Large-Scale Land Subsidence Monitoring and Prediction Based on SBAS-InSAR Technology with Time-Series Sentinel-1A Satellite Data
and Shan Zhao 2,*
Round 1
Reviewer 1 Report (Previous Reviewer 2)
The paper introduces a work of subsidence monitoring based on InSAR and LSTM. But InSAR and LSTM are mature methods, and lots of people used the methods for subsidence monitoring. So, please give the readers more information about what novel work authors did for improving the performance.
In recent years, there have been many articles about subsidence prediction based on LSTM and InSAR. such as: 1)HILL, P., BIGGS, J., PONCE-LOPEZ, V., et al. Time-Series Prediction Approaches to Forecasting Deformation in Sentinel-1 InSAR Data[J]. Journal of geophysical research. Solid earth: JGR,2021,126(3). DOI:10.1029/2020JB020176. 2) CHEN, YI, HE, YI, ZHANG, LIFENG, et al. Prediction of InSAR deformation time-series using a long short-term memory neural network[J]. 2021,42(17/18):6919-6942. DOI:10.1080/01431161.2021.1947540. 3) Li Ruren, Sun Jiayao. Monitoring and prediction of tailings pond settlement based on integration of SBAS-InSAR and GS-LSTM[J]. Metal Mine,2023(1):102-109. DOI:10.19614/j.cnki.jsks.202301011. 4) Chen Yi, He Yi, Zhang Lifeng, et al. Long short-term memory network TS-InSAR surface deformation prediction[J]. Remote Sensing Journal,2022,26(7):1326-1341. 5) Ansari H , Russwurm M , Ali S M , et al. InSAR Displacement Time Series Mining: A Machine Learning Approach[C]// IGARSS 2021. 2021. 6)Liu Q , Zhang Y , Wei J , et al. HLSTM: Heterogeneous Long Short-Term Memory Network for Large-Scale InSAR Ground Subsidence Prediction[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 14. This article use one of LSTM to process InSAR data and predict subsidence, I don't think this article has valuable improvements for this field, and I don't think this article is valuable for remote sensing too. So, I reject this article.
There are too many syntax errors and typos. Please revise this paper carefully.
such as:
1) Abbreviations only need to appear once.
2）Some words are too difficult to understand, for example, line 56, line 138, line151, line 161, line 234, etc.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report (New Reviewer)
1、In section 2.1, it is necessary to refer to the specific time period when the average precipitation next year is between 407 millimeters and 1295 millimeters.
2、Figure 2 shows that the interval time between some images of track 113 is too small, and the interval time between some images is too long. Does the calculated land subsidence time series affect the model results?
3、Why choose volumetric soil water data and monthly precipitation data as environmental factors as input variables for the model?
4、The InSAR technology monitors the subsidence generated by the entire formation, and the depth of four-layer volumetric soil aquifers is only 289 cm. Can it represent the influencing factors of the total deformation variables? What is the meaning of the elevation given in Table 2?
5、The resolution of ERA5 is too much lower than that of Sentinel-1 data, and there is a problem of scale effect. How to solve the problem of low spatial resolution of ERA5?
6、It can provide a detailed description of the SAR data fusion process, demonstrate the magnitude of differences in different track results, and how to use the differences in overlapping region results for correction.
7、How to set parameters in SAR data processing, such as coherence threshold.
8、Both LSTM and TCN are not methods proposed by the author and do not need to spend too much text to describe them, so streamlining is recommended.
9、In section 4.2, “The sum of the cumulative spring-summer deformation variables of the subsidence area land in 2020 and 2021 was divided by two as the average annual spring-summer deformation rate of the subsidence area land, and the sum of the cumulative autumn-winter deformation variables was divided by two as the average annual autumn-winter deformation rate of the subsidence area land.” Should the unit calculated in this way be mm/half a year? But in the following text, it is mm/a.
10、What is the correlation between environmental factors and subsidence data when analyzing seasonal characteristics in combination with environmental factor data and deformation monitoring point data in section 4.2? Table 3 shows that the difference between subsidence and volumetric soil layer water content in spring and summer, autumn and winter is not significant, only precipitation has a significant difference. What is the basis for the direct impact of volumetric soil layer water content on land subsidence?
11、In section 4.3, the accuracy of time series models for different months was tested, and the months were taken from the starting month. What if the accuracy was taken from the middle time? And is it representative to only select points with subsidence rates greater than 10mm/a in the InSAR monitoring results for model testing? Why does the settlement legend in Figure 11 start from 0 and since only points with a subsidence rate greater than 10mm/a were used, why is the conclusion drawn in the later text that LSTM-TCN has better prediction performance in both uplift and subsidence area?
12、Section 4.3 selected four subsidence points in the subsidence centers. What is the basis for this selection, and what are the subsidence levels of the four points?
13、The geographical locations of the four points in Figure 14 are not very clear, and it is recommended to display them in one image. The color scale in the (c)、(d) shows that the positions of C and D are not located at the subsidence center.
There are also errors in terminology in the article, so it is recommended to double-check the entire article.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report (New Reviewer)
A brief summary
This paper addresses an important issue, subsidence monitoring using Sentinel-1 data and state-of the art algorithms. I am impressed with the quality of the article, it has high merit and excellent illustrations. The article is well written and addresses a wide audience, but needs some improvement.
General concept comments
1. Please, rewrite the abstract to make it clearer (too much abbreviations).
2.Please develop the introduction further, give a general background of what has been done so far, as it is difficult to determine what the article adds to science. Also show and highlight the research gap. Add more references to other studies on this issue. Maybe a deeper literature review would help? It is worth taking a broader view that these algorithms such as XGBoost also work well for other applications eg. Bartold, M.; Kluczek, M. 2023. A Machine Learning Approach for Mapping Chlorophyll Fluorescence at Inland Wetlands. (https://doi.org/10.3390/rs15092392) or SantangeloExploring event landslide mapping using Sentinel-1 SAR backscatter products.
3. Also work on the discussion, show what results other researchers have achieved and what you have achieved and why they differ.
4. Describe what software you used, what programming libraries you used, etc.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Round 2
Reviewer 1 Report (Previous Reviewer 2)
1, Some relevant references (which also use one of LSTM to predict subsidence) need to be cited and explain the differences with these works.
2, Compare the performance with other LSTM.
1, Some sentences or words are Chinglish and need to be revised, as line 56-59, line 60-63, line 80-84.etc.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 2 Report (New Reviewer)
Land subsidence occurs mainly due to deep groundwater extraction, while only volumetric soil water data were used as model inputs in the manuscript. It is recommended to supplement deep groundwater data.
It is recommended to check the grammar and words of the whole manuscript.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
The manuscript entitled “Large-scale land subsidence monitoring and prediction based on SBAS-InSAR technology with time-series Sentinel-1A satellite data” revealed and predicted land subsidence in Henan Province using the Sentinel-1A dataset, which is quite meaningful. However, many issues in the manuscript were not well-explained, and maybe it still requires major revisions. The English expression used in this manuscript may be beneficial to have a native English speaker review and edit the text for clarity and accuracy. The main concerns:
1. Analysis of seasonal deformation of land subsidence is insufficient. Only one point P in Zhengzhou could not represent large-scale deformation characteristics. The authors chose Zhengzhou City to present seasonal change, and whether other cities experienced similar seasonal change. If so, the illustration should be added. Because in Section 6, conclusion (2) said “the land in Henan Province showed an upward trend in summer and a downward trend in winter”, which seems to be obtained only based on the analysis of Zhengzhou city.
2. The data acquisition situation of different frames of Sentinel-1 data should be simply listed. How were the images selected, because 364 images should be a part of all the available data?
3. The basic theory of SBAS in Section 3.1 can be shortened, but the explanation of how to mosaic different frames is not clear enough. How should we understand multiple reference points in the same track? Does the frame in the middle have different reference points? What is the reference for subsidence after fusion in different tracks? The prediction is based on time series data, how to make the time series consistent across different frames? Sampling each month couldn’t make them fall in the same time epoch.
4. The illustration of data processing in the last paragraph of Section 3.1 couldn’t be understood well combined with Figure 2. The relationship between the paragraph and figure is weak. How to identify atmospheric noise here? The results shown in Figures 8 and 9 seem to still include the atmospheric term. Is it reasonable for a city to experience such significant land deformation within just one month?
5. The fonts used in Figures 2 and 4 are too small, making them difficult to read, and the same issue exists with the legends in other figures. The colorbars used in all of the velocity maps and cumulative deformation maps should be improved to better illustrate the distribution of deformation. For example, in Figure 7, it is unclear what the yellow color represents, which may confuse readers. In Figure 6, the map indicates an uplift in Anyang, Hebi, Shangqiu, Xinxiang, and Puyang, it also shows subsidence in Anyang and Zhoukou. However, the majority of the map is filled with green, which also represents subsidence. Thus, according to Figure 6, it is difficult to obtain the results illustrated in In Lines 336-337 Page 11, Shangqiu and Xinxiang underwent obvious deformation. The green color with -40 mm/year subsidence also represents subsidence. Additionally, it would be helpful to keep the colorbars consistent across Figures 7 and 11, since they both represent the same data.
6. The conclusion (1) indicated that Shangqiu experienced the most severe land subsidence, while no supporting information about it was mentioned in the text.
7. The properties, such as depth, characteristics, and elevations, of different soil layers in Henan province could be listed in a figure or table, if possible.
Minor questions:
1. Lines 62-63 Page 2: the sequence of these techniques should be D-InSAR, PS-InSAR, and SBAS-InSAR, based on their development. In addition, D-InSAR is not a time-series InSAR technique, and SBAS-InSAR has no full name even if it has been explained in the abstract.
2. Lines 83-84 Page 2: “LSTM has been extensively……prediction problems” could be expressed as “LSTM is specifically designed for sequence prediction problems and has been extensively used in land subsidence prediction.”
3. Line 86 Page 2: “certain requirements on the sequence length and distribution of data” is not clear. Does it mean the length should be not too long and not too short?
4. Line 117 Page 3: “the spatial resolution is 10m” is not proper, and its resolution has a difference in the range and azimuth directions.
5. Line 328 Page 10: “According to Figure 7” should be “Figure 5”.
6. Please make sure the expression of the unit of deformation velocity (mm/year mm/a) be consistent throughout the manuscript.
7. Lines 440-442 Page 15: there is a difference between the selected feature points and total records for prediction models. How to select feature points?
8. Line 454 Page 15: LT4 model has three dilated convolution layers.
9. Line 459 Page 15: Here what about the parameters of the three models? It is unclear.
10. Lines 466-467 Page 17: Table 2 has no information regarding land subsidence in the ranges of 0~100 mm and 100mm~200mm.
11. Line 503 Page 21: “Four models”? which are they separately?
Reviewer 2 Report
1, The paper introduces a work of subsidence monitoring based on SABS-InSAR and LSTM-TCN. But, SABS-InSAR and LSTM-TCN are existing methods. So, please give the readers more information about what authors doing for improving the methods.
2. There are too many syntax errors and typos. Please revise this paper carefully.
