A Deep Learning Application for Deformation Prediction from Ground-Based InSAR
Round 1
Reviewer 1 Report
Dear Authors,
In general, the article was well written, but a review should be carried out on the document in order to improve its presentation, the description of the methods used, results achieved and conclusions.
See below a few questions related to the document:
1 - Page 7 - last paragraph
Which satellite image was used? What is the date of image collection?
2 - Page 8 - Fig. 4
Include data from the image that was used, such as image acquisition date, satellite, etc.
3 - Page 12 - Fig. 20
Same comments as in Fig. 4.
Author Response
Dear Reviewer:Thank you for the comments concerning our manuscript entitled “GBInSAR time series processing method based on LSTM model” (ID: remotesensing-1871938). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Paper we revised with revision mode. The main corrections in the paper and the responds to the reviewer’s comments are as flowing:
- Comment: Which satellite image was used? What is the date of image collection in last paragraph on page 7.
Response: This is a version of the previous paper, my negligence forgot to revise, this sentence “The satellite schematic diagram of the location of the study area” should be deleted. Figure 4 is the image obtained by the drone. I have added the image acquisition time in the new version of the manuscript.
- Comment: Include data from the image that was used, such as image acquisition date, satellite, etc.
Respose: Yes, your comment is right. I have added information about the figure in the new manuscript.
- Comment: Fig. 20 has the same comments as Fig. 4.
Respose: Yes, your comment is right. I have added information about all the figures in the new manuscript.
Special thanks to you for your good comments.
Author Response File: 
Author Response.docx
Reviewer 2 Report
The application of LSTM to GBSAR could be an interesting topic, but the article has to deeply improved.
Major remarks
1) The introduction is definitely too short and lacks a suitable number of references (possibly not only of Chinese authors!), both on GBSAR and LSTM
2) The LSTM method should be better described. Fig. 1 is not enough for this purpose. You should both give an intuitive idea of the method and to provide suitable references on theory and previous applications
3) Eq. 8 reports the normalization of X, but you never say what is X!
4) If I understand correctly, Fig. 21 should demonstrate that your method is consistent with standard process, but I see only the deformation of three points. Probably you mean that they are the differences between your method and standard process, but it is not clear.
5) It is not clear how you obtain the histogram in Fig. 22, I suppose it is the differences between your method and standard process
6) You should better explain how you compare the results of LSTM method with standard GBSAR interferometry
7) Finally, when you claim that your methos is faster than standard interferometric process, it is not enough provide the processing time (4.72 min) and the technical characteristics of the PC, but you should also provide the processing time of a standard interferometric process with the same PC
Minor remarks
1) In abstract you cite LSTM without explicating the acronym
2) “However, these methods require The analysis of the entire data set is inefficient, and with the accumulation of data, the hardware requirements are also high, and the future cannot be predicted, which is not conducive to the prevention and control of geological disasters.” This sentence is grammatically incorrect
3) “at home and abroad” Probably you mean in China and abroad
4) “is the interferometric phase of the winding” winding??
Author Response
Dear Reviewer: Thank you for the comments concerning our manuscript entitled “GBInSAR time series processing method based on LSTM model” (ID: remotesensing-1871938). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. The paper we revised with revision mode. The main corrections in the paper and the responds to the reviewer’s comments are as flowing:
- Comment: The introduction is definitely too short and lacks a suitable number of references (possibly not only of Chinese authors!), both on GBSAR and LSTM.
Response: Yes, your comment is right. I have revised in the new manuscript in first paragraph.
- Comment: The LSTM method should be better described. Fig. 1 is not enough for this purpose. You should both give an intuitive idea of the method and to provide suitable references on theory and previous applications.
Response: Yes, your comment is right. I've added a description of the LSTM approach in Section 2.3.
- Comment: 8 reports the normalization of X, but you never say what is X!
Response: I already stated what X means in the last paragraph of equation 8.
- Comment: If I understand correctly, Fig. 21 should demonstrate that your method is consistent with standard process, but I see only the deformation of three points. Probably you mean that they are the differences between your method and standard process, but it is not clear.
Response: Yes, your comment is correct. I didn't explain it very well. These three points are selected from three different locations of the landslide for study.
- Comment: It is not clear how you obtain the histogram in Fig. 22, I suppose it is the differences between your method and standard process.
Response: Yes, your comment is correct. I did not make it clear. I selected the building stability area in the research area to count the shape variables of this area to evaluate the data processing results. If there is no error in data processing, there should be no shape variable in this region, but there is deformation in this region, which proves that there is error in the data. I evaluate the data processing results by analyzing the standard deviation of this region.
- Comment: You should better explain how you compare the results of LSTM method with standard GBSAR interferometry.
Response: Yes, your comment is right. In the discussion section, I better explain the effect of the combination of LSTM model and GBSAR technology.
- Comment: Finally, when you claim that your methos is faster than standard interferometric process, it is not enough providing the processing time (4.72 min) and the technical characteristics of the PC, but you should also provide the processing time of a standard interferometric process with the same PC.
Response: Your point is correct. I have added in Table 5 a temporal comparison between LSTM and standard GBSAR timing processing.
- Comment: In abstract you cite LSTM without explicating the acronym.
Response: I have explained LSTM (Long short-term Memory) in the abstract.
- Comment: "However, these methods require inefficient analysis of the entire dataset and require high hardware requirements as the data accumulate. Moreover, they cannot predict the future, which is not conducive to the prevention and control of geological hazards." This sentence is grammatically incorrect.
Response: Yes, Your comment is correct. I changed the sentence to :“ However, these methods are computationally inefficient for the whole dataset and require a lot of hardware as the data accumulate. Moreover, they cannot predict the future, which is not good for the prevention and control of geological disasters.”
- Comment:“at home and abroad” Probably you mean in China and abroad.
Response: Yes, your comment is right. I changed the sentence to :“ In recent years, China and other countries have also carried out relevant research.”
- Comment: “is the interferometric phase of the winding” winding??
Response: Yes, your comment is correct. I'm sorry I made a mistake, I changed the sentence to:“ is the interferometric phase of the wrapping.”
Special thanks to you for your good comments.
Author Response File: Author Response.docx
Reviewer 3 Report
Review on “GBInSAR time series processing method based on LSTM
Model” submitted to Remote Sensing
The paper applied the LSTM model to process a time series data acquired by a GBInSAR and reported results from sets of measured data.
The manuscript is not well written, not in English grammar, but in containing many vague words and expressions. One example is the title itself. We cannot understand what the specific meaning of time series processing method, and what is LSTM if we are not, supposed, in the field of machine learning.
The objective of the paper is not well justified, in line of using LSTM. For a ground-based InSAR system, is the atmospheric phase prediction, as outline in Fig. 1, a big issue that must seek the neural network approach? Considering a small operation range of GBInSAR, the authors must give a quantitative assessment of atmospheric phase contribution to the total phase term.
When it comes to deformation prediction, what exactly do you mean “prediction”? Did you establish a kind of predictive model that is injected by the observation data from GBInSAR?
When observing the deformation, it is a common practice to determine the deformation in terms of spatial baseline and then the total deformation within a time scale. Therefore, what is the objective that you require a real-time processing?
The statement given below Fig. 21 that “the time series results obtained by this real-time processing method are consistent with the GBInSAR time series analysis results” is vague.
Does the LSTM’s success rely on the availability of PSs? What about if the scene contains no qualified PSs?
The proposed method based on LSTM is highly site-specified, as the site-limited results stand. Hence, the method does not contribute new technique to the InSAR, which is well-established and mature to deep extent . The paper seems is merely doing LSTM for the sake of LSTM itself.
Author Response
Dear Reviewer:Thank you for the comments concerning our manuscript entitled “GBInSAR time series processing method based on LSTM model” (ID: remotesensing-1871938). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval. Paper we revised with revision mode. The main corrections in the paper and the responds to the reviewer’s comments are as flowing:
- Comment：The manuscript is not well written, not in English grammar, but in containing many vague words and expressions. One example is the title itself. We cannot understand what the specific meaning of time series processing method, and what is LSTM if we are not, supposed, in the field of machine learning.
Response：Yes, your comment is correct. I have changed the title of the manuscript to “A Deep Learning Application to predict from ground-based InSAR”.
- Comment: The objective of the paper is not well justified, in line of using LSTM. For a ground-based InSAR system, is the atmospheric phase prediction, as outline in Fig. 1, a big issue that must seek the neural network approach? Considering a small operation range of GBInSAR, the authors must give a quantitative assessment of atmospheric phase contribution to the total phase term.
Response: Yes, your comment is correct. I have supplemented the experimental part of atmospheric phase prediction in the manuscript, which may better quantitatively evaluate the influence of atmospheric phase on GBSAR system.
- Comment: When it comes to deformation prediction, what exactly do you mean “prediction”? Did you establish a kind of predictive model that is injected by the observation data from GBInSAR?
Response: By prediction, I mean that the formal variables monitored by GBINSAR at three moments are input into the prediction model, and then the formal variables at the next moment are obtained.
- Comment: When observing the deformation, it is a common practice to determine the deformation in terms of spatial baseline and then the total deformation within a time scale. Therefore, what is the objective that you require a real-time processing?
Response：For PS-INSAR of ground-based SAR, when we need to get high-precision deformation results in a period of time, we need to process the data of this period of time by PS-INSAR, which takes a lot of time, but the real-time processing method can get the deformation results in a short time.
- Comment: The statement given below Fig. 21 that “the time series results obtained by this real-time processing method are consistent with the GBInSAR time series analysis results” is vague.
Response: Yes, your comment is correct. I added the GBINSAR deformation curve of three points.
- Comment: Does the LSTM’s success rely on the availability of PSs? What about if the scene contains no qualified PSs?
Response: The method proposed in this paper is mainly for PS points, so it has nothing to do with the number of PS points.
- Comment: The proposed method based on LSTM is highly site-specified, as the site-limited results stand. Hence, the method does not contribute new technique to the InSAR, which is well-established and mature to deep extent . The paper seems is merely doing LSTM for the sake of LSTM itself.
Response: The method proposed in this paper has achieved high experimental results for GBINSAR technology real-time processing, deformation prediction and atmospheric prediction, which proves that LSTM model can play a good role in GBINSAR monitoring.
Special thanks to you for your good comments.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
The authors replied all my issues, now the paper can be published. Just a very small note: Graphs in Figures 4 and 5 should be clearer if y-axis was in logarithmic scale.
