A Scenario-based Case Study: AI to analyse casualties from landslides in Chittagong Metropolitan Area, Bangladesh

Sufi, Fahim; Alam, Edris; Islam, Abu

doi:https://doi.org/10.5194/nhess-2022-90

Preprints

https://doi.org/10.5194/nhess-2022-90

Preprints

01 Apr 2022

| 01 Apr 2022

Status: this discussion paper is a preprint. It has been under review for the journal Natural Hazards and Earth System Sciences (NHESS). The manuscript was not accepted for further review after discussion.

A Scenario-based Case Study: AI to analyse casualties from landslides in Chittagong Metropolitan Area, Bangladesh

Fahim Sufi, Edris Alam, and Abu Islam

Abstract. Understanding the complex dynamics of landslides is crucial for disaster planners to make timely and effective decision that saves lives and reduces the economic impact on society. Using the landslide inventory of Chittagong Metropolitan Area (CMA), we created a new Artificial Intelligence (AI) based insight system for the town planners and senior disaster recovery strategists of Chittagong, Bangladesh. Our system generates dynamic AI-based insights for a range of complex scenarios created from 7 different landslide feature attributes. The users of our system can select a particular kind of scenario out of the exhaustive list of 1.054X10⁴¹ possible scenario sets and our AI-based system will immediately predict how many casualties are likely to occur based on the selected kind of scenario. Moreover, an AI-based system shows how landslide attributes (e.g., rainfall, area of mass, elevation, etc.) correlate with landslide casualty by drawing detailed trend lines performing both linear and logistic regressions. According to literature and the best of our knowledge, our CMA scenario-based AI insight system is the first of its kind providing the most comprehensive understanding of landslide scenarios and associated deaths and damages in CMA. The system was deployed on a wide range of platforms including Android, iOS, and Windows systems so that it could be easily adapted to strategic disaster planners. The deployed solutions were handed down to 12 landslide strategists and disaster planners for evaluations whereby 91.67 % of users found the solution easy to use, effective and self-explanatory while using via mobile.

Received: 09 Mar 2022 – Discussion started: 01 Apr 2022

Download & links

Fahim Sufi, Edris Alam, and Abu Islam

Status: closed

AC1: 'Comment on nhess-2022-90', Edris Alam, 05 Apr 2022

An AI-based system shows how landslide attributes (e.g., rainfall, area of mass, elevation, etc.) correlate with landslide casualty by drawing detailed trend lines performing both linear and logistic regressions

Citation: https://doi.org/10.5194/nhess-2022-90-AC1
CC1:
'Comment on nhess-2022-90', Fatima Zohra, 09 Apr 2022
It looks like a very interesting paper. Specially, I concur with the motivation for this paper which is to allow the strategic decision makers on the run to make evidence based instant decisions using their mobile devices. Provided the fact that most of the AI based insight generation research focused primarily on traditional desktop platforms, I can see the usability of the presented system on mobile Apps based solution. This brings me to the following enquiries:
Are there any other mobile App and AI based landslide analysis systems in existing literature?

While traditional desktop platforms could handle resource intensive computational demands by AI algorithms, how does miniature mobile devices perform on executing AI algorithms, described within this paper?

Figure 11 to Figure 14 shows several mobile deployed solutions. Are these web applications (i.e., cloud-based solutions being executed through mobile browser) or deployed mobile apps in iOS/ Android?
Citation: https://doi.org/10.5194/nhess-2022-90-CC1
- CC3: 'Reply on CC1', Fahim Sufi, 11 Apr 2022
  
  We are grateful that you found our paper interesting. You are absolutely right about the motivation of this paper. Traditionally, the decision makers had to rely on data scientists to analyze data and to generate AI driven insights (which is time consuming). However, our unique solution allows the mobile phone to automatically select the right AI algorithms to execute on the right set of scenario-driven data. The insights are presented in the decision makers mobile phone in a natural language (meaning data scientists are not required to interpret the AI driven insights).
  
  This research would enable decision makers to harness the power of AI on their mobile devices. Hence, using the AI driven insights, the decision makers can make informed and timely decisions.
  
  I appreciated all your valid queries. Following are responses to your three queries:
  
  1) There are very few studies on mobile App based disaster management (e.g., landslide, tornado, flood etc.). However, these studies mainly focused on using GPS capability of mobile for data collection. Following is an example:
  
  • Sujeet Kumar Sharmaa, Santosh K. Misrab, Jang Bahadur Singha, "The role of GIS-enabled mobile applications in disaster management: A case analysis of cyclone Gaja in India", International Journal of Information Management, Vol. 51, No. 102030, 2020
  
  However, the existing studies reported in literature didn't use AI capability on mobile Apps for generating AI based insights. In our most recent studies, we have reported AI based insights on Mobile Apps. Following are some examples our related publications:
  
  • F. K. Sufi and M. Alsulami, "Knowledge Discovery of Global Landslides Using Automated Machine Learning Algorithms," in IEEE Access, vol. 9, pp. 131400-131419, 2021, doi: 10.1109/ACCESS.2021.3115043.
  
  • F. K. Sufi and I. Khalil, "Automated Disaster Monitoring From Social Media Posts Using AI-Based Location Intelligence and Sentiment Analysis," in IEEE Transactions on Computational Social Systems, doi: 10.1109/TCSS.2022.3157142.
  
  In this paper, we used local landslide data of CMA, Bangladesh and made the AI driven insights available to senior decision makers for the very first time to measure the usability of such system.
  2) Computational loads are shared between the Mobile Apps (called Microsoft Power BI Mobile App) and Microsoft Cloud (called Microsoft Power BI Service). The seamless integration between Microsoft Power BI Mobile App and Microsoft Power BI Service provide good experience to the senior decision maker on their mobile app. With increase in computational load, the fully scalable Microsoft Power BI Service (i.e., Microsoft Cloud) dynamically assign more computational resources to provide faster AI driven sights to the decision makers. Further technical details on Microsoft Power BI Service is located at https://docs.microsoft.com/en-us/power-bi/fundamentals/service-basic-concepts.
  3) Figure 11 to Figure 14 demonstrated the usability of the proposed system via deployed mobile apps both in iOS and Android platform through Microsoft Power BI Mobile App. In other words, these are not web applications running through the mobile web browsers.
  
  Microsoft Power BI Mobile App is available in all mobile platforms like windows, iOS, and Android within their respective app stores. More details on Microsoft Power BI Mobile App could be found at https://docs.microsoft.com/en-us/power-bi/consumer/mobile/mobile-apps-for-mobile-devices.
  
  Citation: https://doi.org/10.5194/nhess-2022-90-CC3
CC2: 'Comment on nhess-2022-90', Abu Reza Md. Towfiqul Islam, 10 Apr 2022

Its an intersting topic in the field of natural hazard. A scenario-based study will help us to make a quick decision in disaster management. Both linear and logistic regressions have used to get clear picture in landslide inventories.

Citation: https://doi.org/10.5194/nhess-2022-90-CC2
CC4:
'Comment on nhess-2022-90', Disaster Development, 13 Apr 2022

The techniques applied in developing this AI-based landslide analysis are unique and innovative. Does the presented system support Lift-n-Shift or Plug-n-Play mechanism to the different datasets to obtain AI-driven insights of different locations? What would be the process if someone wants to use the same system on landslide data in another location (e.g., Loess Plateau in China)?

Can I use the same techniques for other disasters like earthquakes, tornados, Flood, etc.?

It would be very useful if the authors can discuss this issue within a discussion or any other relevant sections of the paper.

Citation: https://doi.org/10.5194/nhess-2022-90-CC4
- CC5: 'Reply on CC4', Fahim Sufi, 25 Apr 2022
  
  Many thanks for your interest in our paper. The methods described in this paper can be applied to other disaster databases (i.e., for any disaster types, for any location). As seen from Figure 6 of this paper, the process starts with obtaining the required data from a data store (i.e., Chittagong Metropolitan Area Landslide Data). After obtaining the data from the data store, Transformation, Decomposition Analysis, Regression Analysis are performed for generating the aggregated results.
  
  This data store could be replaced with any other data stores, be it of other disaster types (e.g., earthquake, flood, cyclone, bushfire etc.), or of other locations of landslide incidence (e.g., e.g., Loess Plateau in China). When Transformation, Decomposition Analysis, and Regression Analysis are performed on the new data sets, AI driven insights are generated for those specific records. Hence, the external validity of the proposed experimentation is very high.
  
  We will be more than happy to add these explanations in the discussion part of our updated manuscript. Again, many thanks for your interest in our paper. Please feel free to let me know if you have any suggestions or if you require any clarifications on this paper.
  
  Citation: https://doi.org/10.5194/nhess-2022-90-CC5
CC6: 'Comment on nhess-2022-90', Fahim Sufi, 07 Jun 2022

We are pleased to announce that we have published a relevant study that performs scenario-based analysis of Tornadoes in Bangladesh. It is now avaialbe at https://doi.org/10.3390/su14106303:
Sufi F, Alam E, Alsulami M. A New Decision Support System for Analyzing Factors of Tornado Related Deaths in Bangladesh. Sustainability. 2022; 14(10):6303. https://doi.org/10.3390/su14106303
In this study, we had also used machine learning techniques to discover factors responsible for Tornado related casualties in Bangladesh.

Citation: https://doi.org/10.5194/nhess-2022-90-CC6
AC2: 'Comment on nhess-2022-90', Edris Alam, 20 Jul 2022

Over 100 researchers have already read this research paper and have shown their interest in this work. It received 4 recommendations by renowned researchers in the discipline. The research interest score shown in Researchgate is 6.7 which is greater than the average Scopus index journal papers.
Link: https://www.researchgate.net/publication/359661565_A_Scenario-based_Case_Study_AI_to_analyse_casualties_from_landslides_in_Chittagong_Metropolitan_Area_Bangladesh

Citation: https://doi.org/10.5194/nhess-2022-90-AC2
RC1:
'Comment on nhess-2022-90', Anonymous Referee #1, 27 Sep 2022

This is a very interesting work on critically analyzing landslide events in Chittagong Metropolitan Area (CMA). Provided that landslides and land falls, like many other disasters, have detrimental effects on the economy, infrastructure and precious life, using modern AI-driven approaches to obtaining critical insights into decision-making is a significant research contribution. Particularly, I liked the fact that the solution described in this study provides decision-support in plain english, which might be suitable for a non-technical decision maker. Moreover, deploying the proposed solution and testing them in mobile environments (e.g., iOS, Android, Windows etc.) is innovative and significant. This paper is well written and maintains a cohesive logical flow. Hence, I am in favor of accepting this paper in its current form.

Citation: https://doi.org/10.5194/nhess-2022-90-RC1
- AC3: 'Reply on RC1', Edris Alam, 27 Sep 2022
  
  We are grateful to the reviewer for positive appraisal, and commenting on how the paper contributes to new knowledge. AI based decision making for disaster prediction and preparedness.
  
  Citation: https://doi.org/10.5194/nhess-2022-90-AC3
- CC7: 'Reply on RC1', Fahim Sufi, 28 Sep 2022
  
  Many thanks to the honorable reviewer for this encouraging remarks about our paper. We also greatly appreciated the reviewer's recommendation on accepting this paper.
  
  Citation: https://doi.org/10.5194/nhess-2022-90-CC7
RC2:
'Comment on nhess-2022-90', Anonymous Referee #2, 05 Oct 2022
First, I want to thank the editors for this review invitation, especially on this topic. This paper presents an innovative tool for data insights regarding landslide-related disasters. Before a more detailed list of comments, I would like to share a few questions regarding the research process.

If other researchers want to apply this methodological framework in a different location, what are the main characteristics defining valuable data helpful in performing meaningful insights?

What considerations were made to select the collection of feature attributes used to analyze casualties?

Do the selected KPI follow any variables contributing to exacerbating the disaster condition in past landslide-related events?
Citation: https://doi.org/10.5194/nhess-2022-90-RC2
- AC4:
  'Reply on RC2', Edris Alam, 08 Oct 2022
  Many thanks to the anonymous reviewer for finding our solution and study innovative. Indeed, this system presents a new method for autonomously extracting AI-driven insights interactively from landslide related data using Regressions and Decomposition Analysis. This innovative methodology is now being used in other areas of research like cyclones and other natural disasters as evident from the recent citations of this preprint discussion.
  We appreciated the interest of the reviewer in our approach with three highly legitimate and relevant queries. Our responses with the corresponding queries are briefed below:
  
  Query 1: If other researchers want to apply this methodological framework in a different location, what are the main characteristics defining valuable data helpful in performing meaningful insights?
  It is possible to use the methodology explained in this research to apply on landslides (or even other disasters like Cyclone or Tornado) that happened in other locations. For example, the process of using the same methodology in Tornado related casualty is explained in our following recent publication:
  Fahim Sufi, Edris Alam, Musleh Alsulami, “A New Decision Support System for Analyzing Factors of Tornado Related Deaths in Bangladesh”, Sustainability, Vol. 14, No. 10, 2022 (Impact Factor 3.889).
  
  Similarly, the same method applied in critically analyzing Australian cyclones is explained in our following recent publication:
  Fahim Sufi, Edris Alam, Musleh Alsulami, “Automated Analysis of Australian Tropical Cyclones with Regression, Clustering and Convolutional Neural Network”, Sustainability, Vol. 14, No. 16, 2022 (Impact Factor 3.889).
  
  Moreover, this method could also be used to monitor disasters from any global locations as demonstrated in following publication:
  Fahim Sufi and Ibrahim Khalil, "Automated Disaster Monitoring From Social Media Posts Using AI-Based Location Intelligence and Sentiment Analysis," in IEEE Transactions on Computational Social Systems, doi: 10.1109/TCSS.2022.3157142, 2022 (https://ieeexplore.ieee.org/document/9737676, Impact Factor 4.747)
  
  As it becomes apparent from these recent publications, the dataset is first required to be cleansed and transformed. Then, the Microsoft Power BI's Key Influencer visualization is used to analyze the outcome variable (e.g., Casualty) with respect to a list of available "explain by" variables (e.g., Elevation, Rainfall, Area of Mass, Longitude, Latitude, Number of Injuries, Style, Types etc.). The detailed process in using Microsoft Power BI's Key influencer visualization is explained at https://learn.microsoft.com/en-us/power-bi/visuals/power-bi-visualization-influencers?tabs=powerbi-desktop.
  
  Query 2: What considerations were made to select the collection of feature attributes used to analyze casualties?
  Machine Learning (ML) based feature analysis (e.g., linear Regression or logistic Regression) depends on the availability of many feature attributes for understanding their correlations to the outcome variable. In this study, Casualty was deemed as an outcome variable, since strategic decision makers are always keen on saving precious
  
  Citation: https://doi.org/10.5194/nhess-2022-90-AC4
  - RC3: 'Reply on AC4', Anonymous Referee #2, 11 Oct 2022
    
    It seems that somehow you did not end your answer. Please, let me know if I am wrong and if it has all the content you want to post.
    Thanks for your response. I will be posting a more detailed comment.
    
    Citation: https://doi.org/10.5194/nhess-2022-90-RC3
    
    AC5: 'Reply on RC3', Edris Alam, 11 Oct 2022
    
    Many thanks to the anonymous reviewer for finding our solution and study innovative. Indeed, this system presents a new method for autonomously extracting AI-driven insights interactively from landslide related data using Regressions and Decomposition Analysis. This innovative methodology is now being used in other areas of research like cyclones and other natural disasters as evident from the recent citations of this preprint discussion.
    We appreciated the interest of the reviewer in our approach with three highly legitimate and relevant queries. Our responses with the corresponding queries are briefed below:
    Query 1: If other researchers want to apply this methodological framework in a different location, what are the main characteristics defining valuable data helpful in performing meaningful insights?
    It is possible to use the methodology explained in this research to apply on landslides (or even other disasters like Cyclone or Tornado) that happened in other locations. For example, the process of using the same methodology in Tornado related casualty is explained in our following recent publication:
    Fahim Sufi, Edris Alam, Musleh Alsulami, “A New Decision Support System for Analyzing Factors of Tornado Related Deaths in Bangladesh”, Sustainability, Vol. 14, No. 10, 2022 (Impact Factor 3.889).
    
    Similarly, the same method applied in critically analyzing Australian cyclones is explained in our following recent publication:
    Fahim Sufi, Edris Alam, Musleh Alsulami, “Automated Analysis of Australian Tropical Cyclones with Regression, Clustering and Convolutional Neural Network”, Sustainability, Vol. 14, No. 16, 2022 (Impact Factor 3.889).
    
    Moreover, this method could also be used to monitor disasters from any global locations as demonstrated in following publication:
    Fahim Sufi and Ibrahim Khalil, "Automated Disaster Monitoring From Social Media Posts Using AI-Based Location Intelligence and Sentiment Analysis," in IEEE Transactions on Computational Social Systems, doi: 10.1109/TCSS.2022.3157142, 2022 (https://ieeexplore.ieee.org/document/9737676, Impact Factor 4.747)
    
    As it becomes apparent from these recent publications, the dataset is first required to be cleansed and transformed. Then, the Microsoft Power BI's Key Influencer visualization is used to analyze the outcome variable (e.g., Casualty) with respect to a list of available "explain by" variables (e.g., Elevation, Rainfall, Area of Mass, Longitude, Latitude, Number of Injuries, Style, Types etc.). The detailed process in using Microsoft Power BI's Key influencer visualization is explained at https://learn.microsoft.com/en-us/power-bi/visuals/power-bi-visualization-influencers?tabs=powerbi-desktop.
    
    Citation: https://doi.org/10.5194/nhess-2022-90-AC5
    
    AC8: 'Reply on RC3', Edris Alam, 11 Oct 2022
    
    Please find the attached responses to your queries...
    
    Citation: https://doi.org/10.5194/nhess-2022-90-AC8
- AC6: 'Reply on RC2', Edris Alam, 11 Oct 2022
  
  Query 2: What considerations were made to select the collection of feature attributes used to analyze casualties?
  Machine Learning (ML) based feature analysis (e.g., linear Regression or logistic Regression) depends on the availability of many feature attributes for understanding their correlations to the outcome variable. In this study, Casualty was deemed as an outcome variable, since strategic decision makers are always keen on saving precious lives resulting from landslides. Within our dataset, we only had few available features to analyze (e.g., Latitude, Longitude, Elevation, Area of Mass, Rainfall etc.). After applying our innovative method, our solution found a positive correlation of casualty with "Area of Mass" (as shown in Fig. 5, Row 1 of Table 2, Row 2 of Table 2, Row 3 of Table 2, Row 4 of Table 2, Row 5 of Table 2, Row 6 of Table 2, Row 7 of Table 2), Rainfall (as shown in Row 3 of Table 2, Row 4 of Table 2), and Elevation (as shown in Row 5 of Table 2, Row 6 of Table 2). Even though we utilized all the available features present within our dataset to obtain relationships with the observed variable (i.e., casualty), we considered appropriate data cleansing prior to the automated ML process. As a result of the cleansing process, Elevation and Area of Mass turned out to be the decimal type of data and Rainfall turned out to be integer data types.
  Pre-processing the available dataset with appropriate data cleansing and transformation is the key to obtaining better AI-driven insight on the casualty.
  
  Citation: https://doi.org/10.5194/nhess-2022-90-AC6
- AC7: 'Reply on RC2', Edris Alam, 11 Oct 2022
  
  Query 3: Do the selected KPI follow any variables contributing to exacerbating the disaster condition in past landslide-related events?
  This study didn’t use any KPIs to report past landslide-related casualties. The first sentence in “Section 2.5 Analysis Data with AI” mentioned that "Key Performance Indicator (KPI) visualization was used to analyze casualty...". In fact, it should be rewritten as "Microsoft Power BI's Key Influencer visualization was used to analyze casualty...". As seen from the reference (https://learn.microsoft.com/en-us/power-bi/visuals/power-bi-visualization-influencers?tabs=powerbi-desktop), Key Influencers visualization finds out all the dependent variables along with their relationships to an observed variable. In this study, this Key Influencer visualization found out that "Area of Mass", "Rainfall", and "Elevation" are the three most related feature attributes that have a direct correlation with past landslide-related Casualties.
  
  Citation: https://doi.org/10.5194/nhess-2022-90-AC7
RC4:
'FINAL Comment on nhess-2022-90', Anonymous Referee #2, 14 Oct 2022
This study presents a modern tool to help understand the complex behavior of disasters triggered by natural events focusing on landslides. The paper describes a methodological framework to identify factors contributing to high landslide casualties by finding variable correlations from a collection of data from recorded landslides on CMA. I recommend considering the following comments before publishing this paper to underline the research contribution:
Proofreading general comments:
Please re-read line 26; there may be writing errors.

Please re-read line 30; there may be writing errors.

Line 62. Make sure you settle AI and ML will be used as equivalent concepts if they are technically different.

Please re-read line 62; there may be writing errors.

Line 29. Give some examples and references of the main events of this type.

Line 33. If contributing to understanding your research aim, mention which were those causes.

What do you mean by “scenario-based” in your research? Which is the scenario of analysis?

Line 70. The authors suggest that the application of AI reduces disaster deaths. Do you mean this tool helps understand the factors over which actions can be made to reduce the risk of death?

I recommend using a consistent term from those used in your paper (deaths, casualties), as they could represent a different concept in a disaster context.

Figure 2. Make sure you are using the term vulnerability correctly or somehow define it when describing the photos on your figure. If possible, explain the expected consequences of a landslide scenario in this area shown in the figure, describe how many areas like those exist in the study region, etc.

Make sure to clarify the implications of using techniques other than AI, or why your AI technique enhances/complements other types of analyses presented in the literature.

Please clarify if this type of analyses can be conducted with any available data (as you suggested in your author response), or if minimum requirements (data collection size, completeness, etc.) exist to generate meaningful AI insights.

Please show an extract of the final data used after the preparation process.

Please discuss how your findings about the correlated variables (i.e. “Area of Mass", "Rainfall", and "Elevation”) correspond or differ from the evidence of the variable relationship in other investigations on landslides.

[Discussion section] Highlight the implications of using incomplete datasets when generating AI insights like those used in your research.

Finally, it should be interesting to mention how this tool could be used for future research on identifying the main factors influencing other loss metrics, such as injured people, for those researchers studying resources to cope with post-disaster response scenarios.
Citation: https://doi.org/10.5194/nhess-2022-90-RC4
- AC11: 'Reply on RC4', Edris Alam, 02 Nov 2022
  
  Many thanks for all the constructive feedback. All these suggestions are appropriate and we will be more than happy to address each and all of them in the updated manuscript. In response to “clarify the implications of using techniques other than AI, or why your AI technique enhances/complements other types of analyses presented in the literature.”, we can highlight that this paper only focused on automatically identifying the relationships that may exist between an outcome variable (i.e., landslide related casualty) with a range of other variables (e.g., rainfall, area of mass, Elevation, etc.). Hence, we used a particular AI-based regression tool called “Key Influencer Visualization”. There are many other AI-based as well as non-AI-based statistical techniques that may suit other research objectives. For example, to find the similarity and dissimilarities between past landslides, AI-based automated clustering techniques could be used.
  In terms of “Highlight the implications of using incomplete datasets when generating AI insights like those used in your research.”, we should highlight the fact that AI-based automated insight generation processes as depicted in this research are often referred to as data-driven insight. For data-driven insights, having a robust and complete set of data is often a mandate. In case the data suffers from irregular/missing values (or any other data quality issues hampering the overall quality of the dataset) then several pre-processing techniques (e.g., StandardScaler, MinMaxScaler, StandardScaler, OneHotEncoder, etc.) could enhance the performance of data-driven-insight solutions. The above explanation is also applicable to the honorable reviewer’s point on “Please clarify if this type of analyses can be conducted with any available data (as you suggested in your author response), or if minimum requirements (data collection size, completeness, etc.) exist to generate meaningful AI insights.”. Within the updated manuscript we would like to highlight the fact that having a more robust and comprehensive set of data assures the generation of more meaning of insights with the approach described within this paper.
  
  Citation: https://doi.org/10.5194/nhess-2022-90-AC11
RC5:
'Comment on nhess-2022-90', Anonymous Referee #3, 18 Oct 2022

This work presents a framework for a tool where insights can be derived from data on past landslide events to inform response to future scenarios. While the premise is exciting and such a tool would be useful to inform decision making, I reject this manuscript for publication as I think the underlying methods should be reconsidered. Below, I explain why I have made my decision and provide suggestions for improvement.

My main reason for rejecting the manuscript is that the data inputs do not capture many important charactertistcs of landslides that may influence how impactful, in terms of casualties, a landslide may be. Some examples of these factors include, antecedent soil moisture^1,2 and slope material properties³. In addition, in this study, the only landslide types considered were slides, falls and topples, despite the fact that other landslide types (most notably, debris flows) do occur in Bangladesh. I understand that simple factors may have been chosen for ease of use and general applicability. However, inferring the impact of a future landslide with this tool would be problematic when landslide types, which are not reflected in the training data, are triggered. Debris flows, as previously mentioned, have resulted in many casualties yet their relationship with rainfall and antecedent soil mositure is complex ^1,2. Thus, if your tool is used to estimate impacts from a debris flow users may respond in an inappropriate manner as the tool has not been trained using debris flow data and thus, cannot capture the potential impacts of a landslide of that type.

There is also no assessment of the predictive performance of the tool. As the tool has currently been developed, the results would be misleading and likely improperly used by the anticipated end-user (e.g., emergency managers).

In addition to these general comments, the manuscript lacks clarity which ultimately impacts the reproducibility of the work. One example of this is the lack of explanation for the rainfall data used. It is not clear what rainfall corresponds to in the context of this study (e.g., average, maximum), how it has been measured (e.g., rain gauge, satellite) and at what scale (e.g., specific landslide, adminstrative boundary etc.). Even if the data used is gathered from another source a summary of that data and how it was collected should still be mentioned in the text. While this is one example where clarity could be improved, there were many other instances where further clarification of data and methods used would be helpful.

To summarize, I think the topic of this work should be explored further; however, caution needs to be taken when developing a tool meant to inform those decision makers without technical expertise. In the next iteration of this work, the complex nature of landslide processes needs to be addressed somehow. The methods and data used need to be carefully documented and explained so that others can reproduce the work. The performace of the model should also be assessed and discussed. Limitations and uncertainty in the ‘insights’ being provided need to be clearly communicated to the end user.

[1] Baum, R. L., & Godt, J. W. (2010). Early warning of rainfall-induced shallow landslides and debris flows in the USA. Landslides, 7(3), 259-272.

[2] Wieczorek, G. F., & Glade, T. (2005). Climatic factors influencing occurrence of debris flows. In Debris-flow hazards and related phenomena (pp. 325-362). Springer, Berlin, Heidelberg.

[3] Medwedeff, W. G., Clark, M. K., Zekkos, D., & West, A. J. (2020). Characteristic landslide distributions: An investigation of landscape controls on landslide size. Earth and Planetary Science Letters, 539, 116203.

Citation: https://doi.org/10.5194/nhess-2022-90-RC5
- AC9: 'Reply on RC5', Edris Alam, 01 Nov 2022
  
  First of all, we would like to thank all the reviewers for their constructive comments. This would significantly improve the quality of the final paper. Most importantly, I would like to thank Review 1 and Review 2 for their valuable suggestions on accepting this paper after carefully going through the paper and understanding the merit.
  While reading through the comments of reviewer 3, it appeared that the reviewer did not read some core concepts that had been already mentioned in the manuscript. For example, the main reason for the reviewer’s decision of rejecting this paper seems to be quite odd as the reviewer mentioned “My main reason for rejecting the manuscript is that the data inputs do not capture many important characteristics of landslides that may influence how impactful, in terms of casualties, a landslide may be.” Respectfully, we argue that this paper is a case study as it is clearly stated in the title of the paper⸻A Scenario-based Case Study: AI to analyse casualties from landslides in Chittagong Metropolitan Area, Bangladesh. Because this paper is a case study, it did not consider all possible landslide variables that the reviewer identified (i.e., this is not the focus of the paper). In fact, this paper worked on a publicly available data and focused only on the available set of landslide feature within that data set. It has been clearly pointed out in paragraph 125 as “we obtained publicly available data directly from PDF file (Rahman et al., 2016) and then we transformed the data in a suitable format that allows faster analysis”.
  If we use the same technique on another dataset as a new case study (that includes the suggested landslide parameters like Debris flows, soil moisture etc.), then this methodology would autonomously discover insights from the available parameters. The same methods applied on Tornadoes, Cyclones, Earthquakes, COVID-19, Global News, Political Messages, and even NASA’s landslide data has successfully provided us with deep insights as evident from our following recent publications:
  [1] Fahim Sufi and Ibrahim Khalil, Automated Disaster Monitoring from Social Media Posts using AI based Location Intelligence and Sentiment Analysis, IEEE Transactions on Computational Social Systems, (Accepted, in Press DOI: https://doi.org/10.1109/TCSS.2022.3157142), 2022 (IF: 5.23, Q1)
  [2] Fahim Sufi, E. Alam, M. Alsulami, Automated Analysis of Australian Tropical Cyclones with Regression, Clustering and Convolutional Neural Network, Sustainability, Vol. 14, No. 16, p. 9830, DOI: https://doi.org/10.3390/su14169830, 2022 (IF: 3.889)
  [3] Fahim Sufi, Imran Razzak and Ibrahim Khalil, Tracking Anti-Vax Social Movement Using AI based Social Media Monitoring, IEEE Transactions on Technology and Society (Accepted, in Press DOI: https://doi.org/10.1109/TTS.2022.3192757), 2022
  [4] Fahim Sufi, A decision support system for extracting artificial intelligence-driven insights from live twitter feeds on natural disasters, Decision Analytics Journal (Elsevier), Vol. 5, No. 100130, DOI: https://doi.org/10.1016/j.dajour.2022.100130, 2022
  [5] Fahim Sufi, E. Alam, M. Alsulami, “A New Decision Support System for Analyzing Factors of Tornado Related Deaths in Bangladesh”, Sustainability, Vol 14, No 10, p. 6303, DOI: https://doi.org/10.3390/su14106303, 2022 (IF: 3.889)
  [6] Fahim Sufi, “AI-SocialDisaster: An AI-based software for identifying and analyzing natural disasters from social media”, Software Impacts (Elsevier), Vol 11, No 100319, 2022, DOI: https://doi.org/10.1016/j.simpa.2022.100319
  [7] Fahim Sufi, “AI-Tornado: An AI-based Software for analyzing Tornadoes from disaster event dataset”, Software Impacts, Vol. 11, No. 100357, 2022, DOI: https://doi.org/10.1016/j.simpa.2022.100357
  [8] F. Sufi and M. Alsulami, "AI-based Automated Extraction of Location-Oriented COVID-19 Sentiments," Computers, Materials & Continua (CMC), Vols. 72, no. 2, pp. 3631–3649, 2022. DOI: https://doi.org/10.32604/cmc.2022.026272 (IF: 3.772, Q1)
  [9] Fahim Sufi, Identifying the Drivers of Negative News with Sentiment, Entity and Regression Analysis, International Journal of Information Management Data Insights, Vol. 2, No. 1, 100074, 2022, DOI: https://doi.org/10.1016/j.jjimei.2022.100074
  [10] F. Sufi and M. Alsulami, "A Novel Method of Generating Geospatial Intelligence from Social Media Posts of Political Leaders," Information, vol. 13, no. 3, p. 120, https://doi.org/10.3390/info13030120, 2022.
  [11] Fahim Sufi, AI-GlobalEvents: A Software for analyzing, identifying and explaining global events with Artificial Intelligence, Software Impacts (Elsevier), Vol 11, No 100218, 2022, DOI: https://doi.org/10.1016/j.simpa.2022.100218
  [12] Fahim Sufi, AI-Landslide: Software for acquiring hidden insights from global landslide data using Artificial Intelligence, Software Impacts (Elsevier), Vol 10, No 100177, 2021, DOI: https://doi.org/10.1016/j.simpa.2021.100177
  [13] Fahim Sufi, Musleh Alsulami, Knowledge Discovery of Global Landslides Using Automated Machine Learning Algorithms, IEEE Access, Vol. 9, 2021, Available Online at https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9546772 (IF: 3.367, Q1)
  [14] Fahim Sufi and M. Alsulami, "Automated Multidimensional Analysis of Global Events with Entity Detection, Sentiment Analysis and Anomaly Detection," IEEE Access, Vol. 9, 2021, DOI: https://ieeexplore.ieee.org/document/9612169 (IF: 3.367, Q1)
  
  Citation: https://doi.org/10.5194/nhess-2022-90-AC9
- AC10: 'Reply on RC5', Edris Alam, 01 Nov 2022
  
  In another comment, the review mentioned that the manuscript lacks clarity which ultimately impacts the reproducibility of the work. We can assure that this work is fully reproducable as the source code have been already published with peer-reviewed academic software repositories:
  
  [1] Fahim Sufi, AI-Landslide: Software for acquiring hidden insights from global landslide data using Artificial Intelligence, Software Impacts (Elsevier), Vol 10, No 100177, 2021, DOI: https://doi.org/10.1016/j.simpa.2021.100177
  
  [2] Fahim Sufi, Musleh Alsulami, Knowledge Discovery of Global Landslides Using Automated Machine Learning Algorithms, IEEE Access, Vol. 9, 2021, Available Online at https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9546772 (IF: 3.367, Q1)
  
  Citation: https://doi.org/10.5194/nhess-2022-90-AC10

Status: closed

AC1: 'Comment on nhess-2022-90', Edris Alam, 05 Apr 2022

An AI-based system shows how landslide attributes (e.g., rainfall, area of mass, elevation, etc.) correlate with landslide casualty by drawing detailed trend lines performing both linear and logistic regressions

Citation: https://doi.org/10.5194/nhess-2022-90-AC1
CC1:
'Comment on nhess-2022-90', Fatima Zohra, 09 Apr 2022
It looks like a very interesting paper. Specially, I concur with the motivation for this paper which is to allow the strategic decision makers on the run to make evidence based instant decisions using their mobile devices. Provided the fact that most of the AI based insight generation research focused primarily on traditional desktop platforms, I can see the usability of the presented system on mobile Apps based solution. This brings me to the following enquiries:
Are there any other mobile App and AI based landslide analysis systems in existing literature?

While traditional desktop platforms could handle resource intensive computational demands by AI algorithms, how does miniature mobile devices perform on executing AI algorithms, described within this paper?

Figure 11 to Figure 14 shows several mobile deployed solutions. Are these web applications (i.e., cloud-based solutions being executed through mobile browser) or deployed mobile apps in iOS/ Android?
Citation: https://doi.org/10.5194/nhess-2022-90-CC1
- CC3: 'Reply on CC1', Fahim Sufi, 11 Apr 2022
  
  We are grateful that you found our paper interesting. You are absolutely right about the motivation of this paper. Traditionally, the decision makers had to rely on data scientists to analyze data and to generate AI driven insights (which is time consuming). However, our unique solution allows the mobile phone to automatically select the right AI algorithms to execute on the right set of scenario-driven data. The insights are presented in the decision makers mobile phone in a natural language (meaning data scientists are not required to interpret the AI driven insights).
  
  This research would enable decision makers to harness the power of AI on their mobile devices. Hence, using the AI driven insights, the decision makers can make informed and timely decisions.
  
  I appreciated all your valid queries. Following are responses to your three queries:
  
  1) There are very few studies on mobile App based disaster management (e.g., landslide, tornado, flood etc.). However, these studies mainly focused on using GPS capability of mobile for data collection. Following is an example:
  
  • Sujeet Kumar Sharmaa, Santosh K. Misrab, Jang Bahadur Singha, "The role of GIS-enabled mobile applications in disaster management: A case analysis of cyclone Gaja in India", International Journal of Information Management, Vol. 51, No. 102030, 2020
  
  However, the existing studies reported in literature didn't use AI capability on mobile Apps for generating AI based insights. In our most recent studies, we have reported AI based insights on Mobile Apps. Following are some examples our related publications:
  
  • F. K. Sufi and M. Alsulami, "Knowledge Discovery of Global Landslides Using Automated Machine Learning Algorithms," in IEEE Access, vol. 9, pp. 131400-131419, 2021, doi: 10.1109/ACCESS.2021.3115043.
  
  • F. K. Sufi and I. Khalil, "Automated Disaster Monitoring From Social Media Posts Using AI-Based Location Intelligence and Sentiment Analysis," in IEEE Transactions on Computational Social Systems, doi: 10.1109/TCSS.2022.3157142.
  
  In this paper, we used local landslide data of CMA, Bangladesh and made the AI driven insights available to senior decision makers for the very first time to measure the usability of such system.
  2) Computational loads are shared between the Mobile Apps (called Microsoft Power BI Mobile App) and Microsoft Cloud (called Microsoft Power BI Service). The seamless integration between Microsoft Power BI Mobile App and Microsoft Power BI Service provide good experience to the senior decision maker on their mobile app. With increase in computational load, the fully scalable Microsoft Power BI Service (i.e., Microsoft Cloud) dynamically assign more computational resources to provide faster AI driven sights to the decision makers. Further technical details on Microsoft Power BI Service is located at https://docs.microsoft.com/en-us/power-bi/fundamentals/service-basic-concepts.
  3) Figure 11 to Figure 14 demonstrated the usability of the proposed system via deployed mobile apps both in iOS and Android platform through Microsoft Power BI Mobile App. In other words, these are not web applications running through the mobile web browsers.
  
  Microsoft Power BI Mobile App is available in all mobile platforms like windows, iOS, and Android within their respective app stores. More details on Microsoft Power BI Mobile App could be found at https://docs.microsoft.com/en-us/power-bi/consumer/mobile/mobile-apps-for-mobile-devices.
  
  Citation: https://doi.org/10.5194/nhess-2022-90-CC3
CC2: 'Comment on nhess-2022-90', Abu Reza Md. Towfiqul Islam, 10 Apr 2022

Its an intersting topic in the field of natural hazard. A scenario-based study will help us to make a quick decision in disaster management. Both linear and logistic regressions have used to get clear picture in landslide inventories.

Citation: https://doi.org/10.5194/nhess-2022-90-CC2
CC4:
'Comment on nhess-2022-90', Disaster Development, 13 Apr 2022

The techniques applied in developing this AI-based landslide analysis are unique and innovative. Does the presented system support Lift-n-Shift or Plug-n-Play mechanism to the different datasets to obtain AI-driven insights of different locations? What would be the process if someone wants to use the same system on landslide data in another location (e.g., Loess Plateau in China)?

Can I use the same techniques for other disasters like earthquakes, tornados, Flood, etc.?

It would be very useful if the authors can discuss this issue within a discussion or any other relevant sections of the paper.

Citation: https://doi.org/10.5194/nhess-2022-90-CC4
- CC5: 'Reply on CC4', Fahim Sufi, 25 Apr 2022
  
  Many thanks for your interest in our paper. The methods described in this paper can be applied to other disaster databases (i.e., for any disaster types, for any location). As seen from Figure 6 of this paper, the process starts with obtaining the required data from a data store (i.e., Chittagong Metropolitan Area Landslide Data). After obtaining the data from the data store, Transformation, Decomposition Analysis, Regression Analysis are performed for generating the aggregated results.
  
  This data store could be replaced with any other data stores, be it of other disaster types (e.g., earthquake, flood, cyclone, bushfire etc.), or of other locations of landslide incidence (e.g., e.g., Loess Plateau in China). When Transformation, Decomposition Analysis, and Regression Analysis are performed on the new data sets, AI driven insights are generated for those specific records. Hence, the external validity of the proposed experimentation is very high.
  
  We will be more than happy to add these explanations in the discussion part of our updated manuscript. Again, many thanks for your interest in our paper. Please feel free to let me know if you have any suggestions or if you require any clarifications on this paper.
  
  Citation: https://doi.org/10.5194/nhess-2022-90-CC5
CC6: 'Comment on nhess-2022-90', Fahim Sufi, 07 Jun 2022

We are pleased to announce that we have published a relevant study that performs scenario-based analysis of Tornadoes in Bangladesh. It is now avaialbe at https://doi.org/10.3390/su14106303:
Sufi F, Alam E, Alsulami M. A New Decision Support System for Analyzing Factors of Tornado Related Deaths in Bangladesh. Sustainability. 2022; 14(10):6303. https://doi.org/10.3390/su14106303
In this study, we had also used machine learning techniques to discover factors responsible for Tornado related casualties in Bangladesh.

Citation: https://doi.org/10.5194/nhess-2022-90-CC6
AC2: 'Comment on nhess-2022-90', Edris Alam, 20 Jul 2022

Over 100 researchers have already read this research paper and have shown their interest in this work. It received 4 recommendations by renowned researchers in the discipline. The research interest score shown in Researchgate is 6.7 which is greater than the average Scopus index journal papers.
Link: https://www.researchgate.net/publication/359661565_A_Scenario-based_Case_Study_AI_to_analyse_casualties_from_landslides_in_Chittagong_Metropolitan_Area_Bangladesh

Citation: https://doi.org/10.5194/nhess-2022-90-AC2
RC1:
'Comment on nhess-2022-90', Anonymous Referee #1, 27 Sep 2022

This is a very interesting work on critically analyzing landslide events in Chittagong Metropolitan Area (CMA). Provided that landslides and land falls, like many other disasters, have detrimental effects on the economy, infrastructure and precious life, using modern AI-driven approaches to obtaining critical insights into decision-making is a significant research contribution. Particularly, I liked the fact that the solution described in this study provides decision-support in plain english, which might be suitable for a non-technical decision maker. Moreover, deploying the proposed solution and testing them in mobile environments (e.g., iOS, Android, Windows etc.) is innovative and significant. This paper is well written and maintains a cohesive logical flow. Hence, I am in favor of accepting this paper in its current form.

Citation: https://doi.org/10.5194/nhess-2022-90-RC1
- AC3: 'Reply on RC1', Edris Alam, 27 Sep 2022
  
  We are grateful to the reviewer for positive appraisal, and commenting on how the paper contributes to new knowledge. AI based decision making for disaster prediction and preparedness.
  
  Citation: https://doi.org/10.5194/nhess-2022-90-AC3
- CC7: 'Reply on RC1', Fahim Sufi, 28 Sep 2022
  
  Many thanks to the honorable reviewer for this encouraging remarks about our paper. We also greatly appreciated the reviewer's recommendation on accepting this paper.
  
  Citation: https://doi.org/10.5194/nhess-2022-90-CC7
RC2:
'Comment on nhess-2022-90', Anonymous Referee #2, 05 Oct 2022
First, I want to thank the editors for this review invitation, especially on this topic. This paper presents an innovative tool for data insights regarding landslide-related disasters. Before a more detailed list of comments, I would like to share a few questions regarding the research process.

If other researchers want to apply this methodological framework in a different location, what are the main characteristics defining valuable data helpful in performing meaningful insights?

What considerations were made to select the collection of feature attributes used to analyze casualties?

Do the selected KPI follow any variables contributing to exacerbating the disaster condition in past landslide-related events?
Citation: https://doi.org/10.5194/nhess-2022-90-RC2
- AC4:
  'Reply on RC2', Edris Alam, 08 Oct 2022
  Many thanks to the anonymous reviewer for finding our solution and study innovative. Indeed, this system presents a new method for autonomously extracting AI-driven insights interactively from landslide related data using Regressions and Decomposition Analysis. This innovative methodology is now being used in other areas of research like cyclones and other natural disasters as evident from the recent citations of this preprint discussion.
  We appreciated the interest of the reviewer in our approach with three highly legitimate and relevant queries. Our responses with the corresponding queries are briefed below:
  
  Query 1: If other researchers want to apply this methodological framework in a different location, what are the main characteristics defining valuable data helpful in performing meaningful insights?
  It is possible to use the methodology explained in this research to apply on landslides (or even other disasters like Cyclone or Tornado) that happened in other locations. For example, the process of using the same methodology in Tornado related casualty is explained in our following recent publication:
  Fahim Sufi, Edris Alam, Musleh Alsulami, “A New Decision Support System for Analyzing Factors of Tornado Related Deaths in Bangladesh”, Sustainability, Vol. 14, No. 10, 2022 (Impact Factor 3.889).
  
  Similarly, the same method applied in critically analyzing Australian cyclones is explained in our following recent publication:
  Fahim Sufi, Edris Alam, Musleh Alsulami, “Automated Analysis of Australian Tropical Cyclones with Regression, Clustering and Convolutional Neural Network”, Sustainability, Vol. 14, No. 16, 2022 (Impact Factor 3.889).
  
  Moreover, this method could also be used to monitor disasters from any global locations as demonstrated in following publication:
  Fahim Sufi and Ibrahim Khalil, "Automated Disaster Monitoring From Social Media Posts Using AI-Based Location Intelligence and Sentiment Analysis," in IEEE Transactions on Computational Social Systems, doi: 10.1109/TCSS.2022.3157142, 2022 (https://ieeexplore.ieee.org/document/9737676, Impact Factor 4.747)
  
  As it becomes apparent from these recent publications, the dataset is first required to be cleansed and transformed. Then, the Microsoft Power BI's Key Influencer visualization is used to analyze the outcome variable (e.g., Casualty) with respect to a list of available "explain by" variables (e.g., Elevation, Rainfall, Area of Mass, Longitude, Latitude, Number of Injuries, Style, Types etc.). The detailed process in using Microsoft Power BI's Key influencer visualization is explained at https://learn.microsoft.com/en-us/power-bi/visuals/power-bi-visualization-influencers?tabs=powerbi-desktop.
  
  Query 2: What considerations were made to select the collection of feature attributes used to analyze casualties?
  Machine Learning (ML) based feature analysis (e.g., linear Regression or logistic Regression) depends on the availability of many feature attributes for understanding their correlations to the outcome variable. In this study, Casualty was deemed as an outcome variable, since strategic decision makers are always keen on saving precious
  
  Citation: https://doi.org/10.5194/nhess-2022-90-AC4
  - RC3: 'Reply on AC4', Anonymous Referee #2, 11 Oct 2022
    
    It seems that somehow you did not end your answer. Please, let me know if I am wrong and if it has all the content you want to post.
    Thanks for your response. I will be posting a more detailed comment.
    
    Citation: https://doi.org/10.5194/nhess-2022-90-RC3
    
    AC5: 'Reply on RC3', Edris Alam, 11 Oct 2022
    
    Many thanks to the anonymous reviewer for finding our solution and study innovative. Indeed, this system presents a new method for autonomously extracting AI-driven insights interactively from landslide related data using Regressions and Decomposition Analysis. This innovative methodology is now being used in other areas of research like cyclones and other natural disasters as evident from the recent citations of this preprint discussion.
    We appreciated the interest of the reviewer in our approach with three highly legitimate and relevant queries. Our responses with the corresponding queries are briefed below:
    Query 1: If other researchers want to apply this methodological framework in a different location, what are the main characteristics defining valuable data helpful in performing meaningful insights?
    It is possible to use the methodology explained in this research to apply on landslides (or even other disasters like Cyclone or Tornado) that happened in other locations. For example, the process of using the same methodology in Tornado related casualty is explained in our following recent publication:
    Fahim Sufi, Edris Alam, Musleh Alsulami, “A New Decision Support System for Analyzing Factors of Tornado Related Deaths in Bangladesh”, Sustainability, Vol. 14, No. 10, 2022 (Impact Factor 3.889).
    
    Similarly, the same method applied in critically analyzing Australian cyclones is explained in our following recent publication:
    Fahim Sufi, Edris Alam, Musleh Alsulami, “Automated Analysis of Australian Tropical Cyclones with Regression, Clustering and Convolutional Neural Network”, Sustainability, Vol. 14, No. 16, 2022 (Impact Factor 3.889).
    
    Moreover, this method could also be used to monitor disasters from any global locations as demonstrated in following publication:
    Fahim Sufi and Ibrahim Khalil, "Automated Disaster Monitoring From Social Media Posts Using AI-Based Location Intelligence and Sentiment Analysis," in IEEE Transactions on Computational Social Systems, doi: 10.1109/TCSS.2022.3157142, 2022 (https://ieeexplore.ieee.org/document/9737676, Impact Factor 4.747)
    
    As it becomes apparent from these recent publications, the dataset is first required to be cleansed and transformed. Then, the Microsoft Power BI's Key Influencer visualization is used to analyze the outcome variable (e.g., Casualty) with respect to a list of available "explain by" variables (e.g., Elevation, Rainfall, Area of Mass, Longitude, Latitude, Number of Injuries, Style, Types etc.). The detailed process in using Microsoft Power BI's Key influencer visualization is explained at https://learn.microsoft.com/en-us/power-bi/visuals/power-bi-visualization-influencers?tabs=powerbi-desktop.
    
    Citation: https://doi.org/10.5194/nhess-2022-90-AC5
    
    AC8: 'Reply on RC3', Edris Alam, 11 Oct 2022
    
    Please find the attached responses to your queries...
    
    Citation: https://doi.org/10.5194/nhess-2022-90-AC8
- AC6: 'Reply on RC2', Edris Alam, 11 Oct 2022
  
  Query 2: What considerations were made to select the collection of feature attributes used to analyze casualties?
  Machine Learning (ML) based feature analysis (e.g., linear Regression or logistic Regression) depends on the availability of many feature attributes for understanding their correlations to the outcome variable. In this study, Casualty was deemed as an outcome variable, since strategic decision makers are always keen on saving precious lives resulting from landslides. Within our dataset, we only had few available features to analyze (e.g., Latitude, Longitude, Elevation, Area of Mass, Rainfall etc.). After applying our innovative method, our solution found a positive correlation of casualty with "Area of Mass" (as shown in Fig. 5, Row 1 of Table 2, Row 2 of Table 2, Row 3 of Table 2, Row 4 of Table 2, Row 5 of Table 2, Row 6 of Table 2, Row 7 of Table 2), Rainfall (as shown in Row 3 of Table 2, Row 4 of Table 2), and Elevation (as shown in Row 5 of Table 2, Row 6 of Table 2). Even though we utilized all the available features present within our dataset to obtain relationships with the observed variable (i.e., casualty), we considered appropriate data cleansing prior to the automated ML process. As a result of the cleansing process, Elevation and Area of Mass turned out to be the decimal type of data and Rainfall turned out to be integer data types.
  Pre-processing the available dataset with appropriate data cleansing and transformation is the key to obtaining better AI-driven insight on the casualty.
  
  Citation: https://doi.org/10.5194/nhess-2022-90-AC6
- AC7: 'Reply on RC2', Edris Alam, 11 Oct 2022
  
  Query 3: Do the selected KPI follow any variables contributing to exacerbating the disaster condition in past landslide-related events?
  This study didn’t use any KPIs to report past landslide-related casualties. The first sentence in “Section 2.5 Analysis Data with AI” mentioned that "Key Performance Indicator (KPI) visualization was used to analyze casualty...". In fact, it should be rewritten as "Microsoft Power BI's Key Influencer visualization was used to analyze casualty...". As seen from the reference (https://learn.microsoft.com/en-us/power-bi/visuals/power-bi-visualization-influencers?tabs=powerbi-desktop), Key Influencers visualization finds out all the dependent variables along with their relationships to an observed variable. In this study, this Key Influencer visualization found out that "Area of Mass", "Rainfall", and "Elevation" are the three most related feature attributes that have a direct correlation with past landslide-related Casualties.
  
  Citation: https://doi.org/10.5194/nhess-2022-90-AC7
RC4:
'FINAL Comment on nhess-2022-90', Anonymous Referee #2, 14 Oct 2022
This study presents a modern tool to help understand the complex behavior of disasters triggered by natural events focusing on landslides. The paper describes a methodological framework to identify factors contributing to high landslide casualties by finding variable correlations from a collection of data from recorded landslides on CMA. I recommend considering the following comments before publishing this paper to underline the research contribution:
Proofreading general comments:
Please re-read line 26; there may be writing errors.

Please re-read line 30; there may be writing errors.

Line 62. Make sure you settle AI and ML will be used as equivalent concepts if they are technically different.

Please re-read line 62; there may be writing errors.

Line 29. Give some examples and references of the main events of this type.

Line 33. If contributing to understanding your research aim, mention which were those causes.

What do you mean by “scenario-based” in your research? Which is the scenario of analysis?

Line 70. The authors suggest that the application of AI reduces disaster deaths. Do you mean this tool helps understand the factors over which actions can be made to reduce the risk of death?

I recommend using a consistent term from those used in your paper (deaths, casualties), as they could represent a different concept in a disaster context.

Figure 2. Make sure you are using the term vulnerability correctly or somehow define it when describing the photos on your figure. If possible, explain the expected consequences of a landslide scenario in this area shown in the figure, describe how many areas like those exist in the study region, etc.

Make sure to clarify the implications of using techniques other than AI, or why your AI technique enhances/complements other types of analyses presented in the literature.

Please clarify if this type of analyses can be conducted with any available data (as you suggested in your author response), or if minimum requirements (data collection size, completeness, etc.) exist to generate meaningful AI insights.

Please show an extract of the final data used after the preparation process.

Please discuss how your findings about the correlated variables (i.e. “Area of Mass", "Rainfall", and "Elevation”) correspond or differ from the evidence of the variable relationship in other investigations on landslides.

[Discussion section] Highlight the implications of using incomplete datasets when generating AI insights like those used in your research.

Finally, it should be interesting to mention how this tool could be used for future research on identifying the main factors influencing other loss metrics, such as injured people, for those researchers studying resources to cope with post-disaster response scenarios.
Citation: https://doi.org/10.5194/nhess-2022-90-RC4
- AC11: 'Reply on RC4', Edris Alam, 02 Nov 2022
  
  Many thanks for all the constructive feedback. All these suggestions are appropriate and we will be more than happy to address each and all of them in the updated manuscript. In response to “clarify the implications of using techniques other than AI, or why your AI technique enhances/complements other types of analyses presented in the literature.”, we can highlight that this paper only focused on automatically identifying the relationships that may exist between an outcome variable (i.e., landslide related casualty) with a range of other variables (e.g., rainfall, area of mass, Elevation, etc.). Hence, we used a particular AI-based regression tool called “Key Influencer Visualization”. There are many other AI-based as well as non-AI-based statistical techniques that may suit other research objectives. For example, to find the similarity and dissimilarities between past landslides, AI-based automated clustering techniques could be used.
  In terms of “Highlight the implications of using incomplete datasets when generating AI insights like those used in your research.”, we should highlight the fact that AI-based automated insight generation processes as depicted in this research are often referred to as data-driven insight. For data-driven insights, having a robust and complete set of data is often a mandate. In case the data suffers from irregular/missing values (or any other data quality issues hampering the overall quality of the dataset) then several pre-processing techniques (e.g., StandardScaler, MinMaxScaler, StandardScaler, OneHotEncoder, etc.) could enhance the performance of data-driven-insight solutions. The above explanation is also applicable to the honorable reviewer’s point on “Please clarify if this type of analyses can be conducted with any available data (as you suggested in your author response), or if minimum requirements (data collection size, completeness, etc.) exist to generate meaningful AI insights.”. Within the updated manuscript we would like to highlight the fact that having a more robust and comprehensive set of data assures the generation of more meaning of insights with the approach described within this paper.
  
  Citation: https://doi.org/10.5194/nhess-2022-90-AC11
RC5:
'Comment on nhess-2022-90', Anonymous Referee #3, 18 Oct 2022

This work presents a framework for a tool where insights can be derived from data on past landslide events to inform response to future scenarios. While the premise is exciting and such a tool would be useful to inform decision making, I reject this manuscript for publication as I think the underlying methods should be reconsidered. Below, I explain why I have made my decision and provide suggestions for improvement.

My main reason for rejecting the manuscript is that the data inputs do not capture many important charactertistcs of landslides that may influence how impactful, in terms of casualties, a landslide may be. Some examples of these factors include, antecedent soil moisture^1,2 and slope material properties³. In addition, in this study, the only landslide types considered were slides, falls and topples, despite the fact that other landslide types (most notably, debris flows) do occur in Bangladesh. I understand that simple factors may have been chosen for ease of use and general applicability. However, inferring the impact of a future landslide with this tool would be problematic when landslide types, which are not reflected in the training data, are triggered. Debris flows, as previously mentioned, have resulted in many casualties yet their relationship with rainfall and antecedent soil mositure is complex ^1,2. Thus, if your tool is used to estimate impacts from a debris flow users may respond in an inappropriate manner as the tool has not been trained using debris flow data and thus, cannot capture the potential impacts of a landslide of that type.

There is also no assessment of the predictive performance of the tool. As the tool has currently been developed, the results would be misleading and likely improperly used by the anticipated end-user (e.g., emergency managers).

In addition to these general comments, the manuscript lacks clarity which ultimately impacts the reproducibility of the work. One example of this is the lack of explanation for the rainfall data used. It is not clear what rainfall corresponds to in the context of this study (e.g., average, maximum), how it has been measured (e.g., rain gauge, satellite) and at what scale (e.g., specific landslide, adminstrative boundary etc.). Even if the data used is gathered from another source a summary of that data and how it was collected should still be mentioned in the text. While this is one example where clarity could be improved, there were many other instances where further clarification of data and methods used would be helpful.

To summarize, I think the topic of this work should be explored further; however, caution needs to be taken when developing a tool meant to inform those decision makers without technical expertise. In the next iteration of this work, the complex nature of landslide processes needs to be addressed somehow. The methods and data used need to be carefully documented and explained so that others can reproduce the work. The performace of the model should also be assessed and discussed. Limitations and uncertainty in the ‘insights’ being provided need to be clearly communicated to the end user.

[1] Baum, R. L., & Godt, J. W. (2010). Early warning of rainfall-induced shallow landslides and debris flows in the USA. Landslides, 7(3), 259-272.

[2] Wieczorek, G. F., & Glade, T. (2005). Climatic factors influencing occurrence of debris flows. In Debris-flow hazards and related phenomena (pp. 325-362). Springer, Berlin, Heidelberg.

[3] Medwedeff, W. G., Clark, M. K., Zekkos, D., & West, A. J. (2020). Characteristic landslide distributions: An investigation of landscape controls on landslide size. Earth and Planetary Science Letters, 539, 116203.

Citation: https://doi.org/10.5194/nhess-2022-90-RC5
- AC9: 'Reply on RC5', Edris Alam, 01 Nov 2022
  
  First of all, we would like to thank all the reviewers for their constructive comments. This would significantly improve the quality of the final paper. Most importantly, I would like to thank Review 1 and Review 2 for their valuable suggestions on accepting this paper after carefully going through the paper and understanding the merit.
  While reading through the comments of reviewer 3, it appeared that the reviewer did not read some core concepts that had been already mentioned in the manuscript. For example, the main reason for the reviewer’s decision of rejecting this paper seems to be quite odd as the reviewer mentioned “My main reason for rejecting the manuscript is that the data inputs do not capture many important characteristics of landslides that may influence how impactful, in terms of casualties, a landslide may be.” Respectfully, we argue that this paper is a case study as it is clearly stated in the title of the paper⸻A Scenario-based Case Study: AI to analyse casualties from landslides in Chittagong Metropolitan Area, Bangladesh. Because this paper is a case study, it did not consider all possible landslide variables that the reviewer identified (i.e., this is not the focus of the paper). In fact, this paper worked on a publicly available data and focused only on the available set of landslide feature within that data set. It has been clearly pointed out in paragraph 125 as “we obtained publicly available data directly from PDF file (Rahman et al., 2016) and then we transformed the data in a suitable format that allows faster analysis”.
  If we use the same technique on another dataset as a new case study (that includes the suggested landslide parameters like Debris flows, soil moisture etc.), then this methodology would autonomously discover insights from the available parameters. The same methods applied on Tornadoes, Cyclones, Earthquakes, COVID-19, Global News, Political Messages, and even NASA’s landslide data has successfully provided us with deep insights as evident from our following recent publications:
  [1] Fahim Sufi and Ibrahim Khalil, Automated Disaster Monitoring from Social Media Posts using AI based Location Intelligence and Sentiment Analysis, IEEE Transactions on Computational Social Systems, (Accepted, in Press DOI: https://doi.org/10.1109/TCSS.2022.3157142), 2022 (IF: 5.23, Q1)
  [2] Fahim Sufi, E. Alam, M. Alsulami, Automated Analysis of Australian Tropical Cyclones with Regression, Clustering and Convolutional Neural Network, Sustainability, Vol. 14, No. 16, p. 9830, DOI: https://doi.org/10.3390/su14169830, 2022 (IF: 3.889)
  [3] Fahim Sufi, Imran Razzak and Ibrahim Khalil, Tracking Anti-Vax Social Movement Using AI based Social Media Monitoring, IEEE Transactions on Technology and Society (Accepted, in Press DOI: https://doi.org/10.1109/TTS.2022.3192757), 2022
  [4] Fahim Sufi, A decision support system for extracting artificial intelligence-driven insights from live twitter feeds on natural disasters, Decision Analytics Journal (Elsevier), Vol. 5, No. 100130, DOI: https://doi.org/10.1016/j.dajour.2022.100130, 2022
  [5] Fahim Sufi, E. Alam, M. Alsulami, “A New Decision Support System for Analyzing Factors of Tornado Related Deaths in Bangladesh”, Sustainability, Vol 14, No 10, p. 6303, DOI: https://doi.org/10.3390/su14106303, 2022 (IF: 3.889)
  [6] Fahim Sufi, “AI-SocialDisaster: An AI-based software for identifying and analyzing natural disasters from social media”, Software Impacts (Elsevier), Vol 11, No 100319, 2022, DOI: https://doi.org/10.1016/j.simpa.2022.100319
  [7] Fahim Sufi, “AI-Tornado: An AI-based Software for analyzing Tornadoes from disaster event dataset”, Software Impacts, Vol. 11, No. 100357, 2022, DOI: https://doi.org/10.1016/j.simpa.2022.100357
  [8] F. Sufi and M. Alsulami, "AI-based Automated Extraction of Location-Oriented COVID-19 Sentiments," Computers, Materials & Continua (CMC), Vols. 72, no. 2, pp. 3631–3649, 2022. DOI: https://doi.org/10.32604/cmc.2022.026272 (IF: 3.772, Q1)
  [9] Fahim Sufi, Identifying the Drivers of Negative News with Sentiment, Entity and Regression Analysis, International Journal of Information Management Data Insights, Vol. 2, No. 1, 100074, 2022, DOI: https://doi.org/10.1016/j.jjimei.2022.100074
  [10] F. Sufi and M. Alsulami, "A Novel Method of Generating Geospatial Intelligence from Social Media Posts of Political Leaders," Information, vol. 13, no. 3, p. 120, https://doi.org/10.3390/info13030120, 2022.
  [11] Fahim Sufi, AI-GlobalEvents: A Software for analyzing, identifying and explaining global events with Artificial Intelligence, Software Impacts (Elsevier), Vol 11, No 100218, 2022, DOI: https://doi.org/10.1016/j.simpa.2022.100218
  [12] Fahim Sufi, AI-Landslide: Software for acquiring hidden insights from global landslide data using Artificial Intelligence, Software Impacts (Elsevier), Vol 10, No 100177, 2021, DOI: https://doi.org/10.1016/j.simpa.2021.100177
  [13] Fahim Sufi, Musleh Alsulami, Knowledge Discovery of Global Landslides Using Automated Machine Learning Algorithms, IEEE Access, Vol. 9, 2021, Available Online at https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9546772 (IF: 3.367, Q1)
  [14] Fahim Sufi and M. Alsulami, "Automated Multidimensional Analysis of Global Events with Entity Detection, Sentiment Analysis and Anomaly Detection," IEEE Access, Vol. 9, 2021, DOI: https://ieeexplore.ieee.org/document/9612169 (IF: 3.367, Q1)
  
  Citation: https://doi.org/10.5194/nhess-2022-90-AC9
- AC10: 'Reply on RC5', Edris Alam, 01 Nov 2022
  
  In another comment, the review mentioned that the manuscript lacks clarity which ultimately impacts the reproducibility of the work. We can assure that this work is fully reproducable as the source code have been already published with peer-reviewed academic software repositories:
  
  [1] Fahim Sufi, AI-Landslide: Software for acquiring hidden insights from global landslide data using Artificial Intelligence, Software Impacts (Elsevier), Vol 10, No 100177, 2021, DOI: https://doi.org/10.1016/j.simpa.2021.100177
  
  [2] Fahim Sufi, Musleh Alsulami, Knowledge Discovery of Global Landslides Using Automated Machine Learning Algorithms, IEEE Access, Vol. 9, 2021, Available Online at https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9546772 (IF: 3.367, Q1)
  
  Citation: https://doi.org/10.5194/nhess-2022-90-AC10

Fahim Sufi, Edris Alam, and Abu Islam

Viewed

Total article views: 1,603 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
1,013	527	63	1,603	17	17

HTML: 1,013
PDF: 527
XML: 63
Total: 1,603
BibTeX: 17
EndNote: 17

Views and downloads (calculated since 01 Apr 2022)

Month	HTML	PDF	XML	Total
Apr 2022	206	47	19	272
May 2022	69	17	0	86
Jun 2022	39	11	2	52
Jul 2022	38	12	2	52
Aug 2022	34	35	0	69
Sep 2022	60	21	6	87
Oct 2022	148	65	18	231
Nov 2022	74	33	6	113
Dec 2022	19	9	1	29
Jan 2023	20	13	0	33
Feb 2023	54	37	0	91
Mar 2023	19	7	0	26
Apr 2023	26	21	0	47
May 2023	16	9	0	25
Jun 2023	16	5	1	22
Jul 2023	16	47	2	65
Aug 2023	18	16	1	35
Sep 2023	55	31	0	86
Oct 2023	23	32	1	56
Nov 2023	10	9	0	19
Dec 2023	12	5	0	17
Jan 2024	8	4	0	12
Feb 2024	20	17	2	39
Mar 2024	13	24	2	39

Cumulative views and downloads (calculated since 01 Apr 2022)

Month	HTML	PDF	XML	Total
Apr 2022	206	47	19	272
May 2022	69	17	0	86
Jun 2022	39	11	2	52
Jul 2022	38	12	2	52
Aug 2022	34	35	0	69
Sep 2022	60	21	6	87
Oct 2022	148	65	18	231
Nov 2022	74	33	6	113
Dec 2022	19	9	1	29
Jan 2023	20	13	0	33
Feb 2023	54	37	0	91
Mar 2023	19	7	0	26
Apr 2023	26	21	0	47
May 2023	16	9	0	25
Jun 2023	16	5	1	22
Jul 2023	16	47	2	65
Aug 2023	18	16	1	35
Sep 2023	55	31	0	86
Oct 2023	23	32	1	56
Nov 2023	10	9	0	19
Dec 2023	12	5	0	17
Jan 2024	8	4	0	12
Feb 2024	20	17	2	39
Mar 2024	13	24	2	39

Viewed (geographical distribution)

Total article views: 1,505 (including HTML, PDF, and XML) Thereof 1,505 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 01 Apr 2024

Short summary

Using the landslide inventory of Chittagong Metropolitan Area (CMA), we created a new Artificial Intelligence (AI) based insight system for the town planners and senior disaster recovery strategists of Chittagong, Bangladesh. The users of our system can select a particular kind of scenario out of the exhaustive list of 1.054X10⁴¹ possible scenario sets and our AI-based system will immediately predict how many casualties are likely to occur based on the selected kind of scenario.


Total:	0
HTML:	0
PDF:	0
XML:	0