because that is a foundation of the model. Do you agree? Why?

I might further argue that your model could naturally be expected to perform more poorly closer to the coast where the temperature effect cannot emerge as important because there has not been substantial rainout yet. Other processes would be expected to be important closer to the coast. Do you agree? Why?

327-328

How significant is the slope of 7.54 in terms of identifying re-evaporation?

342-344

The act of evaporation undoubtedly induces isotope fractionation and helps set “initial” isotope values of vapor. However, it is simplistic to view d-excess as a static value – consider, for example, the modeling exercises of Xia and Winnick (2021) and Xia (2023). To what degree can your data set break apart the oceanic vs. continental influences controlling d-excess values?

362-363

This is confusing. So is the T-d18O relationship you infer dominantly coming from condensation reactions or is it incorporating a broader swath of processes with variable influence from temperature?

364-365

I find this pretty challenging. On the one hand, you write here that the d18O-T relationships are dependent on CPs, but in the next sections (paragraphs starting lines 400 and 415), you note that the d18O-T relationships do not appear to be strongly dependent on CPs and in fact can be challenging to usefully apply outside of the region surrounding LIST. How should we reconcile these aspects of the data?

375

Citations are needed here. Where has this assumption been made recently?

380-381

So if changes in air temperature cannot be assumed to accurately predict d18O as the d18O-T relationship changes through time, what implications does that have for your reconstructed isoscapes? Or do you just see this as a challenge under substantially different planetary boundary conditions? Explain further

385

How well, what did others find?

391

Again, if the initial data focus was on monthly inputs, how does this translate into discerning sub-monthly changes in the past? Especially as you note that the model has trouble accurately capturing better than seasonal-scale variability (lines 290-305; Fig. 8)

404-405

So could this instead be taken to mean that climate patterns do not have a strong control on the physical processes underpinning isotope fractionation – that they occur independent of wherever the vapor is coming from?

Perhaps because the CPs tend have uneven seasonal distribution, by breaking apart different d18O-T relationships by CP you are essentially breaking out d18O-T relationships for different seasons, which is what leads to a slightly better performance.

410

If convection strength is an important parameter, why is it not included here? The ERA5 data contain this type of information

419-420

This seems like a significant challenge to the use of the model presented here

424-425

These sentences appear to directly contradict each other

FIGURES

Figure 1

Identify the black lines in panels (a) and (b)

In all figures with maps, clearly identify the LIST site

TABLES

Table 4

Are these coefficients and constant the same for all CPs? I thought that, as in Eqn. 2, the coefficients and constants for Eqn. 3 implied they would also be different for different CPs

The manuscript entitled “Effects of synoptic atmospheric variability on sub-daily precipitation δ¹⁸O-air temperature functions in reconstruction of pre-instrumental δ¹⁸O chronicles across Europe” by Türk et al. assesses the relationship between atmospheric circulation patterns (CPs), temperature and the isotope composition of precipitation from daily to interannual timescale. They use a six sub-daily precipitation isotope record recorded in Luxembourg to train and validate the model. Then, the model is applied to monthly precipitation isotope data of the GNIP network to regionalize the seasonal variability of precipitations isotope composition for the last 150 years.

The principal idea behind the study and the six-year high resolution isotope dataset of precipitation are original and provide a number of interesting features that will be of high interest for the hydrology and paleoclimate communities. However, the manuscript has several issues that need to be solved before publication.

One of the major issues is that the authors switch a lot between different timescales, which is confusing for the reader. For example, most of the introduction deals with variations on monthly timescale, but then the record is on sub-daily timescale. Further, Section 3.1 presents the precipitation and meteorological data on monthly scale, but then, in Section 3.2, an event-based model is applied. The authors should restructure the manuscript, including presentation of isotope and model results on different timescales to better guide the reader.  They may also decide to stick to only one timescale, if only this is relevant regarding the objective of the manuscript.

The authors focus on the relationship between temperature and the isotope composition of precipitation, which is definitely an important factor during precipitation and Rayleigh rainout. However, other processes affecting the atmospheric water vapor from which the precipitation is formed, such as changes in climate conditions in the moisture source regions and contribution of continental evapotranspiration are only shortly mentioned, while their effect on the isotope composition of precipitation remains undescribed. It is not until the discussion section that the reader learns about post-precipitation formation processes, such as rainfall re-evaporation, that can modify the isotope composition of precipitation. I suggest restructuring the introduction section, providing an overview of the processes that can affect the isotope composition of precipitation before, during and after precipitation formation and on which timescale they are relevant.

The authors observe variations in the isotope composition of precipitation with atmospheric circulation patterns. However, the processes behind these isotope variations as well as their relevance on different timescales are not discussed. There is a need for a climate characterization of the different CPs. Also, the authors focus mainly on temperature, but other factors such as rainfall amount and RH during the precipitation event may provide information on local processes such as rain re-evaporation. Also, changes in moisture sources should be discussed in relation to CPs. There is a debate on isotope differences between precipitation derived from Atlantic and Mediterranean air masses, the latter being characterize by a higher δ¹⁸O and d-excess values. Do the authors observe similar isotope differences between air mass sources? Which sources can be attributed to different CPs?

Line-by-Line Comments:

Line 29: In which sense stable isotopes of water are “near-conservative” if isotope fractionation occurs during phase transitions?

Line 47-48: What is the timescale of hydrological processes that is interesting for the community/in this study? Daily/Monthly/Seasonal/Yearly?

Line 55: Why ~20 years if GNIP data exists for 50 years? Give a range?

Line 65: Why are isotope-enabled climate models difficult to constrain? Is it due to the difference in timescale between observations and models or because processes driving isotope variations in precipitation are not well understood? Not clear.

Line 77ff: Your study is based on daily to sub-daily data. Which processes are relevant at this timescale?

Line 77ff: It is not clear how atmospheric circulation patterns influence δ¹⁸O_P. Changes in moisture source, condensation conditions, post-formation processes? Specify this.

Line 81: Why you choose a subjective classification scheme and not an objective criterion.

Line 89: What do you refer to with “precipitation data”? Is it isotope data, samples or meteorological data? Temperature, relative humidity, precipitation amount? Did you use other parameters?

Line 100-104: Can you give more details on the analyses? How many injections per sample? Did you account for the memory effect? What is the frequency of standard analysis?

Line 104: the secondary d-excess parameter is not introduced. Add the formula and explain how it complements the δ¹⁸O and δ²H data.

Line 105: Reference Kirchner (2016) missing in the reference list. Why the amplitude of monthly amount-weighted δ¹⁸O_P is of interest?

Section 2.2: The classification is obtained from a database or have you done it on your own? In the former case, cite the database. In the latter case, specify which criteria were used. Give the period and temporal resolution of this classification. Did you distinguish only the three major patterns or also the sub-types? Meteorological characterization of the CPs would help to better understand the differences. Consider adding a map.

Section 2.3: Why “reanalysis data” in the section title? There is no reanalysis data described in this section.

Line 139: Which GNIP stations did you include and how many? Was there a criterion to include or exclude stations?

Line 168: Give minimum and maximum values instead of the range of δ¹⁸O and d-excess as it might be unevenly distributed around the amount-weighted average value.

Line 170: Not clear which one is high in which season as δ¹⁸O and d-excess show inverse patterns.

Line 172-174: How the LMWL was determined? Based on sub-daily, daily or monthly data or interannual monthly data?

Line 175-181: Is there seasonal variability in the precipitation amount?

Line 189: To which CP refer HCE and LCE? Zonal/Meridional/Mixed? They haven’t been introduced in Section 2.2, aren’t they?

Line 205: Why these values are relevant? Do they reflect the annual average isotope composition of precipitation? Why they deviate from this value?

Line 208: Does the seasonality of CPs influence the sine wave curve calculation?

Line 210-211: Is there a table or figure showing these data for all CPs?

Line 216-220: Is this also reflected in the amount-weighted average for each CP?

Line 224: Does calculating a sine wave curve make sense for HCE of there is little data in winter (or summer?)?

Line 234: Is this expected?

Section 3.2: Here, you evaluated data on weekly scale, while before you presented data on monthly scale, but it is an event-based model… Please restructure to guide the author with the different timescales!

Line 280: Explain this normalization. Is it commonly used? How to interpret this normalized value?

Line 327: This is the first time you mention these post-precipitation formation processes. They should already be introduced in the introduction.

Line 330-331: Do you observe lower slopes in the dry season? Show these results in the Result Section. Why would you expect lower slopes during the dry season, i.e. why rain re-evaporation should be pronounced in the dry season? Is it a temperature or an amount effect?

Line 333-336: Does this exclude the rain re-evaporation process being the key driver of the precipitation isotope composition or do both, moisture source and re-evaporation overlap?

Line 337: What do you mean with “memory effect” here?

Line 346-349: I found the observed d-excess not exceptionally high. It is rather close to the global average. So, there is no contribution of Mediterranean air masses or moisture recycling or are there certain CPs that show higher values?

Line 349-351: Do you observe seasonal variations in d-excess that could be linked to seasonality in continental ET?

Github: It would be great if you could provide a metadata sheet that explains shortly the different files and R scripts.

The paper focuses on the reconstruction of long-term oxygen-18 data in precipitation using sub-daily isotope data for periods of less than 10 years, employing a multiple linear model.

Novelty and scientific significance

The novelty and scientific significance are not well described. Particularly, the application of RSME, which, although highlighted as a novel approach, has been widely used for similar tasks in previous studies or has already been tested (e.g. https://onlinelibrary.wiley.com/doi/abs/10.1002/hyp.14254, https://www.sciencedirect.com/science/article/pii/S016980952300090X). 

Notably, the authors utilize a sub-daily precipitation approach, which is uncommon in the monitoring of isotopes in atmospheric precipitation, where event-based and composite monthly samples are typically collected. More discussion should be added on the advantages of this method compared to event-based and composite monthly samples. Why are these sub-daily data more beneficial for the construction of isotopes and climate-related data since 1881? How can these data be better compared with ice-core, tree, lake sediment, and other types of isotope data used in climate reconstruction? What is the novelty and scientific importance of the method? The linear method is not the best approach to examine the non-linear relationships that can be highly important in relating isotopes and atmospheric oscillations, which mainly have sinusoidal modes. This method has been compared to the AI ML model of Piso.AI, which is based on determining the principal factors controlling the isotopic composition of precipitation and on the prediction function, which is based on non-linear relationships between isotopes and main determinants. This model also does not account for the outliers derived mainly in the winter months.

Scientific Quality 

The scientific quality should be improved, particularly by considering related works and including appropriate references. For example, the temperature effect is not a stationary effect even in continental stations. This can be a reason for the poor prediction of isotope values in winter precipitation. Here, more climate and possibly orographic parameters should be included in the regression. Non-stationarity of isotope values in winter precipitation can be due to a shift towards the precipitation amount effect, and this should be checked and discussed in relation to similar studies. Another point is whether the temperature effect, as the correlation between isotope and air temperature, is a constant function over time. Maybe this effect can be stronger or weaker depending on larger-scale oscillations such as the Multidecadal Atlantic Oscillation or the shorter-term North Atlantic Oscillation. Additionally, explanations should be added on how to relate isotope values in sub-daily precipitation to daily or monthly climate parameters. The paper is missing a strong discussion based on papers that used a similar approach. More references should be added. Even the references on the physical nature of oxygen and hydrogen (as explained in the Introduction) should be revised, and more classical isotope-related studies should be included.

Presentation quality

The introduction doesn't reflect the title of the paper, results, and discussion. A significant part of the introduction is focused on isotopes in streams, but this is not well documented in the results and discussion. The introduction should cover the state of the art related to isotopes and atmospheric circulation, reconstruction of the climate and isotope values, and more clarification on daily circulations should be added. The figures should be improved, for example, by reducing the abbreviations in the legends.

Dear Authors, Dear Editor,

The manuscript (egusphere-2024-4169) presents and evaluates a unique high resolution multiannual precipitation stable isotope dataset from a European observation site and afterwards construct a hybrid model combining circulation patterns and δ¹⁸O -temperature correlation slopes to get monthly isoscapes of precipitation δ¹⁸O values for the entire European continent. While the presented sub-daily dataset is unique and deserve publication, I have concerns regarding the concept of the derivation of the European precipitation δ¹⁸O isoscape. Illustration material (figures and tables) also need more editing and revision. My concerns are detailed below. I think the study needs substantial revision and the revised manuscript should be restricted to the presentation and assessment of the sub-daily dataset.

I note that I cannot make a detailed linguistic revision since I’m not a native English speaker.

General comments

• The assumption that circulation pattern has some predictive power in modelling precipitation stable isotope composition is interesting and agreeable, however I cannot agree that an empirical δ¹⁸O–T relationship can be transferred to thousands of kilometres apart. I mean that the distance and temperature difference between the moisture source area and the location of precipitation formation can be larger between a summer and a winter month belonging to the same CP than between two CPs in a given month. Please have a look on Fig. 14 in https://doi.org/10.1196/annals.1446.019 The maps show the characteristic SLP pattern of selected Hess-Brezowsky types and the spatial clusters of temperature anomalies (and also unresponsive areas!!!) across Europe. In addition, Fig 2 in https://doi.org/10.1016/j.pce.2009.11.013 also warns that the concept of this study is untenable. The maps in top and bottom left show areas where surface temperature under H&B types is significantly different from the rest of data, that is, where a certain CP type is accompanied by specific temperature conditions. I think that it is of indicative value how the areas of European stations with the weakest relationship between surface air temperature and H&B types trends correspond to the areas where the modelling experiment described in the paper performed poorly (Northern Europe, Iberia, Apennines, SE Europe based on Fig 6 , 7 of the manuscript). Seeing this correspondence, it is alarming that the concept of combing Hess-Brezowsky circulation types and CP-specific δ¹⁸O -T slopes derived from the Belvaux dataset can be valid only for a restricted area in Central Europe.
• I suggest carefully checking the isotopic terminology. Usually a word, such as “data”, or “value” or “variance” can be necessary after the delta symbol (see a few examples among the specific comments). In addition, please check the delta symbol carefully because simply “d” is typed at a couple of times (e.g. lines, 73, 75)

Specific comments:

line 25: “assumption-lean” is somehow confusing for me. If you mean that there is no assumption, then I do not agree. (See my detailed comments above)

lines 49-53: This recent paper https://doi.org/10.1007/s13137-023-00224-x is a highly relevant reference for this statement.

line 60: Unclear meaning: “…the influence of climate forcing”

line 65: I re-read the PisoAI paper and think that it is not a pertinent reference to support this statement. Nelson et al. did not constrain any isotope-enabled climate model.

lines 141-142: What does “full year” mean? 12 consecutive months? I mean it could be from Jann to Dec, or Feb to Jan are equivalently, OK? Or how did you treat a year when monthly perc is 0mm in one or two months? How do you treat a station which recorded from Jan to June in 2000 and July to Dec in 2001?

line 153: Which “DEM grid”?

line 155: I’d add “values” or “data” to the end of this sentence.

line 167: The sentence says that the end of the sampling interval is December 2022, while in line 97 it was written “the end of the sampling campaign in January 2023”. Please clarify it.

line 170: Please add a comma before “respectively”.

line 209, 220, 223, 355, 371: Please add “data” or “value” after both δ¹⁸O and d-excess

lines 223-224: Does this high negative phi value acceptable for HCE? I assume that the phase should be close to 12 months (or less assuming a semiannual cyclicity for instance). Moreover, an estimated 98-monzh phase (periodicity?) based on a 62-month dataset is definitely uncertain.

lines 267-268: Similarly “Systematic overestimations towards the lower end of values” i.e. he winter minima was reported by other models (e.g. https://doi.org/10.1073/pnas.2024107118, https://doi.org/10.1007/s13137-023-00224-x ) It might deserve a bit more discussion.

lines 324-326: This is very speculative. Probably the interannual difference could be sufficient to explain the difference. You can try to eliminate the interannual difference calculating the mean weighted δ¹⁸O value for the same period.

lines 327,435 As far as I know “in-cloud evaporation” is rarely an issue. Authors might think about “sub-cloud evaporation”.

lines 333-335: If I understand well, then this is an argument against the significant role of sub-cloud evaporation, contradicting with the previous statements.

line337-338: Unclear meaning. This sentence is confusing.

lines 362-364: Why? What is the supporting evidence?

lines 379-380: I understand this statement, but how is it related to the δ¹⁸O -T relationship? The lower δ¹⁸O values during colder climate is not equivalent with the constancy of the δ¹⁸O -T relationship.

lines 391-392: It is debatable. The model skill was not validated at the sub-monthly scale.

line 394: I suggest omitting “WMO” before “Climate Explorer”. But anyway, the original datasets of historical temperature records should be mentioned here rather than a web application to analysis climate data statistically.

line 437: The statement needs revision. Currently it says that average δ¹⁸O and d-excess values of the Atlantic Ocean is -8.1 and 10.7‰.

Figures and Tables

Figure 1: The first sentence of the caption says that time series are between 2017 and 2022, however in section 2.1 it was written that sampling started in December 2016 and ended in January 2023. Please clarify it.

Figure 6: This map suggests a reasonable validation only for the surrounding of the Belvaux station or. This pattern suggests that the constructed model can be suitable only for the surrounding of the training station (Central Europe?).

Figure 8: What does the capital letter in the brackets show? If country code, as I assume, then it needs double checking. Similarly the elevation data for Athens-Pendeli needs checking.

Figure 9: What are the blank areas in Norway, Bosnia and Italy?

I suggest combining Table 1 and Table 2.

In addition, negative values look strange for amplitude in Table 2. Please check them.