An Efficient In-Situ Debris Flow Monitoring System over a Wireless Accelerometer Network

Round 1

Reviewer 1 Report

Brief summary:

In the manuscript titled “An Efficient In-Suit Debris Flow Monitoring System over Wireless Accelerometer Network“ the authors present a novel monitoring system for debris flow occurrences base on in-situ accelerometers measurements                .

 

Broad comments:

The manuscript present an interesting method to detect the occurrence of debris flows within an instrumented channel. Along all manuscript English language needs to be revised. In figures the fonts need to be checked since they seem to be too small.

Paper organization: I would suggest to the authors to reorganize the manuscript according with the journal guidelines that foreseen the following scheme: Introduction, Materials and Methods, Results, Discussion, Conclusions.

Since the authors want to present a novel monitoring system for debris flows detection, in the Discussion section it would be useful to present how responsive is the detection system (i.e. the resulting time-lag between the event occurrence and its detection).

Follow some general comments divided according to the paragraphs of the manuscript:

1 Introduction
The presentation and characterization of debris flow phenomena must be extended. Please refer also to at least one classification. Moreover, This section lacks a properly developed “state of the art” section: previous  detection techniques must be presented giving to the reader an idea of the actual scenario in this field.

 

2 Experiment field description
A geological setting should be presented together with the characterization of previously occurred debris flows.

 

Detailed comments are given in the attached pdf file.

 

Comments for author File: Comments.pdf

Author Response

We appreciate both the interest and the constructive criticism you presented. 

We have addressed the major concerns of the comments. More specifically we have made the following modifications: 

Q1. I would suggest to the authors to reorganize the manuscript according to with the journal guidelines that foreseen the following scheme: Introduction, Materials and Methods, Results, Discussion, Conclusions.

Answer 1. Thank you for the suggestion on this issue. To comply with the journal guidelines, we adjusted the structure of paragraphs as the case you pointed out. So this issue had been solved now.

Q2. Since the authors want to present a novel monitoring system for debris flows detection, in the Discussion section it would be useful to present how responsive is the detection system (i.e. the resulting time-lag between the event occurrence and its detection).

Answer 2. Thank you for pointing out this critical problem. Yes, we totally agree with the comments and now added the contents in Sec. 5 Discussion. Concretely, we present the detail evaluation results regarding the processing time efficiency and confirmed that (near) real-time debris flow monitoring can be achieved. Also, we present ideas of future works, such as adding new sensors to further improve monitoring performance and robustness to environmental noise; in addition, Moreover, it can be anticipated that as we collect more accelerometer monitoring data the system performance can be further improved.

Q3. Introduction
The presentation and characterization of debris flow phenomena must be extended. Please refer also to at least one classification. Moreover, This section lacks a properly developed “state of the art” section: previous detection techniques must be presented giving to the reader an idea of the actual scenario in this field.

Answer 3. Thank you very much for your constructive comments. We have added important background information of debris flow classification in the beginning part. Also, the "state of the art" information of the computerized debris flow monitoring systems had been added. We covered both hardware monitoring platform review (sensors, and data transmission systems) and machine learning algorithms survey for the task in the Sec. 1  Introduction. We hope the current version can show an overview to the reader that machine learning system development for debris flow disaster monitoring became an active research theme.

 

Q4. Experiment field description
A geological setting should be presented together with the characterization of previously occurred debris flows.

Answer 4. Thank you for the comments. We have rewritten the related section in manuscript, i.e. in Sec. 2. The key information is that [A particular characteristic of Sakurajima’s debris flows is that the presence of large volumes of accumulated volcanic ash on the steep mountain slopes results in many debris-flow disasters even when only a small amount of rainfall occurs. ]. Hope this version can ease the understanding of readers.

Finally, thank you so much for carefully check the whole manuscript. Thank you for the great support to us to improve the quality of the manuscript. We have taken all the comments and made revision accordingly. Hope this updated version can be presented in the Remote Sensing journal.

Best regards,

 

Jiaxing YE, PhD

The National Institute of Advanced Industrial Science and Technology (AIST), Japan

1-1-1 Umezono, Tsukuba, Ibaraki 305-8568 Japan

Reviewer 2 Report

Paper must be corrected by a native English reader.

Accelerometers should be housed and mounted separately from the solar panel/RF antenna to avoid wind perturbations.

The results are not very convincing IMHO, since we can see from the Fig 10 that wind+rain sensors by themselves are pretty good indicators of possible flows, maybe better than the accelerometer network.

Author Response

We appreciate your great efforts in checking the manuscript and the constructive criticism you presented.

We have addressed the major concerns of the comments. More specifically we have made the following modifications:

Q1. Accelerometers should be housed and mounted separately from the solar panel/RF antenna to avoid wind perturbations.

Answer 1. Thank you for presenting the idea to reduce vibration signal interference. Yes, that is true that in our sensing unit design the accelerometer can be easily affected by wind perturbations. We didn't fully optimize the design based on an assumption that the wind-induced vibrations are inherently different from debris flow-induced vibrations in terms of tri-axial signal presentation, and we expect that the advanced data-driven machine learning algorithms, such as deep learning can eliminate the wind noise effect at debris flow identification stage. According to the results, we can see that actually there is no false alarm generated by a strong wind. Again, thank you very much for the suggestion. We will keep the idea in mind when we can update the design and make a new series of sensing units.

Q2. The results are not very convincing IMHO, since we can see from the Fig 10 that wind+rain sensors by themselves are pretty good indicators of possible flows, maybe better than the accelerometer network.

Answer 2. Thank you for the comment. Fig. 10 presents some strong correlations between rain/wind and debris flow strikes. However, as we see the charts carefully, we can find that heavy rain and strong winds often hit together; furthermore, we find that rain falls are necessary for inducing a debris flow, but not sufficient, such as in the case on Jun. 10th. By applying a threshold on rain/wind data, there will be too many false alarms. Such fact suggests that rain/wind gauge data cannot provide adequate information for debris flow prediction and we need to incorporate more channel of information to achieve the task.

Finally, thank you for the great support to us to improve the quality of the manuscript. We have taken all the comments and made revision accordingly. Hope this updated version can be presented in the Remote Sensing journal.

Best regards,

Jiaxing YE, PhD

The National Institute of Advanced Industrial Science and Technology (AIST), Japan

1-1-1 Umezono, Tsukuba, Ibaraki 305-8568 Japan

Reviewer 3 Report

First person perspective is not a good way to write an academic article, there are many grammar errors found. What is the required time for processing your data from this system? Is all the handling are automatically? What is the procedure to distinguish noise from traffic or earthquake shaking? How to measure the magnitude of debris flow from this system? Those are factors that need to be resolved.   

Author Response

We appreciate your great efforts in checking the manuscript and the constructive criticism you presented.

We have addressed the major concerns of the comments. More specifically we have made the following modifications:

Q1. there are many grammar errors found. 

Answer 1. Thank you for pointing out this issue. We have done all our best to correct the typos and grammar errors in the manuscript throughout the past 10 days. 

Q2. What is the required time for processing your data from this system? 

Answer 2. Thank you very much for your constructive comments. 

As you pointed out, the data transmission and analysis processing efficiency is a critical issue for disaster monitoring because the provision of timely and effective debris flow progress information is crucial to avoid or reduce the damages. Therefore, we added a new discussion section to present processing time evaluation results. We evaluated the proposed algorithms on two computers with different specification. As a result, both hardware platforms can complete the analysis quite fast and the result manifests that near real-time debris monitoring can be achieved. 

Q3. Is all the handling are automatically? 

Answer 3. Yes, the data capture, transmission, and analysis are all automatically.

Q4. What is the procedure to distinguish noise from traffic or earthquake shaking?

Answer 4. As shown in Fig. 2, the study area is located halfway up the mountain and therefore there is no traffic noise. In our current data collection, there is no Earthquake signal being captured. While we believe the debris flow vibration pattern could be discerned from that of earthquake because the earthquakes are a commonly impulsive signal with wide variations and short period. On the contrary, debris flow-induced vibration can be more "continuous". We hope that the discriminant information between those two types of the signal can be characterized by the statistical features presented in Tab. 1. We will keep an eye on this matter and hope we can show the debris flow/earthquake pattern classification results in our future works.

Q5. How to measure the magnitude of debris flow from this system? 

Thank you for the comment. Unfortunately, the current system can only output binary judgment representing debris flow occurrence or not. We agree that the magnitude of debris flow is quite important information, while the performance may be not satisfying if we use accelerometer data to predict magnitude of debris flow. In our plan, this theme could be tackled by adding new sensors, such as ultrasonic water depth sensor. We believe those direct measurement would greatly contribute to debris flow magnitude estimation. 

Finally, thank you for the great support to us to improve the quality of the manuscript. We have taken all the comments and made revision accordingly. Hope this updated version can be presented in the Remote Sensing journal.

Best regards,

Jiaxing YE, PhD

The National Institute of Advanced Industrial Science and Technology (AIST), Japan

1-1-1 Umezono, Tsukuba, Ibaraki 305-8568 Japan

Round 2

Reviewer 2 Report

No further comments.

Author Response

We appreciate your valuable comments regarding the manuscript and the constructive criticism you presented.

We have addressed your major concerns and we have made modifications accordingly as follows:

Q1. The length of data is 5 minutes, and take how long to process data into information then?

Answer 1. Thank you for the comment regarding the data process efficiency.  We have evaluated the proposed algorithm by using a high-end desktop PC with i9-7900K CPU with 128GB memory and a mid-range laptop PC with i7-7500U CPU with 24GB memory. Both hardware specifications can complete the data analysis process within a minute. That means before the following 5-minute data clip is captured, the analysis can be done and thus the in-situ monitoring can be achieved.

Q2. Is data from all the sensors been processed by the same system or by paralleled machine?

Answer 2: Thank you for the key comment regarding practical implementation. In our hardware system design, one data collection hub computer is connected to 5 sensing nodes via wireless communication. We have validated the stabilizability of sensing network and it worked fine even under extreme weather conditions.

Q3. For an effective monitoring system,  the accuracy ratio has to be high and low for a false alarm. Most of all, the time consumption has to be as less as possible and this is not addressed within the manuscript.  

Answer 3. Yes, we agree that the evaluation metric is quite critical and both recall rate and precision should be investigated for detection problem. In this study, we actually presented two results, which are shown in Fig. 11 and Fig. 12, respectively. Fig. 11 showed that all three debris flow occurrences can be successfully detected and that is the event-based detection result. Moreover, our system can generate a detection indication for every 5-minute, and thus we further examine detection performance on each 5-minute data segments. The most efficient metric to evaluate detection algorithm is the Receiver operating characteristic (ROC) curve, which exhibits both the recall rate (True positive rate, shown on Fig. 12 vertical-axis) and false alarm rate (Fig. 12 horizontal-axis) simultaneously. An optimal classification system is anticipated to generate bigger area under the ROC curve (often simply referred to AUC). And from Fig. 12, it is obvious that the proposed method (Spectrogram + CNN, in blue) achieved the biggest area under ROC curve and thus proposed monitoring data analysis algorithm outperformed all other methods. 

Finally, thank you again for 2nd round review and all your comments helped to us to improve the quality of the manuscript. Hope this updated version can make all remained issues clear. 

Best regards,

Jiaxing YE, PhD

The National Institute of Advanced Industrial Science and Technology (AIST), Japan

1-1-1 Umezono, Tsukuba, Ibaraki 305-8568 Japan

Reviewer 3 Report

The length of data is 5 minutes, and take how long to process data into information then? Is data from all the sensors been processed by the same system or by paralleled machine? For an effective monitoring system,  the accuracy ratio has to be high and low for a false alarm. Most of all, the time consumption has to be as less as possible and this is not addressed within the manuscript.  

Author Response

We appreciate your valuable comments regarding the manuscript and the constructive criticism you presented.

We have addressed your major concerns and we have made modifications accordingly as follows:

Q1. The length of data is 5 minutes, and take how long to process data into information then?

Answer 1. Thank you for the comment regarding the data process efficiency.We have evaluated the proposed algorithm by using a high-end desktop PC with i9-7900K CPU with 128GB memory and a mid-range laptop PC with i7-7500U CPU with 24GB memory. Both hardware specifications can complete the data analysis process within a minute. That means before the following 5-minute data clip is captured, the analysis can be done and thus the in-situ monitoring can be achieved.

Q2. Is data from all the sensors been processed by the same system or by paralleled machine?

Answer 2: Thank you for the key comment regarding practical implementation. In our hardware system design, one data collection hub computer is connected to 5 sensing nodes via wireless communication. We have validated the stabilizability of sensing network and it worked fine even under extreme weather conditions.

Q3. For an effective monitoring system, the accuracy ratio has to be high and low for a false alarm. Most of all, the time consumption has to be as less as possible and this is not addressed within the manuscript.

Answer 3. Yes, we agree that the evaluation metric is quite critical and both recall rate and precision should be investigated for detection problem. In this study, we actually presented two results, which are shown in Fig. 11 and Fig. 12, respectively. Fig. 11 showed that all three debris flow occurrences can be successfully detected and that is the event-based detection result. Moreover, our system can generate a detection indication for every 5-minute, and thus we further examine detection performance on each 5-minute data segments. The most efficient metric to evaluate detection algorithm is the Receiver operating characteristic (ROC) curve, which exhibits both the recall rate (True positive rate, shown on Fig. 12 vertical-axis) and false alarm rate (Fig. 12 horizontal-axis) simultaneously. An optimal classification system is anticipated to generate bigger area under the ROC curve (often simply referred to AUC). And from Fig. 12, it is obvious that the proposed method (Spectrogram + CNN, in blue) achieved the biggest area under ROC curve and thus proposed monitoring data analysis algorithm outperformed all other methods.

Finally, thank you again for 2nd round review and all your comments helped to us to improve the quality of the manuscript. Hope this updated version can make all remained issues clear. 

Best regards,

Jiaxing YE, PhD

The National Institute of Advanced Industrial Science and Technology (AIST), Japan

1-1-1 Umezono, Tsukuba, Ibaraki 305-8568 Japan