Satellite Advanced Spaceborne Thermal Emission and Reflection Radiometer Mineral Maps of Australia Unmixed of Their Green and Dry Vegetation Components: Implications for Mapping (Paleo) Sediment Erosion–Transport–Deposition Processes

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This study is very interesting and important for the geoscience information extraction of remote sensing data. The author tried to reevaluate the mineral products suitable for mapping (paleo)geomorphic features, including the silica index (~quartz sand content), AlOH content and composition (e.g., kaolinite, muscovite/illite) and water content. The unmixing strategy was used to suppress/remove the effects of both the green and dry vegetation components. This paper has been written good. However, some minor issues need to further demonstrated before publication.

(1)    Some legends were missing in figure, resulting in comprehension difficulty. For example, The band ratio was applied to AlOH content in figure 6c,6d. however, how to evaluate these two grey-scale maps? There are the same problem in fig.7 and fig. 8. The author should try to quantify evaluate these maps, especially, the difference between maps.

(2)    In discussion, although, band ratio was very useful. Some machine learning methods have been employed to extract geoscience information based on remote sensing, these work should be discussed.

Author Response

Thanks for your review.  Much appreciated. Responses to your issues as follows:

• A statement in the caption of Figure 6 now includes “Brighter tones in the grey-scale images equate to greater amounts of AlOH mineral content.” The difference between the AlOH content maps (6c and 6d) is a linear combination of the dry (and to a much lesser extent) green vegetation products (Equations 1, 2 and 4). Figure 4 shows this result by presenting the green and dry vegetation products as well as the AlOH content products, before and after unmixing. Given this demonstration, the later Figures focused on the derived geoscience information.  Including the dry and green vegetation products in these later figures, as well as (ideally) the mineral products before unmixing, would generate very large figure multi-panel mosaics, showing the same “unmixing” story but with the negative consequence of losing the ability to see the geoscience information.   For example, in Figure 6, the key message is that paleo-drainage across the Canning Basin (yellow polygons in Figure 6a) is revealed by vegetation unmixing the AlOH content (Figure 6d).
• This paper is about vegetation unmixing – it is not a comparison between ratios and other information extraction methods like “machine learning”. Having said this, I have expanded the Discussion to include a statement about more automated methods because the method described here is subjective in that the gains in Equation 4 are determined by the user based on image-visual cues. That is, a more objective method is required, which down the track a “machine learning” system could be trained to do.

Reviewer 2 Report

Comments and Suggestions for Authors

Dear author, dear Tom,

Thank you for a paper in which you present the improvements on an already impressive product. I had a lot of pleasure reading it!

My comments on the manuscript:

- you seem to present 2 papers in a single manuscript, the first being the improved ASTER products and the second the geological findings, where especially the novelty on paleochannels is competing with the first novelty. Given that several geological studies are needed to explain the improvements you have brought to the ASTER mosaics, it can go as is. Still, it would also be possible to have each novelty focussed in a shorter paper (where the paleochannels story leans on the improved products and cannot be published without).

- Recently, a mosaic of EMIT was released (albeit as a picture rather than as an online map like the ASTER mosaics (https://earth.jpl.nasa.gov/emit/news/27/nasa-sensor-produces-first-global-maps-of-surface-minerals-in-arid-regions/). The other hyperspectral sensors have global coverage, but will not be able to make a global covering map, likely not even continental.

- There is also a recent mosaic made with Sentinel-2 and Landsat. These do not have the discriminative capacities of ASTER when it comes to geology, but the high volume acquired by these two sensors led to different mosaicking techniques (pixel selection) that allow the suppression of vegetation by selecting the most favourite pixels from a collection. Have a look at "Roberts, D.; Wilford, J.; Ghattas, O. Exposed soil and mineral map of the Australian continent revealing the land at its barest. Nat. Commun. 2019, 10, 11".

- Other recent work with time-series of Sentinel-2 has shown that soil moisture is a major source of variation in spectral indicators (in your results nicely shown by the Canning data). Longer-term seasonal variations seem to have a more pronounced effect than single rainfall events. I assume this is what you mean with "climate-driven rainfall" in conclusion #7, yet it would be helpful to better specify the time and space dimensions of weather influences on the ASTER mosaic results. This also holds for the beginning of the discussion, where you touch upon this subject (line 705).

- table 1: column 6 seems to have the wrong label, it mentions band 2208 when I think it should be 2165. Looking at where the ASD indices should have a correlation of 1, It might even be possible that the labels of columns 5 and 6 have been switched. Please have a careful look at it.

- some figures could be presented a bit bigger / sharper; note that MDPI allows to use the full page width for figures. Except for figures 5 and 6, which span a page already, some others could positively be shown larger.

-l700: the word "clearer" is a bit vague, I think a more accurate word can be used here.

-l706: please shortly repeat here why some products do not need vege unmixing (or did not improve by it).

 

Comments on the Quality of English Language

No issues, just two typos:

l56: "3especially"

l228: "two e 2165-2200"

Author Response

Thanks for your support and comments of this work.  Comments to your points as follows:

1. Throughout this work (3 years in the formulation) I could series a series of linked papers.  But as you suggest, a demonstration of the effectiveness of the method  was required. The  geological (mineral exploration) opportunities are left for other paper/s, especially as the subsurface (3^rd) dimension could also be incorporated using similar drill core VNIR-SWIR-TIR data.
2. I have seen the work of Roberts et al using Sentinel-2 and Landsat data which is similar to earlier but unpublished work done by Japanese colleagues (at JOGMEC) using ASTER to find the “barest” pixels.
3. There were quite a few unexpected “water-related” results. For example, in the NGSA field sample data, all the “raw” ASTER SWIR bands as well as most of the indices tested in Table 1 are significantly correlated with the depth of the 1900 nm absorption. A more detailed look by the various satellite HS sensor teams at weather/climate time-space drivers on the satellite SWIR signal is arguably needed. For example, I note how NASA’s preliminary EMIT global desert mineral maps appear to go “black” in fringing higher rainfall (more vegetated) zones. Is this just a problem with Tetracorder because it does not properly account for (unmix) the climate-related vegetation cover?   
4. Table 1 is now fixed. Very observant. Columns 5 and 6 are correct.
5.  I will leave it to MDPI editors to maximise figure size.  
6. Changed to “more accurate geoscience information is generated”
7. Changed to “These “affected” products are driven by the wavelength regions used for the mineral indices and whether these overlap with either green or dry vegetation spectral features. That is, where the vegetation is effectively “aspectral”, then it has no apparent effect on the mineral index. Thus, the SWIR and TIR mineral indices (e.g., the ASTER AlOH and silica indices, respectively) are more prone to dry vegetation effects while VNIR mineral indices (e.g. iron oxyhydroxides) are more prone to green vegetation effects.”
8. Typos are now fixed.

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript is well-written and organized. In this study, ASTER indices for mapping dry vegetation, vegetation unmixed and regional “wet” and “dry” zones were developed that include possible implications for mapping sediment erosion-transport-deposition processes.  It is novel and applicable method  for improved mapping and solutions to geoscience challenges. The manuscript can be accepted after some minor corrections. Details of corrections are provided for better presentation of the manuscript as follows: 

1. Please reduce the numbers of keywords to most important one. You provide 15 keywords which some of them repeated for example AlOH content and AlOH composition. 

2. Line 51-52 please remove @ before 15 m, 30 m and 90 m. I can not understand the meaning.  Line 119, too. 

3. Please define objectives according to research questions that you provided in the last part of your introduction. 

4. After introduction, please provide some information about the geological setting of your study area, as a readership I would like to know about the geological characteristics of the study area before realizing the materials and methods. 

5.  Figure 1, yellow polylines hardly can been seen. Please increase the resolution of the image. 

6. Please provide a flowchart of the methodology that you applied. It can provide an overview of the techniques you used.

7. Please increase the resolution of Figures 3, 5 and 8, some of the band ratios hardly can be seen. 

8.  Discussion and conclusions are well-written. 

Author Response

Thanks for your useful comments. Specific responses as follows:

1. Key word number reduced.
2. @ removed.
3. New paragraph inserted at the end of the Introduction describing the two primary objectives.
4. A “Geological Setting” prior to “Materials and Methods” is a challenge given the work spans the entire Australian continent. Book reference [48], “Shaping a Nation: A Geology of Australia” is cited in the Introduction. The first map of Australia is presented in Figure 5, which also makes the order/timing problematic.  More detailed study area geology maps (e.g. Figure 7) are even later in the paper compounding this problem. To solve this, (i) geologic/tectonic/geomorphic/climate context (with references) is provided both in the Introduction; as well as (ii) the first paragraph of each of the three each sub-sections in the “Results” (one for each of the three more detailed examples of improved mapping), with accompanying published geoscience maps (where required).
5. Yellow (and cyan now for paleorivers) lines have been increased in thickness in Figures 1 and 4.
6. Is a flowchart diagram necessary given the unmixing methodology comprises a maximum of two processing steps?
7. MDPI will maximize the figure size where possible.