Urban Flood Mapping Using SAR Intensity and Interferometric Coherence via Bayesian Network Fusion

Round 1

Reviewer 1 Report

Review of Ms. Ref. No.: Manuscript ID: remotesensing-592965

Type of manuscript: Article

Title: Urban flood mapping using SAR intensity and interferometric coherence via Bayesian network fusion

Authors: Yu Li, Sandro Martinis, Marc Wieland, Stefan Schlaffer, Ryo Natsuaki

 

Overview

In the present manuscript authors describe a Bayesian approach to merge two variables provided by a remote sensing source in order to better estimate the extension of urban flooded areas. The proposed method was tested over two study areas by using data provided by two different sensors.

The manuscript is quite well introduced, described and discussed. Nevertheless, some clarifications, discussions and results should be added for a better understanding of outcomes and evaluation about the feasibility of an operational implementation within the context of emergency actions to be taken in case of flood (see comments below for details).

In my opinion, the manuscript is acceptable for publication, but with minor revisions.

 

Broad comments

(1) The feasibility of an operational implementation of the proposed method should be more deeply investigated and discussed. The current version of the manuscript does not clearly stress this issue.

Is the proposed method applicable in real time for flood management purposes considering the acquisition times of SAR observations? In the paper, authors refer to “near real-time”: which is the time interval between flood occurrence, observation of sensors and transmission, receiving and  acquisition of data? Are these times suitable to the dynamics of a flood event in a urban area and the consequent actions, emergency operational responses of authorities in charge of public safety?

(2) Some objective criteria (i.e., statistical scores) are proposed throughout the paper to evaluate the performance of the proposed method. Nevertheless, these scores were not introduced by a brief description (e.g., formula and range of theoretical values ranging from the worst to the best). This issue hampers to have a full comprehension of the performance of the proposed method and the meaning of each scores. Moreover, a term of comparison should be added to the statistical analysis in order to better evaluate the improvements of the Bayesian approach (for instance, by showing scores for flood maps based on just one variable, such as intensity or coherence).

The addition of the abovementioned proposals would help to better stress the usefulness of the proposed probabilistic approach and its added value with respect to a flood mapping based on just one variable.

(3) Authors should add a brief comment about the content of Section 4.3. Do the cited limitations hamper an operational implementation of the proposed method?

 

Specific comments

Lines 18-20: May this issue be also related to a longer time (thus, more compatible with SAR acquisition times) for standing floodwater in flooded rural areas than in flooded urban areas?

 

Lines 27-30: The citation of values for the statistical scores (without a prior description) could be too specific for the abstract. Please consider to replace the score values with a general description of performances.

 

Lines 36-37: Can the selected study areas (i.e., Houston and Joso) be considered coastal cities? Otherwise, the citation is not proper and should be removed.

 

Lines 44-45: Could some references about the cited missions be added?

 

Line 45: Could authors provide a quantification for the “revisit periods”?

 

Line 45: Could authors provide a quantification for “near real-time”?

 

Line 51: Authors should briefly recall the meaning/description of “supervised” and “unsupervised” methods.

 

Line 84: Authors should briefly recall the meaning/description of “temporal baseline” and “spatial baseline”.

 

Lines 118-120: The discussion about this issue should be enlarged based on general comment (1). 

 

Line 124: “However, it cannot be directly applied in complex urban environments”. Authors should briefly describe the reasons for this issue.

 

Lines 126-127: “We extended their work to include flood detection in urban areas and made it automatic so that sensor and study area independent”. Authors should briefly introduce and clarify how they performed the extension of the cited past work.

 

Lines 128-129: This sentence is not clear. What do author mean with “incorporate the contexture information from a long-range point of view”?

 

Line 145: Should “x₁” be “x_i”?

 

Lines 161-168: In the introduction, authors cited just one study (i.e., reference [35] at line 122) about the combination of intensity and coherence to provide information about flooded areas. Are there in the specific literature further past studies about this issue? Or is this approach a novelty?

 

Lines 231-232: “data with large temporal baseline are used”. Could authors provide a quantification for the “large temporal baseline”?

 

Lines 250-252: “The fully-connected CRF is adopted to refine the flood probability by integrating the long-range spatial information”. This process should be briefly described. 

 

Lines 261-262: The extension of the study area (km²) should be here specified. Could authors provide additional information about the period of standing floodwater, response time of this urban catchment to a rainfall event?

 

Lines 271-272: “the coherent (built-up) area thresholding was set as ?=0.5”. How was this setting chosen? Is it based on a common use in the specific literature? Please add a clarification.  

 

Lines 277-278: The extension of the study area (km²) should be here specified. Could authors provide additional information about the period of standing floodwater, response time of this urban catchment to a rainfall event?

 

Lines 313-314: The cited scores should be introduced by a brief description (i.e., formula and range of theoretical values ranging from the worst to the best).

 

Line 324: Should “RBG” be “RGB”?

 

Lines 349-350: Do the cited scores refer to Fig. 3a? Was the computation of scores in Table 5 based on data included in the area displayed in Fig.2? Please specify.

 

Lines 355-356: Could authors provide an explanation about the reasons for such a result?

 

Line 361: Remove “.” in “Figure. 3b”.

 

Line 368: How was the mean computed? Does it refer to a spatial average? Were the cited mean values computed over data included in each landscape displayed in Fig.4? Please specify.

 

Line 397: Could authors provide a quantification for the “medium resolution data”?

 

Lines 425-426: The zoom-in of yellow boxes should be provided also for panels (f) and (g).

 

Lines 431-432: Could authors provide an explanation about the reasons for such a result?

 

Lines 465-468: The discussion about this issue should be enlarged based on general comment (1).

 

Lines 495-496: The discussion about this issue should be enlarged based on general comment (1).

 

Lines 501-503: The discussion about this issue should be enlarged based on general comment (2).

 

Line 510: A brief description about benefits which are expected by the new missions should be added.

Author Response

Dear Reviewer 1,

We have revised the manuscript according to your comments and provide our responses in a point-by-point manner. Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The manuscript "Urban flood mapping using SAR intensity and interferometric coherence via Bayesian network fusion" deals with a very interesting subject of great interest to both the SAR remote sensing and hazards and risk management communities. The reading of the paper itself is attractive and the methods are well applied. However, there are small parts of the paper that requires better explanation and changes for clarification and allowing reproducible research. Therefore, the manuscript deserves publication after minor changes. 

Below some concerns.

L30: For instance, the abstract could conclude with the "lessons learned" and the new contributions of this interesting research (into a broader context);

L34: the introduction should highlight (when possible) the advantages of using the Bayesian approach rather than intensity backscattering only; 

L45: please spell out sensors names;

L149: from both the intensity backscattering and interferometric coherence;

L138: please explain in which environment this can be implemented;

L245: explain which conditions would be recommended (in discussion section too);

L260: please provide geographic coordinates;

L262: which Sentinel-1, 1A or 1B? or both? please explain;

L263: considering dual-polarization acquisition, why only considering VV?

L266 please provide more info about parameters and sources;

L267: explain better linear to dB (which equation?) and size;

L269: resulting in pixel with which dimension?

L270: geocoded with or without a DEM?

L276: please provide geographic coordinates;

L278: write down "seven ALOS-2/PALSAR-2";

L279: which acquisition mode? single, dual-polarization? why choosing HH only? please consider same concerns above;

L295: consider explaining importance of orbits;

L338: please consider adding an empty image (letter D) containing just arrows indicating the proper location you gave details in the text;

Fig4: please consider adding full dataset and either an arrow or texture indicating moisture (flood event and also some precipitation records prior data acquisition);

Fig7: check values in your figures and cross-checked it with those presented in Table 3 for example (some of them are different); also check comments above comment;

L471: please consider in this section the Sentinel-1 mission and the importance of having constellation with more satellites; make an overview of forthcoming missions;

L504: check typo;

L510: these missions should appear in discussion sections first;

L511: same above;

L513: make conclusions in a broader context and applicability for low-income countries with the current available SAR datasets;

Author Response

Dear Reviewer 2,

We have revised the manuscript according to your comments and provide our responses in a point-by-point manner. Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors present an urban flood mapping methodology (as an extension to previous work by the same team) based on Baysian network fusion using SAR coherence and intensity.

The paper is well written and the methods are sound. Using both SAR coherence and intensity for urban flood mapping seems promising and indeed would be a game-changer if operationally applicable.

There are a couple of points/concerns that should be addressed before publication:

In the introduction, the authors should try and put the method of choice in context with other self-learning algorithms. They should then justify why the Baysian network fusion is a better choice; The authors should remove the word "complicated" in association with urban since the urban landscapes they present are not really complicated in terms of urban environments; I think it would be really helpful in all the flood result figures to add a sub-figure with just a base map or high resolutionsatellite image, and then also add that same base map to all the sub-figures showing the flood maps. This will help a lot in visualizing where and under what land cover flooding took place;   

Author Response

Dear Reviewer 3,

We have revised the manuscript according to your comments and provide our responses in a point-by-point manner. Please see the attachment.

Author Response File: Author Response.pdf