Impact of Rainfall-Induced Landslide Susceptibility Risk on Mountain Roadside in Northern Thailand

Round 1

Reviewer 1 Report

This paper applies ANN model to predict risk of rainfall-induced landslide based on the data collected in Northern of Thailand. Please revise the paper base on the following comments.

It was mentioned that there were 11 papers applying ANN to forecast landslide so that what makes this paper difference from the previous papers. Number of input parameters and datasets? Please elaborate.

Line 142, a total of 13 literature reviews from 2019 to 2021 summarized. All these papers should be cited properly.

There was no discussion on Fig 5. Fig 5 should be discussed.

In 5.1 data collection, how can the risk data be related to Table 1? There were 7 landslide events collected. However, 7.45% risky areas which is about 744 data points is presented. Please elaborate.

It is not clear how many hidden layers and hidden neurons were used. 5 neurons in hidden layer 1 and 1 neuron in hidden layer 2? Sometimes only one neuron is enough and provides high accuracy by increasing number of hidden neurons.

There was no discussion about ANN model validation.

I understand that this is a classification problem with the outputs of only risk and no risk so that how can the risk percentage be obtained? What does it mean by the number?

Please revise Fig 12c.

In Fig 15, how are the other factors considered in this chart? There are other factors that can be applied e.g. land cover. It is recommended to apply more factors to do sensitivity analysis in this section to present better results.

Author Response

The revision will be presented in the attached file.

Author Response File: Author Response.pdf

Reviewer 2 Report

Thank you for submitting your manuscript to the Infrastructures journal. Generally, the topic fits into the scope of the journal. However, the content needs some revision for clarity. Moreover, nearly all figures are blury.

The topic under investigation is relevant and important, and the modeling approach is innovative. For the reviewer remains unclear why not established models like scoops3d haave been used.

In the literature review, it is important that the scientific novelty of the work is established through a critical analysis of related literature. Furthermore, in the introduction must be given a general overview on the subject under review, and the motivation why the study is performed as well as the detailed scope of the study. Thus, the main questions of the reviewer are: What is the scientific motivation for the study? Which scientific question(s) shall be answered with this? What is your scientific hypothesis that you wish to answer with the investigation? Putting the scientific motivation will also help you to identify the novelties that characterises a scientific publication.

Regarding the methodology, it should be explained which data are the 9602 data, means which type and in which resolution. Anyhow, information on the resolution of the data sets must be given. Moreover, it must be added information on the methodology for model calibration and validation, and information on uncertainties.

In the conclusions, in addition to summarising the actions taken and results, please strengthen the explanation of their significance. It is recommended to use quantitative reasoning comparing with appropriate benchmarks, especially those stemming from previous work.

Author Response

The revision will be presented in the attached file.

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper has serious methodological flaws.

1. Lots of speculations about the role of variables are done without proper references to support those assumptions. 
2. How was the importance of the variable determined? You have to determine the significance of the variable and retain only those which are statistically significant. The variables were chosen based on speculation. 
3. The selection of landslide risk as a dependent variable is too subjective. Additionally, it was input as a binary (0 or 1) variable, but when the results were analyzed, it was presented as numerical (decimal) values between 0 and 1. You cannot input the data as binary and predict the result as a continuous variable. Also, while observing the nature of input data, most of them were “no risk” (Figure 12). The input data already looks flawed.
4. The paper lacks the background on how data sets were collected, such as what kind of raster data (digital elevation model, type, satellite sensor, resolution etc). It only mentions that it was done in QGIS. For example, how was slope derived? What is the source? Also regarding rock type, how was it used as an input data? The paper has no mention of its role in the result and discussion section (Figure 15). 
5. The rainfall data used also has serious flaws. The rainfall data from a few stations were used to create the isohyets via interpolation to feed into the model (~9000 data points). It involves too much extrapolation.  
6. Reference section has many articles which are not peer-reviewed including blogs (reference entry 29), and are inaccessible .

Author Response

The revision will be presented in the attached file.

Author Response File: Author Response.pdf

Reviewer 4 Report

This paper reports on the development of the risk model for forecasting the risk of landslides on the mountain roads in the northern Thailand. The model uses a two-layer artificial neural network (ANN). Several variables were considered, and the data included a total number of 9,602 data points. As could be expected, the cumulative rainfall and slope of the road forecast best the landslide risk. The results could be used in developing risk maps and early warning systems.

Name: (1) Are the authors studying the impacts of landslide risks (on people, nature, economy), or the effects of different variables on the landslide risk? I recommend changing the name accordingly.

Keywords: (2) Northern Thailand. No-code platform?

Introduction: (3) The authors should describe in Introduction the scientific contributions of their findings. Now they give a target of their work (a research problem), but how they strive for this target. What have they done? They should also write at the end of Introduction a short paragraph showing the content of their paper, e.g. in the form “The rest of the paper is structured as follows: Section 2 describes the factors that effect on the landslide risk. Section 3 …”.

Further comments:

(4) Are the “thirteen literature reviews” that Figure 3 is based on, in the list of references. They should be, and they should be given in this connection.

(5) Discussion on “coding” ANN algorithms (line 163 and forward) seems out-of-date. The programs for this have existed at least since 1990’s.

(6) Landslide risk in the input data: This is explained in a somewhat controversial way. Line 200 says “data collection of key variables in the study areas from past landslide incidents”. This lets us understand that only the cases (and locations) of landslides are considered. However, Fig. 12 tells that there were a landslide risk only in 7.45% of cases. I guess, it means that in this many cases, the landslide has occurred. This brings in the basic difficulty in this case: how to make the model forecast the risk (from zero to one) with the 0/1 training data.

(7) A clear description of the input data set is needed. There are some hints in the text that wake up the curiosity, but give no answer. E.g. lines 206-7: “along with collecting the recorded local damage photos and news to convert them into datasets for modelling”. To me this tells that the input data is arranged according to locations (numbering 9602). Is it so? How the timely behaviour (repeated landslides) are taken into account? A good figure would clarify the situation. It is hard to believe that these “damage photos” are included in the input data.

(8) Lines 277-8: “the road areas with metamorphic and sedimentary rock soil were selected, which will be discussed later”. I cannot find this in this paper. If it is a potential future research field, list it at the end of Conclusions together with other similar ideas.

(9) Results. Under Figure 13, it is said that Figure includes the results from 25 tests. I can see only nine tests. From Section 5.3, we learn that the authors study only the effects of two variables on the landslide risk. On line 291, what does this sentence “by preparing a new set of data” mean? How does the ANN manage with this new data? What accuracy was reached in training? How was it in testing?

(10) The authors are using a conventional feedforward ANN. Did they consider any other approach? I would like to see some text about what made them to choose this approach.

This is an interesting paper on the application of the neural network in risk forecasting. It needs some revision to enhance its readability. The language is reasonably good. I found a couple of typos, but a good proofreading is enough to correct them.

Author Response

The revision will be presented in the attached file.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Thank you very much for the revision. However, there is not much improved in the revision. 

-The data collection part is not clear. This is an important part. There are sections 4.2 and 5.1 for data collection. Section 5.1 should be relocated to section 4. This presents the range of the datasets (I believe) but it is not well presented and unclear.

-X axis and Y axis in Fig 13 are missing.

-Can this ANN model be adopted for other countries? How can this model be applied to other areas? Please elaborate the limitations of this study in the conclusions.

Author Response

Thank you for your comments, the revision is attached.

Author Response File: Author Response.docx

Reviewer 2 Report

Thank you for providing the revised version. The revised version has still gaps and unclarities.

One comment is still open: it should be explained which data are the 9602 data, means which type. For me the methodology is still not clear. How was considered the soil properties, the digital elevaton model etc? Those data are usually needed for landslide modeling.

The majority

Author Response

Thank you for your comments, the revision is attached.

Author Response File: Author Response.docx

Reviewer 3 Report

Authors have not addressed the earlier comments in the manuscript. The manuscript shows very few changes from the previous one. 

Author Response

Thank you for your comments, the revision is attached.

Author Response File: Author Response.docx