Terrain Proxy-Based Site Classification for Seismic Zonation in North Korea within a Geospatial Data-Driven Workflow

Round 1

Reviewer 1 Report

Site response is of key importance in seismic hazard analysis. To obtain a high-resolution site response is quite challenging because of the lack of data, particularly for the North Korea region which is inaccessible due to military security. The current study aims to construct VS30 maps for three major cities (Pyongyang, Kaesong, and Nampo) in North Korea by using only a digital terrain model and its terrain features which can be obtained by satellite observation.

The interrelationships between VS30 and terrain proxy (elevation, slope, and landform class) are verified for defining the input layer in Machine learning (ML)-based regression models, in Seoul and partial Incheon where the VS30 data is available. The landform class is a new proxy of VS30 and sub-grouped according to the correlation with grid-based VS30. The best-fitting regression models were designed by training the geospatial grid in South Korea.

The methods and results were presented reasonably. The topic conforms to the subject of the Journal. The manuscript provides a new way to assess the seismic site effect. I am glad to read it.

Major comments:

1. The regression models were designed using the data in South Korea. As the author claimed: “The proposed model was possible because of the geomorphological and geological similarities between South Korea and North Korea.” May you should provide or introduce the geological background in the Korean Peninsula. This is the key point to assure your model still works in North Korea.
2. In table 5, The R² are all less than 0.25, implying that regression model fitting results are not so satisfying. The authors should discuss more on this point.

Minor comments：

1. Line 15， should rephrase the sentence.
2. Section 2.1, is not so related to the point of the study and should be shortened. Perhaps you could give the geological background here.
3. Figure 2, should give the unit of the color bar.
4. Check equation 1.
5. Line 238, should rephrase the sentence and explain clearly what is Vsi.
6. Line 304, how did you determine the number of hidden layers?
7. Should give more details about K-fold cross-validation, such as the number of samples and the specific number for K.
8. Figure 8, better to show the distribution of the residual.

Author Response

Authors appreciate reviewer 1's valuable recommendations. Please find the following authors' responses.

Author Response File: Author Response.docx

Reviewer 2 Report

1. The empirical relationships between SPT-N and Vs were employed to estimate the Vs30 in the training area. Generally speaking, these empirical relationships include some amount of error, but the authors do not discuss the amount of error.
2. According to Table 5, the regression models constructed in this study are not suitable to estimate the Vs30. R² of all models are quite low, and RMSE are quite large. Hence, the reviewer has doubts on the reliability of the regression models constructed by the authors. As the reviewer mentioned in the previous comment, the Vs30 in the training area was not directly observed in this study. The Vs30 was estimated from SPT-N, and the empirical relationship includes error. However, the authors employed the estimated Vs30 for machine learning without taking care of the error. Therefore, the constructed regression models are not reliable to estimate the distribution of Vs30 in North Korea.

Author Response

Authors appreciate reviewer 2's valuable recommendations. Please find the following authors' responses.

Author Response File: Author Response.docx

Reviewer 3 Report

The manuscript entitled “Terrain Proxy-Based Site Classification for Seismic Zonation in North Korea within Geospatial Data-Driven Workflow”, by H.-S. Kim, C.-G. Sun, M.-G. Lee and H.-I. Cho, presents an interesting work.

In general, the manuscript should be acceptable for publication but some serious problems must be repaired prior to publication. It needs some significant improvement. Some suggestions are as follows:

1. Please use different terms in the “Title” and the “Keywords”.
2. The abstract should state briefly the purpose of the research, the principal results and major conclusions. An abstract is often presented separately from the article, so it must be able to stand alone.
3. The English language usage should be checked by a fluent English speaker. It is suggested to the authors to take the assistance of someone with English as mother tongue.
4. You could enrich the scientific literature.
5. The authors could take into account the following publication:
   -Chousianitis et al. (2016): Assessment of earthquake-induced landslide hazard in Greece: From Arias Intensity to spatial distribution of slope resistance demand. B Seismol Soc Am, 106 (1): 174 – 188.
   -Okada et al. (2021) The Potential Role of News Media to Construct a Machine Learning Based Damage Mapping Framework. Remote Sens. 2021, 13, 1401.
6. In the P21 L646, please correct the name of journal from “C.A.T.E.N.A.” to “Catena”

Author Response

Authors appreciate reviewer 3's valuable recommendations. Please find the following authors' responses.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The paper was modified properly following the reviewers' comments.