Predictive understanding of socioeconomic flood impact in data-scarce regions based on channel properties and storm characteristics: application in High Mountain Asia (HMA)

Khanam, Mariam; Sofia, Giulia; Rodriguez, Wilmalis; Nikolopoulos, Efthymios I.; Lu, Binghao; Song, Dongjin; Anagnostou, Emmanouil N.

doi:https://doi.org/10.5194/nhess-25-3759-2025

Articles | Volume 25, issue 10

https://doi.org/10.5194/nhess-25-3759-2025

© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.

Special issue:

Hydro-meteorological extremes and hazards: vulnerability,...

https://doi.org/10.5194/nhess-25-3759-2025

© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 25, issue 10

Research article

|

06 Oct 2025

Research article |

| 06 Oct 2025

Predictive understanding of socioeconomic flood impact in data-scarce regions based on channel properties and storm characteristics: application in High Mountain Asia (HMA)

Mariam Khanam, Giulia Sofia, Wilmalis Rodriguez, Efthymios I. Nikolopoulos, Binghao Lu, Dongjin Song, and Emmanouil N. Anagnostou

Download

Final revised paper (published on 06 Oct 2025)
Preprint (discussion started on 08 Aug 2023)

Interactive discussion

Status: closed

RC1:
'Comment on nhess-2023-120', Jakob F. Steiner, 26 Sep 2023

Review attached.

Citation: https://doi.org/10.5194/nhess-2023-120-RC1
- AC1: 'Reply on RC1', Mariam Khanam, 24 Nov 2023
  
  We want to thank the reviewer for taking the time to provide such insightful, detailed comments and recommendations. We provided a complete response to each major point raised in the discussion in the attached PDF document.
  
  Citation: https://doi.org/10.5194/nhess-2023-120-AC1
CC1:
'Comment on nhess-2023-120', Donghui Shangguan, 01 Feb 2024
The manuscript “Predictive understanding of socioeconomic flood impact in data scarce regions based on channel properties and storm characteristics: Application in High Mountain Asia(HMA)” by Khanam et al. used LYI and ML methods to evaluate and predict the flood impacts and risk due to precipitation in HMA. This work is first time to evaluate socioeconomic impacts of flood hazards in data scarce region. However, it is not good writing. The structure is not reasonable. And the XGboosting tools is not clear to solve what? Thus, I would suggest it should be major revision.

General comments

In HMA there are also GLOF which risk the human being and infrastructure. If possible, please include evaluating the socioeconomic flood impact.

Data-Scarce regions should be clear (which data or which type of data). In HMA, population is scarce. And Socio activity is also low.

Specific comments
1.Section 2.2 methods. This title is not reasonable. Is section 2.3(Machine learning model) methods? In addition, the dataset and methods in this section should be divided, for example, 2.2.4 exposure(population) is datasets.
2.Line 135 Why is it classified by LYI values(<2,2-3, and >3).
3.Line 217 While XGBoosting is …，this sentence is incomplete.
Line 217 this section (machine learning model) is a little difficult to understand the role that is plays.
Citation: https://doi.org/10.5194/nhess-2023-120-CC1
- AC2: 'Reply on CC1', Mariam Khanam, 07 Feb 2024
  
  Dear reviewer,
  Thank you for the insightful comments and suggestions. We appreciate your time and effort. Please find the response in the attached file.
  
  Citation: https://doi.org/10.5194/nhess-2023-120-AC2
RC2:
'Comment on nhess-2023-120', Anonymous Referee #2, 18 Feb 2024

1. Clarity and Structure: The abstract is well-structured, presenting the problem, the proposed solution, and a few findings. However, some sentences are complex (starting from the title), and more concise wording could enhance clarity,
2. Methodology: The use of the Lifeyears Index (LYI) as a measure for socioeconomic flood impact is well explained. It would be beneficial to provide a brief explanation of how the geomorphologically guided machine learning approach works, even if it is in a bit summary.
3. Data: The abstract mentions training the model with over 6000 flood events from 1980 to 2020, but It is mentioned that the model shows variability from 1980 to 2022 as temporal Coverage. So what's the correct timeline?
4. Conclusion: A brief conclusion summarizing the main contributions and implications of the study would be beneficial.

Citation: https://doi.org/10.5194/nhess-2023-120-RC2
- AC3: 'Reply on RC2', Mariam Khanam, 06 Mar 2024
  
  We want to thank the reviewer for taking the time to review the manuscript. We appreciate the insightful comments and suggestions. Please find the authors' responses to the specific comments in the attached file.
  
  Citation: https://doi.org/10.5194/nhess-2023-120-AC3

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

ED: Reconsider after major revisions (further review by editor and referees) (22 Mar 2024) by Elena Cristiano

AR by Mariam Khanam on behalf of the Authors (01 Jun 2024) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (20 Jun 2024) by Elena Cristiano

RR by Anonymous Referee #3 (14 Jul 2024)

RR by Anonymous Referee #4 (15 Jul 2024)

Suggestions for revision or reasons for rejection

The authors present an approach to map socioeconomic flood risk in the HMA using a ML model. While the idea seems interesting and valuable, I think that the paper still requires a lot of work before publication in this journal.
There are some major points that lead me to this decision. The most important is that you trained the XGBoost model on a single nation, achieving what seems to be overfitting and then you deploy it to a way bigger region, without any sort of validation. This leads to the possible conclusion that any results that you report, while interesting they might be, they can be wrong or biased towards similar mechanics as you would have in Nepal.
Moreover, there is no clear description on how you trained and validated the model and how you assessed the parameter importance.
Overall, the paper is also improvable in terms of writing, with several redundant phrasings and unclear sections.

I also leave hereafter several minor comments:

Figure 2: from the figure it is not clear why is the LYI both an input and an output of the model. Despite the caption saying to refer to the text, this should be more clear in the image itself as well

Line 145: What is the point of using district-level data if you then aggreagate it at the watershed level, especially for the testing on the HMA region?

Line 147: How do you weight differently the districts?

Table 1: There's no need to repeat the "Y =" in the descprition, since you already have it on the left

Line 169: you often use capital letters for Low, Middle, and High risk. Consider using lowercase letters. This applies also for the rest of the paper.

Lines 194-195: In theory, you can also fully automate from terrain data if you use the contributing areas, since you can delineate them based on the topography. I guess you could argue that in your way you obtain a better representation of the bankfull discharge, but you would have to prove it, for example by comparing the two approaches. It is thus not clear why you did not use the other approach.

Line 209: You are comparing with HAND, which is not exactly a standard inundation model, as it only models the water depths with several assumptions. I think it is overall fine to compare with it but mention at least its limitations.

Figure 3: please reduce the size of the plot, as there is no need for it to be that big.

Lines 211-212: Didn't you just say that you used a modified version of this index? If that is the case, then your claim is wrong since only the GFI has been validated but not the FGP.

Section 2.3.2: please add the formula used for the FGP calculation in the text, rather than in the figure. You can leave the figure that explains the variables that you use.
Figure 4a: what is w=alfa*A^beta? I understood from the text that the reference height was determined from the landscape rather than from formulas. See also my previous comment on it.
Figure 4b: Include in the caption that you are also showing the corresponding orthophotos of the sites you are considering for the flood maps.
Moreover, the legend is quite small; consider increasing it.

Line 249: I don't understand the need to clarify that one variable is in the x axis and the other in the y axis. Consder removing them for clarity.

Eq 2: I would move this equation to line 241, just before explaining it.

Figure 5: it would be useful to color the different areas under the curve with different colors or patterns to help the reader.

Section 2.4.1: There's no indication on how the model is trained, validated and tested; how many samples are used for each; how the results are assessed in terms of metrics.
This section is also very unclear when you describe the comparison with the DFO database.

Figure 6 is not clear enough from the legend and caption.

Section 3.2: How was this variable importance assessed?

Lines 344-346: This should go in the methodology section. Moreover, classification metrics with more than 2 classes should be better discussed as they are not as straightforward as with only 2 classes.

Figure 7: the F score, which should go from 0 to 1, has values above 3000. Please correct the legend.

Table 2: "a classification model" is too generic. Please specify that this is for your trained model, when applied to the test dataset (or validation?, this is not clear)

Line 366: This seem to indicate overfitting. Did you use a validation dataset to limit this?

Section 3.4: Maybe it would be valuable to add some correlation plot to understand if there is indeed a match between DFO and LYI, instead of having a table which has redundant entries.

Hide

RR by Anonymous Referee #5 (28 Sep 2024)

ED: Reconsider after major revisions (further review by editor and referees) (01 Oct 2024) by Elena Cristiano

AR by Mariam Khanam on behalf of the Authors (04 Apr 2025) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (04 Apr 2025) by Elena Cristiano

RR by Anonymous Referee #5 (07 Apr 2025)

ED: Publish as is (22 Apr 2025) by Elena Cristiano

ED: Publish as is (25 Apr 2025) by Paolo Tarolli (Executive editor)

AR by Mariam Khanam on behalf of the Authors (06 May 2025) Manuscript

Short summary

This study comprehends and predicts the socioeconomic effects of floods in the High Mountain Asia (HMA) region. We proposed a machine learning strategy for mapping socioeconomic flood damage. We predicted the life year index (LYI), which quantifies the financial cost and loss of life caused by floods, using variables including climate, geomorphology, and population. The study's overall goal is to offer useful information on flood susceptibility and subsequent risk mapping in the HMA region.