The paper “Large-scale flood risk assessment in data scarce areas: an application to Central Asia” presents a comprehensive analysis in a region that is rarely given attention in natural hazards research. The authors identified the best available global datasets, combined them with local sources and connected the whole through a long modelling chain. The study is mostly sound methodologically and is an important contribution. However, the paper itself is not well structured and quite difficult to navigate. Therefore, a lot of my comments pertain to structuring of the paper as well as too much of some details, the lack of certain other details, and terminology used. Below, I describe the main issues found in the sections, then I discuss some overarching issues, and end with some minor comments.

Abstract

The last paragraph is rather out of place and should be in introduction or conclusions. Please replace this paragraph with some of the results of the assessment, including your projected climate change impacts.

Section 3.1

The datasets are well-known and widely used, so the section can be reduced to single paragraph that refers to details in the table. References and spatial resolution should be mentioned in the table. Fig. 2 is not needed, as you have 19 figures and it adds nothing to the analysis. Climate projections should be mentioned in the table, and the paragraph on this moved to section 4.1

“We used observed data from the KNMI Climate Explorer to assess and correct the ERA5-Land extreme precipitation estimates due to the discrepancy between point station data and grid averaged data”. The authors do not elaborate or cite any literature here, despite potentially significant influence on the results. Bias-adjusting climate data is a major undertaking and don’t see how the authors did it having only observational point data.

Section 3.2

The text on how the data were obtained (or not obtained) could be shortened, with non-essential information moved to discussion. It would be best to merge Fig. 3 and 4, and use a different colour for population, as the bright blue will make everybody immediately assume it shows the flood hazard map. Further, the paragraph from L203 is really difficult to understand in terms of what data were collected (extent, resolution, timeliness) and what is its use in the model.

Section 3.3

I think that this section should be limited to the hydrological models. The other paragraphs are very confusing as to what are your exposure data and how the vulnerability models look like as some local modifications are mentioned, but not described. The authors should rather continue with section 4.1 after the hydrological model description, and then create a new section that collects together information on your exposure data and vulnerability models that is now spread across the paper, and then continue with section 4.2. In this way, a clear structure will appear: flood hazard modelling described comprehensively in section 3, and the transition to flood risk in section 4.

Section 4.1

The section first summarizes the methods in a figure, then in a list and then a full description. I suggest to remove the list and divide the rest of the section into subsections corresponding to the 5 steps. As noted earlier, the climate change aspect should have its own subsection with a full description.

Here, there a crucial aspect of how the authors created the climate projection dataset. It is a very practical “solution” to modify ERA5 as if it was being bias-adjusted. However, only one reference is cited, which neither covers climate changes nor temperature. I haven’t seen this approach used for climate change projections, which normally use hindcasts for both historical and future periods, for modelling consistency. The authors should elaborate here and provide more explanation why this approach is used and whether it was applied in literature before.

Flood protection: the authors scale the protection using correlation between FLOPROS data and population density. However, FLOPROS itself doesn’t contain any actual data for Central Asia: it was created using correlation with GDP per capita. Then, the authors combine it with some local data. Due to the influence of flood protection on results, the map of the protection assumptions should be presented, or the data made available online. The way it is now, I can’t really assess this aspect of the authors’ work.

Section 4.2

I’m missing here information on what are the asset types covered, and how were the damage functions derived. How do the authors know that they are applicable to Central Asia?

Paragraph L407-416 is generally duplicated from the previous, though with some minor changes. Please verify which version is correct and remove the other one.

Section 5.1.1

Description of calibration in L458-471 should be in the methodology. Additionally, the authors should describe what parameters were subject to calibration (or provide suitable reference) as well as what period was calibrated. Then, description in L471-483 should be in the discussion. Then, though the authors show example figures, no statistical analysis is presented. The figures should rather go to a supplement, and replaced with tables or graphs showing summary performance of the calibration & validation indicating correlation, bias and/or metrics.

Section 5.1.2

As before, the introductory paragraphs should be part of the methodology section.

Then, it is not clear how the 2005 event is used in calibration. As I note in the paragraph below, the authors apparently compute observed loss in “current” value incorrectly, therefore spoiling the whole calibration. Then again, not sure how it feeds into calibration.

The authors write about “trending” and “trended reported losses”. I suppose you mean price inflation and deflated reported losses. Then, the calculation is wrong – it apparently applies the typical error of using foreign-currency losses and applying a local deflator or vice versa. In case of Tajikistan, high inflation is matched by loss of the value of the local currency (somoni) relative to the US dollar. Hence, the 2005 losses will only be 8-11 million USD in 2022 using the consumer price index, or 11-14 million USD using the GDP deflator (using data from IMF’s World Economic Outlook). Therefore, the result wouldn’t be far from the modelled result. Unless what the authors did is to exposure-adjust the 2005 losses, which would be somewhat consistent with a 4-fold increase in Tajik GDP since then (in US dollar terms).

The same goes for the subsequent analysis of 7 events (Table 3), hence a much better and clearer description how observed losses were adjusted to be comparable with the model, including original local-currency losses and the adjustments made.

Finally, authors should reduce the extensive discussion of data issues and leave it for the methodology and discussion sections.

Section 5.2

In contrast to other sections, there is very little comment on the figures, especially in 5.2.2, particularly in contrast to extensive descriptions in 5.3.

Section 5.3

A lot of the information in this section repeats the methods, or should be included in that part of the paper. Otherwise, there are 3 tables here showing the details on different scenarios. They should be rather in the supplement, while a table (or graphs) should contrast the scenarios with each other. Further, the use of per mille should be rather replaced by percentages (also next to numbers), making the results more self-explanatory. Finally, the authors lump together in the 2080 scenario the effect of climate change and exposure change (for one sector only). The authors should present those effects separately, and the exposure scenario preferably with contrast to the ‘present-day’ losses pertaining only to the residential sector.

Also, the authors suddenly mention here results of earthquake risk, which is not the topic of the paper. Related text should be moved to the discussion.

Section 6.1

This part contains information that should be part of introduction (motivation of the study) or conclusions.

Section 6.2

This section should be much expanded with elements that are currently in other parts of the paper (as mentioned above in the review). It (and 6.1) shouldn’t be in a list format, but as plain text. Information about availability of the data should be in the “Data availability” section in the end.

Geographical names

In the paper, the authors apply the geographical terminology inconsistently. Spelling mistakes and incorrect names are multiple. The authors should consult, in particular, the ISO 3166-1 and 3166-2 standards. Consequently:

• For consistency, use “Kyrgyzstan” short name from ISO rather than the long name “Kyrgyz Republic”, as you use short name for the other countries;
• Use “Region” rather than “Oblast”. The latter is a Russian name used only in two out of five countries in the study, so its use is inappropriate. ISO 3166-2 uses “Region” as the English name for the first-order administrative units of all five countries.
• Check the spelling of all regions mentioned in the paper according to ISO 3166-2, which provides the correct forms for the national languages of each country.
• The capital of Kazakhstan is called “Astana” again, since 2022.
• Region codes in figures 17-19 sometimes follow ISO 3166-2 codes, but mostly not. Please correct this according to that standard for consistency and to facilitate reuse.

Terminology

The authors’ use of “Large-scale” is problematic. Though the term is by now commonly used in papers, it is imprecise. In geography, it has actually reverse meaning (a large-scale map covers a very small area). In one place, the authors define it as “hundreds of thousands of square km”, though the study area covers 4 million km², in another they call it “country-scale”. Given the size of the study area, which is comparable with the entire European Union, the heavy use of global datasets, and the moderate resolution of meteorological inputs (0.1°) and the flood maps (3”), I strongly suggest for the authors to replace “large-scale” with “continental-scale” in the title and through the whole manuscript. That would introduce precision and improve the visibility of the paper.

The authors use “AAL” but the more common acronym is EAD (expected annual damage).

Minor

L88: “global” rather than “regional” (cf. section 3.1)

L292: how was the 0.1° climate data made to fit the 1 km grid of the hydrological model? The latter must use some geographical projection, correct?

L322: “0.00083° “ would be somewhat better called “3 arc seconds”

L378: “time horizon 2080” -> what does that mean, exactly? Normally, climate projections look at a particular window, e.g. 2071-2100.

L486: where parameters related to lakes and reservoirs part of the calibration?

Fig. 10: it is pretty pointless making a figure just to display a single data point per panel.

Authors should leave statements like the ones in L600-L602 for the conclusions.

Fig. 14: 4 out of 5 are national capitals, but not in case of Turkmenistan. Is Ashgabat not at risk? Also, the axis with the return period should be logarithmic for better presentation.

Fig. 15: as above, logarithmic scale is needed, plus a general improvement of the quality of the graphs.

Figures in general: some figures provide a scale, some not. It should be included in all of them and in general they should be, preferably, made more homogeneous in appearance. Also, if the authors use some kind of background image from external source, it has to be credited. If it’s simply the global DEM from Table 1, it still should be mentioned.

General comments

This manuscript presents a probabilistic assessment of fluvial flood risk for the countries in the region of Central Asia. My opinion is that the work is a relevant contribution to improve flood risk knowledge in Central Asia, providing a consistent transboundary risk assessment for all the region.

The overall methodology is appropriate for the task and makes use of well-established models and datasets, integrating where possible global-scale a local-scale data. The inclusion of several risk parameters in the analysis (e.g. damage for different economic sectors) is appreciable

Having said that, I think that the manuscript needs to be improved in some parts. The descriptions of some components of the methodology are rather short or incomplete and should be expanded. Similarly, the presentation of results should provide a more detailed overview of the different outcomes.

Specific comments

Abstract:  it is not well structured as it is now. Background information (lines 22-36) should be reduced or moved in the introduction (for instance, the work done in the project on pluvial flood and seismic risk). At the same time a short summary of the main elements of the methodology  and main results should be added.

Figure 1: please add country names and, if possible, other details such as the location of main rivers (in particular those mentioned later in the text) and/or main urban areas

Page 4 L135-137: can you please explain how you used observed data from the KNMI Climate Explorer to correct ERA5-Land extreme precipitation estimates?

Figure 2 is not relevant within the description,  so I suggest removing it

Figures 3 and 4: I think they can be merged

Page 5: please provide a reference for the GlobeLand30 dataset. Is it based on static or dynamic-over-time land use information?

Page 5 L 144 “This dataset has been made accessible to the United Nations (UN) through the UN-ESCAP Statistics Division and the UN_ESCAP ICT and Disaster Risk and Reduction Division”: please move this detail in the Data Availability section

Page 5 L 150 the reference for the original MERIT DEM should be Yamazaki et al (2017)

Page 5 L 163-168: I think that the details on the method applied to combined local-scale information on flood defences with the FLOPROS, WorldPOP and Landsat HBASE datasets (now in page 11 L333-348) should be moved here

Page 9: the description of the CA2D model is perhaps too long compared to the description of the other models

Pagge 9 L250-254: I suggest moving this description in Section 3.2. Also, can you provide more information on the exposure database? For instance, does it include building-scale or aggregated information? Which infrastructures are considered? In the following sections you name different types of infrastructures, they should be described here

Page 11 L301 : Generalized Extreme Value

Section 4.1 is quite long. Consider splitting it into two or more subsections (e.g. hydrological modeling , hydraulic modelling and stochastic analysis)

Figure 5 is not much useful so it could be deleted.

Page 11: did you undertake any calibration of the CA2D model (e.g. the roughness parameters)? Also, how did you identify  the river sections mentioned in the description?

Page 11 L333-348: see my previous comment about the opportunity of moving this description to Section 3.1. Besides that, how did you implement flood protections in the risk modeling framework? Did you explicitly include flood protections in CA2D simulations  (e.g. by modifying the DEM) or only in the risk analysis (i.e. assume that the all floods below the design level would cause no damage)?

Page 12 L375-383: there is little information on the climate scenarios. Could you please describe the underlying models that were used to define these scenarios (GCMs. RCMs) and provide references? How did you downscale future climate scenarios to match ERA5-land data resolution?

Section 4.2: some comments here: 1) there seems to be no description of how you applied SSP scenarios in the risk analysis, please provide details. 2) provide a complete list or table of all the risk parameters (impact for different economic sectors, population affected, mortality etc.). 3) put the description of calibration/validation in a dedicated subsection 4) it is sometimes difficult to understand which data were used for calibration and validation of the model (here and in Section 5 too), please try to make it clear.

Page 13 L430: you mention a list of historical events and reported losses, can you provide a reference for this information, is it coming from local governments?

Section 5.1.1: change section name to “hydrological model”

Figure 7: make sure that the Y axis for precipitation and temperature data is readable in all panels. Also, the small maps on the top left side of each graph are difficult to interpret (is it the river basin?). Perhaps it could be useful to put a separate map to locate the different river sections in the region.

Figure 8: the trend lines should be more visible

Section 5. 1.2: please put in dedicated sections the calibration of vulnerability curves and the validation of flood extent maps (they should be presented before the validation of modelled risk estimates).

Section 5.1.2: Here you  mention vulnerability functions for infrastructures and crops, they should be described in Section 4 (I undestsnf they are taken from previous studies but you should specify, for instance, if you used separated functions for each country).

Section 5.1.2: Can you describe how the observed flood extent for the 2005 event was derived? Did you carry out a quantitative comparison of observed and modelled flood extent model?  It would be useful to calculate some performance metrics (see Alfieri et al 2014 for instance) and check if the observed underestimation of impacts might depend on underestimation of flood extent. Also, please include the reported the flood footprint in the map in Figure 9.

The graphs in Figure 10 are not useful because there's only one point for each graph. I would replace them with a table or directly include the numbers in the text. In case the event is reported in global loss datasets (for instance, in the International Disaster Database EM DAT, https://www.emdat.be/) this could be used as an additional reference for comparison.

Page 19 L595: do you mean that you have calibrated the loss model based on this comparison? Or are you referring to the calibration done on the vulnerability functions mentioned before?

Page 20 Possible explanations for underestimated impacts are that pluvial flooding impacts were not considered, as well as impacts ng in the minor drainage network not included in hydraulic modelling.

Figure 14: the graphs in this figure are not much informative because water depth changes in each pixel of the model; instead, showing flood extent graphs over an area (a country or a river basin) would be more useful.

Section 5.3.2: The Global Assessment Report on Disaster Risk Reduction 2015 (GAR2015) produced risk profiles for all countries in the world for different natural hazards including floods. I think it  would be interesting for the reader to see how GAR estimates compare with those of the manuscript (country profiles are available at https://www.preventionweb.net/english/hyogo/gar/2015/en/home/data.html )

Section 5.3.2: Table 4 and figure 15 show more or less the same information so I would keep the figure in the text and perhaps move Table 4 in a supplement. The same applies for Table 5 and figure 16

Table 6: Comparing Tables 4, 5, 6 it seems that the absolute losses decrease in the 2080-SSP1 scenario, whereas relative losses increase. Can you explain this behavior? Did you observe a similar behaviour in other future scenarios? You might also want to check if your results are consistent or  disagree with existing global studies (e.g. Dottori et al 2018, with apologies for the self-citation).

Section 5.3.2. Currently this section only shows results for overall economic damage. I would recommend to provide an overview of all the results produced by the analysis (e.g. human fatalities, breakdown of damage per each economic sector considered, impact on infrastructures etc.). This information would be useful because it is rarely reported in similar studies in literature. Also, you should include in the discussion the results for future scenarios other than 2080-SSP1 (if deemed important, additional results could be added in a Supplement)  

Data availability: Please provide here the details for accessing all the datasets used in the study (or explain why they are restricted). For instance, several global datasets are freely available (Table 1),

Section 7: you could include a short summary of the main outcomes (e.g. countries with higher relative impacts, risk hotsposts)

References

Alfieri, L., Salamon, P., Bianchi, A., Neal, J., Bates, P., and Feyen, L.: Advances in pan-European flood hazard mapping, Hydrol. Process., 28, 4067–4077, https://doi.org/10.1002/hyp.9947, 2014.ferences

Yamazaki, Dai, Daiki Ikeshima, Ryunosuke Tawatari, Tomohiro Yamaguchi, Fiachra O'Loughlin, Jeffery C. Neal, Christopher C. Sampson, Shinjiro Kanae, and Paul D. Bates. "A high‐accuracy map of global terrain elevations." Geophysical Research Letters 44, no. 11 (2017): 5844-5853.

Dottori, F., Szewczyk, W., Ciscar, JC. et al. Increased human and economic losses from river flooding with anthropogenic warming. Nature Clim Change 8, 781–786 (2018). https://doi.org/10.1038/s41558-018-0257-z