Hazard Assessment Based on the Combination of DAN3D and Machine Learning Method for Planning Closed-Type Barriers against Debris-Flow

Round 1

Reviewer 1 Report

This paper develops a framework to generate hazard assessment maps using the results from DAN3D. The width, volume, and impact load of a debris-flow can be calculated, which are used to determine the location and design parameters of a barrier. This is an important topic since barrier is an effective mitigation against debris-flow and there has not yet been any clear guidelines for designing a barrier. The results of this study provide a solution to this problem to some extent. The paper is generally well-organized and well-written. This paper can be accepted after the following issues are addressed, as specified below in the following comments.

1.     Line 20: You stated that “the framework results show agreement with the observed data”, but no observed filed data were mentioned in the text. You are suggested to provide the observed data regarding the Mt. Umyeon debris-flow, such as the width, volume, force and moment of the debris-flow to compare with the calculated results as shown in Table 2. If you don’t have these data, the statement in Line 20 should be deleted.

2.     Isn’t it clear that the debris-flow particles from the same source should be classified into one cluster? Therefore, it seems to be unnecessary to apply the k-means clustering algorithm for the classification.

3.     Two criteria were provided to determine the merging of the two debris-flow, but it is unclear whether or not both criteria are required to be satisfied or only one of them. The same problem is also for the criteria to determine the splitting of a debris-flow.

4.     You only discussed the combined parameters of the multiple debris-flow clusters that collide against a barrier in succession. How to determine the combined parameters if multiple debris-flow clusters (without merging) colliding against a barrier at the same time? For example, if the debris-flow clusters from Source 1 and 2 (Fig. 11) do not meet the merging criteria, but they collide Barrier X simultaneously, how to calculate the combined parameters in such a case?

5.     Several typos are present. Here are some examples:

Line 284: “3.2.5” should be “3.2.4”

Line 323: “ the deposited” should be “deposited”

Line 380: “demonstrates” should be “demonstrate”

Line 448: “shown that the” should be “shown the”

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

see the attached file

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

This manuscript aims to propose a framework to generate hazard assessment maps using the results from DAN3D numerical analysis. The framework incorporated algorithms and machine learning methods. A case study of the debris-flow at Mt. Umyeon, Seoul, Korea, in July 2011, was used to verify the applicability of the model. Results show agreement with the observed data. I carefully read this manuscript, although the manuscript includes interesting information related to assess hazard for designing barrier, there are several concerns about the manuscript need to be covered. The following comments may help to improve the work of the authors.

 

General comments:

As far as the reader understood, machine learning method and computer algorithm have been incorporated into the framework to automate the process. However, it is not clear how we can integrate these algorithms to generate hazard assessment maps. The step-by-step procedures of the proposed framework should be added and clarified in details as shown in recent publications (Lyu et al. 2019). Lyu, H.M., et al (2019). Scenario-based inundation analysis of metro systems: a case study in Shanghai. Hydrology and Earth System Sciences, 23, 1-15. https://doi.org/10.5194/hess-23-1-2019 Graphs illustrate the clusters performance in the k-means clustering algorithm need to be added to clarify the performance and efficiency of this algorithm in this process. In the workflow of k-means clustering algorithm, what are the stopping criteria and how it is evaluated in this framework? Otherwise, all information related to the utilized software and the scikit-learn Python library should be added. The reviewer has no problem with using machine learning models propose by the autohor. However, it is not clear how to adjust the database in the machine learning technique. The statistical analysis of the utilized parameters need to be added. Otherwise, the specification of the main parameters that used in this AI model should be added, please refer to the following publications in international.

https://doi.org/10.1007/s00366-019-00855-5;

https://doi.org/10.1109/ACCESS.2019.2930520;

https://doi.org/10.3390/app9040780.

The optimization of parameter should be conducted for the AI analysed results (Jin et al., 2019; Zhang et al., 2019a-c; Yin et al., 2017).

Zhang P, Yin Z-Y, Jin YF, Chan T (2019). A Novel Hybrid Surrogate Intelligent Model for Creep Index Prediction Based on Particle Swarm Optimization and Random Forest. Eng. Geol., https://doi.org/10.1016/j.enggeo.2019.105328

Jin YF, Yin Z-Y, Zhou WH, Horpibulsuk S (2019a). Identifying parameters of advanced soil models using an enhanced Transitional Markov chain Monte Carlo method. Acta Geotech., DOI: 10.1007/s11440-019-00847-1.

Jin YF#, Yin Z-Y, Zhou WH, Shao JF (2019b). Bayesian model selection for sand with generalization ability evaluation. Int. J. Numer. Anal. Methods Geomech., 43(14): 2305-2327.

Jin YF#, Yin Z-Y, Zhou WH, Huang HW (2019c). Multi-objective optimization-based updating of predictions during excavation. Eng. Appl. Artif. Intell., 78: 102-123.

Yin Z-Y, et al (2017). An efficient optimization method for identifying parameters of soft structured clay by an enhanced genetic algorithm and elastic-viscoplastic model. Acta Geotech., 12(4): 849–867.

It is not clear how to simulate the cluster width in the machine learning techniques. The variations of the important input parameters versus the output prediction should be added as histogram or detailed statistical characteristics of each dataset. It is not clear, what is the conditions that are useful for the applications of this model. Otherwise, no any clear information related to the ground conditions. It is not clear, how the authors evaluated the performance of the machine-learning model? How we can verify from these results? Thus, the utilization of this framework is difficult to follow and difficult to be applied. Some recent publication related to application of machine learning in engineering in details can be found in following international journal publications, please refer to these publications.

https://doi.org/10.1007/s00366-019-00855-5;

https://doi.org/10.1109/ACCESS.2019.2930520;

https://doi.org/10.3390/app9040780.

9 Recent development for risk or hazards assessment, please refet to the follow publications in international journals.

https://doi.org/10.1016/j.tust.2018.10.019;

https://doi.org/10.1016/j.scs.2019.101682;

http://dx.doi.org/10.1061/(ASCE)CO.1943-7862.0001757.

 

Specific comments:

This abstract includes two gaps that need to be covered, (1) the main points of the established framework need to be clarified, (2) no any effective or clear results can be found. The methodology needs to be defined in a good manner. Abstract is not clear and need to be observed in a good manner. Abstract should be written as following sequence: 1) the gap or the problem investigated, 2) methodology used to solve the problem, 3) key points of conclusions and 4) practical suggestions/recommendations. The introduction is interested with useful information related to the numerical DAN3D model. However, no any information related to the applicability of machine-learning technique in this manner. Otherwise, no any previous researches are illustrated to clarify the ability of the AI techniques in the prediction of the width of the debris-flow or to find the optimal location for barriers. Line 29: what is the meaning of (pp. 1-7) and line 56 (pp. 135–158)? Line 93: what are these assumptions? Line 94: it is not clear how the initial landslide volume can be divided into N number of particles? This framework should be illustrated in details. Otherwise, the step-by-step procedures of the established framework should be considered. Line 153: how we can verify that the stopping criterion is satisfied? Line 154-155: “The results of the k-means clustering algorithm are the centroids of the clusters and labels for each data point that assigns the data point into one of the clusters” this sentence should be clearly discussed in more details. 5 needs more explanations. Scale size of Fig. 9 need to be added. 10 should be clarified in a good manner.

In section 4, the quality of the English used is variable, with some poor usage. please enhance the quality of this section, e.g., line 405.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The revised version of the paper seems now clear: if the writer has understood, authors use model-based hazard assessment for planning the position of debris-flow control-works. As the adopted debris-flow routing model (DAN3D) shows some insufficiency, they combine it with a machine learning method. If it is this the case, this proposal should be more evident, both in the title and in the text. For istance, the title could be “Hazard assessment based on the combination of DAN3D with machine learning methods for planning barrier against debris flow”. Again the connection  between DAN3D and machine learning method is not clear and should be clearly explained (Section 3). Moreover, the computation of load the recent works of Huebl et al. (2017) and Rossi and Armanini (2019) could help. In addition, Figure 10 seems very similar to Figure 2 of  Suda et al. (2009).

Lines  15-17 The writer does not agree with “however,  solely  using  DAN3D  or  other  numerical  simulation  methods  would  be  both  insufficient  and  inefficient  in  finding  the  optimal  barrier  location.”. The DN3D model has the insufficiency in the sense it does not compute the width of area inundated by debris flow. Other models computes it. Please rewrite the sentence.

Line 12: it is better “The location and the design of a barrier”

Line 44: volume, height and impact forces are not parameters but characteristics quantities of debris flow. They becomes parameter in the case of control-works planning and design.

Line 59: “cells model” instead of “cell”.

Line 62: “by using debris-flow case  studies” is unclear.

Line 75: “introduced“ is better than “further explored”

Line 84: “constructability” is not a well suited word. Perhaps the “building possibility”?

Line 86: “required parameters” see the comment for line 44

Line 483-484: it is not clear how Figure 14 shows the increase of debris-flow clusters.

References

Hübl, J., Nagl, G., Suda, J., Rudolf-Miklau, F., 2017. Standardized stress model for design of torrential barriers under impact of debris flow (according to Austrian standard regulation 24801. Int. J. Erosion Control Eng. 10 (1), 47–55.

Rossi G,  Armanini A. (2019) Impact force of a surge of water and sediments mixtures against slit

check dams. Science of the total environment. 683, 351-359

Author Response

Please see the attachment.

Reviewer 3 Report

The authors just replied the detailed comments. But the General comments in the first round review, they did not do response or rebut.

Please response and revise. 

Comments for author File: Comments.docx

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 3

Reviewer 3 Report

Now it is acceptable for publication.