Parcel-Level Flood and Drought Detection for Insurance Using Sentinel-2A, Sentinel-1 SAR GRD and Mobile Images

Round 1

Reviewer 1 Report

please see the attachment file.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors explore methods for verifying and validating flood and drought damages to crops.  This work shows the advantage of using remote sensing when the sensors are not hindered by cloud cover and coarse resolution.  The article could be stream-lined and the focus shifted to the final solution the use of CART algorithm with SAR imagery makes the most sense.  When SAR is not available, they identify CNN-based methods are mobile photos as an alternative.   There are so many variables the imagery, the disaster, the algorithm that it is challenging to stay focused on the meaningful results.  The main results for imagery impacted by cloud cover is that X out the total number of fields could not be evaluated using that approach.  The article would be stronger if the solution – CART on SAR imagery is an option for Y of the total number of fields and/or is highly accurate when used.  The authors have a contribution to make to public service – accurately and efficiently evaluating disaster claims, that story can be told more succinctly.

 

 

Introduction – English grammar checks

Line 20

However, disasters like floods and droughts run havoc and swindle rice production from time to time.

However, disasters like floods and droughts wreak havoc and can greatly reduce rice production.

 

Line 45 I think the point of the sentence is that 10m resolution is higher than 30m.  Rewrite for clarity.

These bands have a spatial resolution of 10m which is comparatively low compared to the visible bands and NIR bands provided 46 by Landsat-8 with 30m.

These bands have a spatial resolution 10m of which is relatively high compared to the visible bands and NIR bands provided 46 by Landsat-8 with 30m.

 

Line 47 – the concept of accuracy and resolution are being confounded in this sentence.

Therefore, Sentinel-2A can achieve better accuracy for computing NDVI and NDWI compared to Landsat-8.

Therefore, Sentinel-2A provides higher resolution NDVI and NDWI information.

 

Line 57 …important data regardless of cloud cover [18].

In such a case, Synthetic Aperture Radar (SAR) is instrumental in gathering important data regardless of cloud [18].

Line 58   SAR penetrates clouds and provide observations of earth features obscured by other sensing systems.

SAR provides the facility to penetrate through clouds to observe earth features as the wavelength of bands in SAR images are longer, ranging from centimeters to meters [19].

 

Line 84   Describing how the rest of the paper is structured is not necessary.

The rest of this paper is structured as follows: Section 2 describes the datasets and algorithms that formulate our method; Section 3 details the method design; Section 4 shows the results; comparison is made along with discussion in Section 5, and lastly concludes with future remarks in Section 6.

Line 244   …, Sentinel-1 is hardly affected by clouds, which raises questions related to the necessity for using mobile images.

…., Sentinel-1 is hardly affected by the clouds, which questions the necessity of using mobile images.

Line 249    smoothen is not a word. … to smooth the image using a kernel..

Line 254    unclear sentence         with such exceptions negligible

 

Table 5 has two rows of the same data as Table 4  - a single table would suffice.

 

Line 288 – the implication is that field visits are conducted and area measured.  Given the range of dates when the disaster can occur there should be some explanation about the field methods, lag time between the event and the field visit.

The second step performs field validation of the satellite-based methods of Sentinel-2A and Sentinel-1. The area of disaster-affected land plots is measured on the fields to validate the area coverage shown by the methods.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper presents a combination of different methodologies (and input datasets) to detect flood and drought in Thailand. The authors use a combination of Sentinel-2, Sentinel-1 and mobile phone images.

The techniques are solid, the manuscript well-written, and the problem interesting.

Here are my suggestions for improvement.

Major issues / comments

1.       There is no mention of CART’s hyperparameters nor what you did to optimize them. You only mention the training and test data. Did you use e.g. an internal validation set?

2.       One issue that I consider somewhat important pertains to data privacy. Considering that the whole process was executed on the cloud, i.e., GEE, did you upload the reference data on the platform? Did you take care to strip any personal information of the farmers?  

3.       Section 3.2: why did you use thresholding when for SAR data you used supervised learning? Considering you have labeled data, one may think that e.g. a random forest or a decision tree could use NDVI, NDWI and the rest of the Sentinel-2 data as reference data to automatically identify affected areas. This needs justification in my opinion in the text.

4.       CART methodology is presented as the de facto methodology, have you tested other supervised learning approaches? I know that this was included as future work, but this is an important aspect that could be quickly tested given GEE’s abilities.

Moderate and minor remarks:

·         Line 19, “2017 / 18-2021 / 22” is not clear to me

·         Line 24 “field verification” => you mean on the ground checks from the authorities?

·         Line 54 to Line 57: Repetition of sentence

·         Line 72: “four” but you enumerate only three (S2A, SAR, photos)

·         Line 95: Out of curiosity, do they need to hand draw this? Do you not have access to the polygons? How do you verify this process?

·         In my humble opinion, Table 1 is unnecessary when submitting to the Remote Sensing MDPI journal. I think that most readers are familiar with the MSI of Sentinel-2. If you think it’s important you can always direct the reader on where to find the configuration or put it in the appendix.

·         Similarly, NDVI is a very well-known index, you may want to consider shortening its description.

·         Line 235: if the probability is zero => not near zero with a threshold?

·         Section 3.5: More info should be given about the collection of reference data. Did you use any special tools to collect the reference data? Was it done by a single person or multiple? If multiple, how did you decide on ties? Etc.

·         Line 268 to 269? How do you perform the clipping? Do you retain the borders or only inside points? Which method did you use?

·         Line 285 to 286: Why was Sentinel-1 unavailable in some cases? GEE problem?

·         Equation (3), I suppose that the number of failed image collections is known here, what is its value? (2?) I am only stating this because the definition of the equation is in the results section. If you don’t wish to include results maybe place this in the methodology section?

·         Equation (4), P1 and P2 are not defined in the text.

·         Line 309, experiments = > experiment results

·         Doesn’t Table 5 contain the results of also Table 4? Consider removing the duplicate information.

·         Table 10; I don’t understand the 99.44% for CART, isn’t it 117 / (199 – 21)? Is something wrong here?

·         Similarly for Table 11, shouldn’t it be 135 / (150 – 14) = 99.26% instead of 100%?

·         Line 405 “Figure Figure”

·         Line 405 “the nearly same” => “nearly the same”

·         For section 5.3 and the inability to guide the farmers about the angle at which to take the photo, we have worked on a project to include an augmented reality component to aid the farmers in taking geotagged photos. Perhaps such a scenario could also help you.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

This manuscript structure is a little bit confusing and needs more clarification on methodology. Please find the attached pdf for more comments. 

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

I appreciate the changes to the abstract and discussion.  It appears edits were only made where specific comments or proposed edits were suggested. A thorough revision of the English would have improved the readability of the manuscript.  

Reviewer 4 Report

The updated manuscript is much improved on the previous versions. The authors also address all the reviewer's comments well.