Use of Mamdani Fuzzy Algorithm for Multi-Hazard Susceptibility Assessment in a Developing Urban Settlement (Mamak, Ankara, Turkey)
Round 1
Reviewer 1 Report
The authors provide a description for a multi-hazard assessment in a part of Ankara. This involves flood mapping and landslide susceptibility mapping. While the flooding part is described elsewhere, this paper deals with the landslide susceptibility and the combination of the two hazard maps.
However, neither approach is particularly innovative. The landslide susceptibility map is done using a well-documented approach, so there is no new research to present. Moreover, there is hardly anything that would suggest why a particular set of parameters is being used, and the set of landslides is extremely small. This makes me question the validity of the approach.
The combination of flooding and landslide uses simple if-then combinations, which are not particularly interesting either. In addition, there are many questions for the reader, such as why there is flood hazard in the middle of a lake? Figure numbers are not consistent. Accuracy assessment is questionable.
I am afraid, in the current state, I must reject publication.
More detailed comments are on the attached pdf file.
Comments for author File: 
Comments.pdf
Author Response
Dear Reviewer,
Thank you very much for your time and efforts. Your contributions are extremely valuable and we think that the quality of our manuscript has been improved significantly by taking your suggestions into account.
Please see our answers to your comments in the attached file. We hope that the revisions we made in the manuscript may satisfy you.
Kind regards,
Authors
Author Response File: 
Author Response.docx
Reviewer 2 Report
The paper presents a study that aims at providing a « multi-hazard level » map by combining a landslide susceptibility map and a flood susceptibility map, by means expert-derived rules implemented in a Mandami fuzzy inference system. The landslide susceptibility map is construct by means of a logistic regression model with LULC, DTM-derived features and lithography as dependent variables. The flood susceptibility map was produced by a previous work.
The fuzzy inference system is considered as the main originality of this work.
Overall appreciation
The application issue seems of primary importance and the overall methodology described Figure 2 seems adequate to attempt the objective. The fuzzy inference system seems particularly interesting to implement expert knowledge in a multi-hazard evaluation context.
However, detailed implementation and evaluation of each methodology step seem questionable. In fact, the method does not seem to apply any fuzzy principle and this is perhaps the main problem of this work. Some aspects of the paper structure and redaction can be improved too.
Major comments
Materials ans Methods
Size of the grid cells: the spatial resolution of the logistic regression input layers (size of grid cells) is not clear: such resolution is set to 5m (line 220), but lithology and LULC layers have a spatial resolution of 10m.
Is the DTM spatial resolution is set to 5 m to obtain primary and secondary derivatives of a 10m-equivalent resolution?
LULC map: line 276 : the authors should give more details on the training set defined to train the random forest algorithm. Were the examples defined by photo-interpretation by the authors or from an existing, previously produced, LULC map? How many examples were defined for each LULC class? Etc.
LR application:
what is the spatial unit considered in the model: the manually delineated polygons (with the computation of a summary statistics of the dependent variables – mean, median, … - on these polygons)? Or the 10x10m cells? I think the second solution is the one applied by the authors. In this case, how many positive samples were used? How to be sure that the negative examples are actual negative ones? was any “mask” defined to select the negative examples in areas that are actually not susceptible to undergo landslide? In fact: the authors insist on the fact that it is very difficult to identify landslides in urban areas (consequently there could be a significant number of landslides that are not mentioned by the manual delineation presented Fig. 1, and the random selection of negative example could fall in one of them) ; the results of figure 42 show that there is an important part of the study area with a high landslide risk (are the “negative” examples that fall in high landslide risk areas adequate for the model construction?) ; Are the LR results very sensitive to the random selection of the negative examples? In fact, the number of negative examples (equal to the one of the positive ones, considered as “low” by the authors themselves in the conclusion section) is very low in comparison to the extent of the area without landslides. Consequently, there is little chance that the set of negative samples be representative of the study area … did the authors repeated the modeling step with different sets of randomly chosen negative samples to investigate model sensitivity to the selection of the negative samples? The lithology map used as an input layer of the logistic regression (Fig. 10) provides information with a significant lower resolution than of the other input layers. Re-sample this layer with a 10m grid permit the LR to be applied, but what can be the impact on the LR results (smoothing effect, etc.)?
Mandami fuzzy method: In fact, the method does not seem to apply any fuzzy principle and this is perhaps the main problem of this work. Modalities (Low, Moderate, High), corresponding to landslide and flood susceptibility levels and to MHL, seem to be equal interval crisp modalities. The authors mention a defuzzification method (centre of gravity) but the result of the MHL map is presented with (crisp ??) modalities, whereas the defuzzification aims at providing numeric results. Moreover, the T-norm and T-conorm for, respectively, fuzzy deduction and rule conclusion aggregation, are not mentioned.
If I am wrong, the authors should clarify this, and provide a Figure comparable to the Figure 8 in the paper of Akgun et al, 2012, specifying the membership functions …
Results
LULC evaluation: the authors should give the result of the evaluation of the Sentinel-2 classification (classification precision, kappa).
LR evaluation:
was a cross-validation performed? As the authors do not mention it this suggests that the ROC curve was obtained by using the training sets (positive and negative examples), which greatly overestimates the predictive power of the classifier; The ROC curves in Fig. 13 is a multi-class (number of classes >2) version of the ROC curve whereas the authors aim at evaluating a binary classifier; The ROC curve seems to highlight a kind of “threshold effect” (“stair step” behavior), the true positive rate passing from 0 to 1 by slightly changing the value of the classification threshold. What is the classification threshold corresponding to this change in true positive rate? Is this phenomenon due to the small area associated to landslides in comparison with the study area? Concerning the definition of the landslide susceptibility modalities (Low, Moderate, High), I think it should take into account the threshold mentioned in the precedent comment rather than being based on the equal intervals principle (?) in order to avoid risk under or over-estimation (?) The manually delineated polygons presented in Fig. 1 could be added in the Fig. 42 to verify that they are included in the high susceptibility modality.
MHL results: see previous comments.
Discussion
Discussion section is quite poor and provide additional results. It should discuss the methodological points, the implementation choices, as well as the results validity.
The responses to the previous comments should enriched this section.
Paper structure
- Lines 82 to 96: significant part of this paragraph could be moved to Method section
- Lines 309 to 313 : this was already said previously and this could suggest that the authors of the present did this work.
- Lines 325 – 327: could be moved to Background section
- Paragraph “Urban Transformation Plan” could be moved to the study area description. He last sentence can be moved to the Conclusion section.
- Lines 416 – 418 : should be moved to the Discussion section.
Minor comments
Lines 51, 83, … is “actual” is for “up-to-date” ? “Actual” do not seem appropriate.
Line 42: parenthesis to be closed
Line 86 : to my knowledge there is no L2B pre-processing level for Sentinel-2 images. Do the authors refer to the L2A level?
Line 53: Sentence to be rewritten.
More generally, English should be revised.
Figure numbers in the text and captions do not correspond.
Author Response
Dear Reviewer,
Thank you very much for your time and efforts. Your contributions are extremely valuable and we think that the quality of our manuscript has been improved significantly by taking your suggestions into account.
Please see our answers to your comments in the attached file. We hope that the revisions we made in the manuscript may satisfy you.
Kind regards,
Authors
Author Response File: Author Response.docx
Reviewer 3 Report
This submission traces a paper "Use of Mamdani Fuzzy Algorithm for Multi-hazard Assessment in a Developing Urban Settlement (Mamak, Ankara, Turkey)". The main objective of this study was to introduce for production of a multi-hazard map for a settlement area employing a Mamdani fuzzy inference algorithm.
The article is interesting and well structured. If we disregard the stated quality of research, I would just like to propose several suggestions to the author(s) of the article.
Firstly, as a research paper, this submission needs to critically assess work previously carried out in the scientific field. Although this has been done to a limited extent in the introduction and background section, some key points are missed. Perhaps the most significant articles in the research field, similar worldwide researches, etc. It would be useful to describe the aim of this paper. Besides, please justify convincingly why this manuscript (method, thematology, etc) connected with ISPRS International Journal of Geo-Information’s content and scope.
A better presentation of your results and an extensive discussion would improve your paper.
Perhaps the greatest disadvantage in the work is that the article lacks novelty and applied methodologies. Moreover, it presents serious flaws in all the sections:
- poor literature review;
- confusing and well-known methodology;
- the results of manuscript research are interesting but their discussion seems to be a little bit rushed. I propose to the authors to be more specific, explanatory and simplified to be easily understandable from the readers.
- poor discussion section: only general statements without links to results from this specific research
- I cannot find any general conclusion for this work. In my view, this approach is too "site-dependent". the authors should discuss the reason why they consider their approach more feasible than others in the literature, comparing, in particular, their methodology to the other methods for scientific research.
Author Response
Dear Reviewer,
Thank you very much for your time and efforts. Your contributions are extremely valuable and we think that the quality of our manuscript has been improved significantly by taking your suggestions into account.
Please see our answers to your comments in the attached file. We hope that the revisions we made in the manuscript may satisfy you.
Kind regards,
Authors
Author Response File: Author Response.docx
Reviewer 4 Report
Major comments
1- The approach proposed in this paper is a good example of simplified solution to compare different hazards susceptibilities, therefore it can be really useful in long-term projects for hazard prevention and risk mitigation. Nevertheless, the research presents numerous inaccuracies, lacks and superficialities. First of all, the terms ‘risk’, ‘hazard’ and ‘susceptibility’ have been used incorrectly and often the ‘hazard’ and ‘risk’ or ‘hazard’ and ‘susceptibility’ have been confused (i.e. 117-118, 188-189, 326-327,…..). In my opinion these mistakes are not acceptable in publications focused in susceptibility zoning. Moreover, as consequence, the use of a hazard map or susceptibility map if confused maybe induce in different interpretation by the operator in the spatial planning. A good description of the different use of those terms are reported in (Fell et al. 2008). Moreover the error is present in the title too, the use of the expression Multi-hazard assessment has been confused with the expression multi-hazard susceptibility mapping. It is not correct use ‘hazard’ because it reveals an analysis which consider, not only the spatial prediction of the event, but also the magnitude and the frequency (time) of the event. The analysis taken has been aimed in susceptibility mapping with no references in time and magnitude (intensity). Anyway, If I misinterpreted, no information about magnitude and frequency have been reported in the text. In both cases, I think, the lack is not acceptable.
2 - The second comment regards the validation procedure. Unfortunately a susceptibility map cannot be presented without a validation which is often evaluated using the ROC curves and dividing the dataset into training and validating. The validation and the interpretation of the success rate curve and of the prediction rate curve were not present in the paper, only the accuracy of the method has been considered. Moreover, the use of a statistical method LR for landslide susceptibility mapping require a significant number of landslides, I guess, only 8 landslides cannot support the final result. Finally, the use of statistical methods induce errors related to the statistical nature of the analysis, therefore using two consequential statistical methods RF and LR, the final result is affected by the sum of the respective errors which should be evaluated. Please consider this point.
3 - The second comment regards the combination of the landslide and flood susceptibility maps developed by two different methods, the first based on a statistical approach and the second based on experts consideration. They are two approaches completely different and therefore I think not comparable but in absence of data it maybe acceptable. Please highlight this point in the text and explain the reason why you choose the LR. In every susceptibility analysis a critical point is the classification of the final index into classes of susceptibility (low…...high). Please explain if you used the susceptibility index of the map published by Sozer et al. 2018 or if you reclassified the flood susceptibility map from 5 to 3 classes using only the final map.
4 – In general the paper describe a very long analysis which is composed by numerous steps. For this reason each step has been not described in detail but they are necessary. Please go in detail, in particular in Materials and Methods section.
Minor comments
keywords: Please, verify the number of keywords according to the journal requirements
cite: verify the bibliography style
76-77: This affirmation is not completely sure, please explain
84-91: The paragraph reports info about the data collection, it should be moved to Materials and Methods
93: Please, add detail and move to material and methods. Please add also details in the resampling procedure
123-124: maybe you are referring to vulnerability as exposure and damages (losses). Please explain.
221: Please add more information about data, report some metadata i.e. the pixel size and the toos and functions used for resampling procedure.
222: Please, report more information about the images used for landslide detection. The time reference of the images are very relevant and probably changing the time lapse of the satellite images, more landslide maybe detected.
248-253: Please take care about the use of numerous different causes to evaluate the susceptibility given from the DTM due to redundancy.
Suggestion to read
Fell, Robin, Jordi Corominas, Christophe Bonnard, Leonardo Cascini, Eric Leroi, and William Z. Savage. “Guidelines for Landslide Susceptibility, Hazard and Risk Zoning for Land Use Planning.” Engineering Geology 102, no. 3–4 (2008): 85–98. https://doi.org/10.1016/j.enggeo.2008.03.022.
Author Response
Dear Reviewer,
Thank you very much for your time and efforts. Your contributions are extremely valuable and we think that the quality of our manuscript has been improved significantly by taking your suggestions into account.
Please see our answers to your comments in the attached file. We hope that the revisions we made in the manuscript may satisfy you.
Kind regards,
Authors
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Dear authors
I truly appreciate your efforts in improving the paper, and addressing most of my concerns. However, I still find the number of landslides (8) extremely low, and therefore a landslide susceptibility analysis based on this to be highly unreliable. I also believe that the term Mamdani fuzzy algorithm is misleading, as it is still only a sequence of If - THEN statements without any fuzzyfication, since there are only two possibilities each (LS and FH). While I acknowledge that the analysis could be useful, I must question the scientific merit. Perhaps the authors can find a journal that is more local to the area to which they apply their method.
I am sorry to not have a more positive review.
Author Response
Dear Reviewer,
We have tried to explain our approach further to clarify your concerns in the attached document. We hope that the clarifications we made would eliminate your concerns about our manuscript.
Kind regards,
Authors
Author Response File: Author Response.docx
Reviewer 3 Report
I've appreciated the efforts made by the authors to follow my suggestions. In my view, the paper has been improved. The authors' response to reviewers is convincing.
For these reasons, I suggest accepting the paper in its present form.
Author Response
The file is uploaded.
Author Response File: Author Response.docx
Reviewer 4 Report
The first revision increase significantly the quality of the paper submitted, anyway it still presents some points to be revised.
You can find new few comments in the document attached. Thanks
Comments for author File: Comments.docx
Author Response
The file is uploaded.
Author Response File: Author Response.docx
