Mapping and Monitoring Small-Scale Mining Activities in Ghana using Sentinel-1 Time Series (2015–2019)

Round 1

Reviewer 1 Report

Dear Authors,

Major remarks

            I think that your manuscript entitled  “Mapping and Monitoring Small-Scale Mining Activities in Ghana using Sentinel-1 Time Series 4 (2015-2019)” brings new and valuable ideas and results to remote sensing and galamsey’s evaluation.

This manuscript is quite systematic and thoughtful, having interesting results, analytical part as well as  experimental one (Sentinel-1 observations).  

This work is important from both scientific and practical point of view, and is suitable for publication in the Remote Sensing journal in present state. I like this paper.

P.S. line 739: Replace ghana by Ghana, and check the text again when collaborating with Editors at final stage

                                                                                                                                                                                                                                                            Sincerely yours

Reviewer

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

Congratulations on your interesting paper, which deals with a highly relevant topic. In my opinion, your paper shall be published as is.

You elaborately discuss three key aspects of the subject matter: You derive the algorithm, you use. You investigated and analyzed the pertubations that occur in several of the used Sentinal-1 datatake of the study area. You present your mesurement results on the detection of mining activities. It is commendable that you draw conclusions not only on the technical aspects of detection accuracy but also on the temporal progression of mining activities in the study area and on the effectiveness of local provisions to prevent illegal mining.

I just have one point of critique. Once you identified several parameters in both polarizations (VV and VH) which are similarly suitable as decision criterias for the detection of mining activities, it seems to me that you stopped at half of the way when you do not take into consideration that a joint evaluation of these parameters (by techniques like e.g. support vector machines) might be an instrument to futher improve the detection accuracy. Thus, I encourage you to investicate this topic too and to present your results in a future paper.

Typos:

page 3, line 97: missing unit: "12.5 spatial resolution"

page 5, Figure 2: typo: "SRMT DEM" -> "SRTM DEM"

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

English are well written, easy to read and to follow-up along the manuscript. In my opinion the research as proposed has to be not focus on "illegal minning activities" but in ecosystem dynamism. Google Earth images phtointerpretation well good enough to document the lack of vegetation cover, no need to use the Sentinel-1 images.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 4 Report

Review:

Detection of illegal mining using radar satellite remote sensing is very interesting subject matter and as a one of the remote sensing applications, matches the scope of the journal. Authors perform analysis based on amplitude information from SLC products of Sentinel-1 mission. However, overall research is on the basis of comparison two methods of histogram thresholding between two classes “agro-forest” and “water” in reference to two polarization modes VV and VH. The huge part of the paper is analysis of atmospheric effect affecting radar images named unfortunately as “anomalies”. In my opinion, this analysis is not a subject of the presented manuscript. For sure authors have put a big effort to process the SLC products acquired from four years, however the analysis of change detection seems to be basic and well-known for the remote-sensing researchers. It would be valuable to extend the research by adding the phase analysis, not only amplitude. I mean to focus not only on coherent change detection but also on coherent change detection.

After making changes to the structure in its current form, the article could be published in a magazine addressed directly to mining or government institutions.

Comments

1. Materials and Methods section contains four subsections. Subsection “Satellite imagery” is divided into Sentinel-1 Time Series and SRTM Digital Elevation Model. There is some inconsistency. SRTM DEM is not satellite image, is a product generated based on satellite image. I recommend reconstructing the subsection “Satellite imagery” ie. create one common subsection for Materials and name simply as Data or Data input and next put there Sentinel-1 Time Series, SRTM Digital Elevation Model and Reference Data.

2. Additionally, I would move the content of subsection 3.1 to subsection 2.2.1 since subsection 3.1 presents Materials not Results. Results should provide a concise and precise description of the experimental results. See more remarks to subsection 3.1 in comment 7.

3. Authors chosen Single Look Complex products while GRD (Ground Range Detection) products are also available for the study area. GRD products comparing to SLC products are:

• smaller 4 times (SLC scene ~8GB, GRD scene ~2Gb)
• partially radiometrically corrected
• not complex (real and imaginary) data type, contain only amplitude information
• not divided into subswaths and bursts
• resampled to 10x10m pixel size (SLC are about 5x20m resolution and authors made resampling to 20x20m pixel size)
• easier/not complicated to pre-process into Sigma or Gamma Nought (much less steps to perform)

Can authors explain what was the reason to choose SLC data instead of GRD?

4. Figure 2
   • Authors mentioned that for some unique dates, it was necessary to download two consecutive scenes to cover study area (lines 144-147). There is missing function that indicate process of connecting two consecutive scenes. Please, place this step in workflow. Was it performed using tool Slice Assembly or Mosaic or other way?
   • Please provide more detailed information of speckle filtering. Was it single product or multitemporal filtering? What was the size of the filter window?
   • For S-1 data is recommended to remove thermal noise (tool S-1 Thermal Noise Removal in SNAP). Using this tool atmospheric effect (named as “artefacts or anomalies” by authors) is removing significantly, as well as noise on the borders of bursts. This is missing in the workflow. Did the authors consider removing the thermal noise?

5. Line 211

What types of vegetation are replaced to bare soil or to water surface?  The study area is covered mainly by trees (deciduous and evergreen forests) as is mentioned in line 136. Does it mean that illegal mining is placed instead of trees?

6. Sections 2.4.1-2.4.4

It seems that methodology is focused only on detecting changes from “agro-forest” to “water”. Changes to “bare soil” are omitted, which brings the whole article to detection water bodies. 

7. Section 3.1

I would strongly advise the authors to rewrite this subsection. Firstly, I strongly recommend NOT to use words “artefacts”, “anomalies”. Presented by authors atmospheric effect is TYPICALLY for Synthetic Aperture Radar data. Secondly, only simply data summary, how many images from which months were used and how many were excluded due to the atmosphere, is needed. Figure 4 in this form is not necessary. For example, j) presents strong clouds occurred during investigated period.

8. Lines 321-323

I think that separability analysis should be extended by bare soils class.

9. Lines 323-325

It goes without saying. The minimum, mean and maximum values will never be in a different order.

10. Figure 8

The results show changes from “agro-forest” to “bare soil”. There is no water bodies, despite the analysis they were carried out on the basis of the "water" class. Please correct this inconsistency.

11. Subsection 4.1

Analysis o susceptibility of Sentinel-1 data to cloud cover is not subject matter of the manuscript. It can be mentioned, but not so extensively described.

12. All Figures, Schemes and Tables should be inserted into the main text close to their first citation.

• Figure 1 should be moved to subsection 2.1 Study Area
• Figure 2 should be moved to subsection 2.2.1 Sentinel-1 Time Series

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 4 Report

Thank authors for answering the comments and made corrections. The manuscript looks clearer.  I thought so, that SLC products were used by authors for wider research. The remarks in lines 583-586 are valuable.

Comment 10 – figure 8

Reviewer:

The results show changes from “agro-forest” to “bare soil”. There are no water bodies, despite the analysis they were carried out on the basis of the "water" class. Please correct this inconsistency.

Authors:

Unfortunately, we couldn’t find the “bare soil” the reviewer referred to in Figure 8. However, as explained, the dominant cover that remains after vegetation removal is water.

See Line 226-227

Reviewer:

Thank you for the figure 2 (in answers to the reviewer), it explains a lot. At the first look at figure 8 (manuscript version 1), white pixels (in red oval) on the Google image seem to be a bare soil, rather than water. Usually, dark pixels on compositions from optical sensors represent water bodies. Maybe, authors can provide high resolution image for this example, or if it possible the color composition including Near Infrared band to omit this kind of misinterpretation. The authors could insert this visualisation in 2.2.3 Reference data subsection for example.

Comments for author File: Comments.pdf

Author Response

Please see the attached

Author Response File: Author Response.docx