GENERAL COMMENTS

The presented dataset stores high quality soil physical data. The description of measurement methods and models applied to compute soil hydraulic parameters by fitting the moisture retention and hydraulic conductivity are detailed and clear. Structure of the manuscript is logical. The main strength of the database is the data on unsaturated hydraulic conductivity. This way the presented work and dataset will attain international interest.

The data could be easily accessed. Organization of the six data tables within the dataset is logical, the tables can be merged by the Sample_ID column.

A paragraph could be added about data quality check under materials and methods, because that could strengthen that the dataset was rigorously checked and the way the check was performed can be very informative for the readers and serve as a guideline. A final data check would be useful to secure that all data is correct. The detailed review can be found under SPECIFIC COMMENTS.

SPECIFIC COMMENTS

L24-25, L55, L78, L79, L82, L96 and entire text, please specify if you refer to soil profiles or soil samples, the word “data sets” is not enough specific.

L97: … basic soil properties such as soil texture … or something similar

L101-102: please add reference or some examples for the two level texture information, because it is not widely used.

L127: … mixed average soil sample … is it correct?

L134-149: all is clearly described, just a table providing an overview about the methods would be very informative, because for the readers it is a very valuable information what method was used for which soil property. Please add information about the measurement method of N and S, as well – because those are also included in the BasicProp.csv file. Please consider if the method used by soiltexture package can have limitations. Some other methods exist, which might result in a more accurate conversion to USDA silt and sand content. It is possible that in your case there would not be significant difference between different methods, but for other cases there might be. Readers might follow the procedure you published, so it worth to mention other methods, e.g.: Nemes et al (1999) https://doi.org/10.1016/S0016-7061(99)00014-2.

L150: Before “2.4 Fitting models to measured data” subsection could you please add a separate subsection on how quality of the data was secured? Could you shortly describe what rules were applied during checking the data?

L181-182: please add reference and equation used to compute parameter Ks of the PDI model.

L184-186, Table 1: please add meaning of VGM and PDI to have the table self explanatory.

L190-193: would be informative to add 4.1-4.3 tables from 2023-012_Hohenbrink-et-al_Data-Description.pdf file here.

L194: It might worth to consider to create a metadata .xml file following the INSPIRE metadata guidelines (ISO 19115 and ISO 19139) and add it to the dataset.

L216-219 and Figure 2. : please consider to provide this information according to USDA texture classes (based on the USDA sand, silt and clay fractions), because that is internationally used, the German texture classes are not widely known out of Germany. I see that for Figure 3. it might not make sense to use the USDA standard because than you might have only three fractions and Figures 4 and 5 is easier to interpret if meaning of texture classes can be read from Figure 3.

L241: circles on Figure 4 are hardly visible, maybe Figure 4 could be edited somehow to let easier distinguish between circle, triangle and square.

L244: Please shortly add why number of dewpoint measurements ranges between 1 and 8.

L263: … range for coarser texture classes … Do you agree?

L268-271: if th_1_8, th_2_5 and th_4_2 columns of Param table were computed with PDI model, please add “_PDI” as last characters to those column names.

L272-273: please add very short explanation for why filed capacity and wilting point vary widely within texture classes. This is obvious for experts in soil physics but not that trivial for researchers from other environmental fields.

L308: please consider e.g. the work of Twarakavi et al. (2010) (https://doi.org/10.1029/2009WR007939 ) - or possible other papers in this topic – and rephrase the sentence accordingly.

L311: Do authors plan to add soil depth, chemical soil properties - e.g. pH or calcium-carbonate content - or taxonomical information to the dataset in the future? If soil depth is available it might be easy to add to the BasicProp.csv table, it could be an important data column.

Result of checking the database:

1. there is a negative theta value in RetMeas.csv, please check and revise/correct.
2. there is a negative value for S in BasicProp.csv, please check and revise/correct.
3. Sum of USDA sand and silt and clay is 99.9 and 100.1 for some samples, it might worth to correct them to sum up to 100.

The manuscript reports on a decently homogeneous data collection of soil physical and hydraulic properties that, in many aspects, provides more detailed data than what is available in most existing and internationally available databases. However, in certain aspects it provides less information, or potentially uses a weak solution for data harmonization. If unresolved, these can become its limitations, and eventually limit the anticipated benefit from reporting very detailed water retention data. This contribution of data is very welcome in the literature, but it would be desirable to report the data and some of the methods in more detail. Three generic and a number of specific comments follow.

• Some data, primarily particle size distribution data should be reported in more detail. I don’t see trail of reporting more than sand-silt-clay contents, whereas the original set of measurements should be reported. There are emerging directions of research that would utilize that. It should also be communicated which of the silt-sand data pairs are from original measurements, and which have been a product of interpolation that adds additional noise.
• In addition, instead of the rather standard, flat data description, it is better value to go in depth on the exact steps that involved data manipulation, and present the outcome in a convincing way. I specifically refer to the particle-size conversions and its uncertainty, as well as the derivation of bulk density and return to these in the detailed comments section.
• Several statements made would have been true in the 1990s, but not anymore. I highlight some among the detailed comments. The authors should revise those and bring the statements up to standard according to the state-of-the-art in the 2020s.

Specific comments:

Abstract: Include a brief reference to the geographical extent (i.e. the contributing list of countries, with Germany dominating)

L63: I believe Brazil has non-tropical data in HYBRAS as well. Please check and remove the word ‘tropical’ if necessary.

L66: replace ‘commonly’ with ‘openly’

L72: …as it has been often done historically.

L75: … which are often not recorded at the time of sampling.

L85-89: IT has been identified that ROSETTA’s data is also gepgraphically skewed. It would be worth exploring where the 235 samples with unsatK come from.

L99: i.e. the evaporation

L99: dew-point potentiometry (Campbell et al., 2007)

L101: these are not two ‘levels’, but two standards. Use e.g.: “provided according to both the German and the USDA classification systems, and the…”

L103: please avoid using text like “strong foundation”. The users and history will decide that.

L104-105: delete this sentence, it is repeated from earlier. Remove the dependence of the next sentence on this sentence (reference to further purposes).

L109: This is the 3^rd mention of “various original purposes”. Perhaps remove both earlier mentions.

L129: here you mean aggregated DISTURBED samples, is the correct?

L131-132: To me the “accuracy….smaller than” structure limps. Revise? Uncertainty in their geo-position is less than 100m? Etc. Btw, is this true for all samples? If not, please state.

L143-144: Please list which methods those were. Was PARIO also involved?

L147: To my understanding using the “soiltexture” R-package means that essentially a log-linear interpolation. Is that correct? If so, I have to be critical of the approach. More advanced approached have already been used to re-classify European data more than 20 years ago. The key is to reduce resulting biases. If alternatives have been looked at, please justify why still this approach was to be used.

L148-149: Re bulk density calculations: was the missing volume due to the earlier positioning of HYPROP tensiometers accounted for? Perhaps so. Please state for the record, so that others also think about it in the future.

L169: near-saturated conditions…. (please also consider explaining in half a sentence why PDI enables that prediction better)

L170-171: Sure, but please provide a very short summary of the method and the choice of -6cm.

L190: Has an SQL-supported, searchable single-file database format been considered?

L191: “Soil texture” is derived information, especially after interpolations. Is the raw particle-size data reported? In what format? How many points typically? It would be best practice to report it, so that future users can make their own choices of interpolation, as well as just have the more detailed particle-size data.

L191: Still about sol texture: Is it reported in the database for which samples the original measurements were according to the German standard (silt at 63 microns) and for which those were according to the USDA/FAO standard (50 microns). Obviously the interpolation is for the other.

L198: replace “contained” with “available”

L201-204: Obviously there is large disparity in geographical distribution. My first gut feeling was: why not to limit the data to Germany? - but that would lead to loss of data. As an alternative, the authors could/should provide some information (data distribution, similarity in methodology, standards, use of particle-size interpolation (see above), etc.) that helps the eventual user decide against cutting off the Canadian, Japanese and Israeli data – citing methodological inhomogeneity - right away prior to running an analysis. For me, for instance, losing ca. 10% of the data in exchange for gaining more homogeneity seems like little cost to pay.

L215: this is only true if we cut “natural soils” at the boundaries of temperate climate. Sandy clays and that region of the texture triange that is blank here are frequent in the tropics.

L214-229 (section 3.2): This is a rather flat statistical summary that could be greatly shortened or just relied on in a small table. Instead, it would be much more useful to read about the handling of the raw particle-size data (why not include in the datadase?), interpolation (convince the user you chose the right method, provide which point was interpolated for how many samples that now carry extra uncertainty, etc.). I find the currently provided detail to be insufficient. You work with international data, and the particle-size conversion/harmonization aspect has been a bottle-neck in every one of such projects earlier (e.g. HYPRES, EU-HYDI) where data harmonization took place at all.

L227-229: Definitely delete this. This is basically coded into the texture classes’ definitions or even their names. As if one said that “sand content was higest in sands”.

L241-243 and Figure 4: the circles, trianges and squares are only identifiable under extreme magnification. Please find another way of identifying them. Perhaps only refer to the pF ranges? Or colors?

L248: (a) Please define what ‘dry range’ means. (b) What would be against suggesting that the mini-disk infiltrometer could be used for this in the laboratory?

L261-263: Can you please suggest why that is? Is it more realistic, or only a fall-out of model constraints?

L268: Why not just call them “commonly derived properties” …to evaluate the ability of a soil…..

L273-274: PAW is most often not the highest in the finest textured soils, but rather the intermediate to intermediately fine textured soils. Can you please refine the statement, or justify your stated finding?

L288: saturation levels compared

L293: saturation levels

L295-298: This statement and follow-up elaboration does not make much sense in 2023. Most internationally used databases hold data _only_ of undisturbed samples. Some old ones have some disturbed ones. If you want to keep this statement, please give justice to the internationally known databases and cite which ones are based on disturbed and which are based on undisturbed samples. The statement here suggests as if this database is unique in this sense, whereas it is not.

L304: use comma before and after “similarly to Weynants et al. (2009)”

L307: Remove “Besides”

L308-309: This sounds like a statement from the 1990s. Please remove/revise/update according to the state of the art. (1) ALL “continuous” PTFs use particle size data, and not only classes or texture groups; (2) already in the late 90s some studies have evaluated the benefit from using finer-resolution particle-size data, as well as alternative representations of particle size distribution in PTFs (e.g. geometric mean and std of the curve), (3) the effect of using (and misusing) different classification systems was also evaluated at least 2 decades ago. You can use the Rawls and Pachepsky 2004 book as base reference, and find the most relevant literature therein.

L311: re: “more accurate PTFs”: That always depends on the bottleneck in each source database and PTF development tool, as well as noises and biases in a database. Here I think there are at least two obvious data-bottlenecks: (1) not publishing the as-detailed-as-possible, original particle-size data (if that is the case), and (2) using a relatively weak interpolation technique to get from one classification system to the other. Otherwise yes, the data collection has good values.

L321: and provide

L322: handle such dependencies.

L325: which led to more

L327-328: Sure, but that is easier said than done. It can also be addressed by using “local” PTF solutions instead of the usual “global” ones across the data domain. A local type PTF algorithms can work with dense data where it is dense within the overall domain, and scarce data where it is scarce. It is easy to set such techniques up to quantify that and communicate it together with the estimate, along with estimation uncertainty. The state-of-the-art has changed in the last 20 years.

L332: repeated from an earlier comment: what is against spelling out that the mini disk infiltrometer wold be suitable to respond to the need for “near saturated hydraulic conductivity”

L356: refine this to “compared to most data in existing databases”, because there are actually evaporation-based data in some of the relatively newer databases like EU-HYDI, but even HYPRES has such already from the 1990s.

L368: delete “also”