Smallholder Crop Type Mapping and Rotation Monitoring in Mountainous Areas with Sentinel-1/2 Imagery

Round 1

Reviewer 1 Report

General comments:

The paper presents a Random Forest classification for mapping agricultural areas in mountainous regions in China using Sentinel 1 and 2 images. The introduction presents a literature review of the main points covered in the paper. However, the knowledge gaps are not unified in a single paragraph, making it challenging to understand which knowledge gaps you investigated with the research, and how this approach would differ from other published papers. Moreover, in the introduction section, some paragraphs are wordy. It would be easier to read if you focus on the key message of each paragraph.

The methodology section is scientifically sound, and the authors properly answered the questions asked in the previous review. A key of interpretation should be considered, making it easier for non-local readers to understand the complexity of the study region.

Results are well written, focusing only on the findings of the paper.

The discussion section is the main point that needs attention in this paper. This is because a large part of what is written should not be in the discussion, as a scientific discussion should focus on interpreting the results based on another published research.

The conclusion section should focus on answering the research question.

Regarding the writing, there is still some room for improvement as you should decide if you are writing in the first or the third person. Mixing writing styles makes the article difficult to read.

Specific comments:

Lines 77-79: It would be relevant to show which are these papers and present what classes they use.

Line 80: The use of the word “suffer” does not seem suitable.

Lines 79-86: These lines should be in the same paragraph, as they present the knowledge gaps of the paper. It would be important to organise all knowledge gaps in one paragraph, concluding in how your paper will contribute to reducing these gaps.

Lines 120-129: All this information must be referenced, or you must mention how you acquired them. Since many Remote Sensing readers are unaware of the study site. Adding these references would make the text more reliable.

Line 157: This is not a commonly used expression in scientific journals.

Line 163: You should decide to write in the first or third person your paper.

Lines 182-183: “as the change of cropland is negligible within several years” must be referenced.

Lines 186-187 and 198-199: It would be relevant to see your key of interpretation with photos of each crop, RGB image of each crop (at the proper coordinate). This would be important for readers from other countries to see the complexity of separating these crops.

Table 3: You should discriminate at each phenological stage you acquired your ROI, as this can interfere with the final accuracy result.

Lines 212-215: You do not have to explain the RF usage again, as you did in the Introduction. Just explain detailed how it works theoretical and how it is applied in GEE.

Line 264: “overall accuracy” Use only the acronym throughout the text.

Lines 291-294: This is a conclusion.

Figures 8 and 9: Is it possible to put these figures together? It would be interesting in terms of comparison.

Lines 331-353: This is not a discussion. In a scientific discussion, focus on how your results compare to others in the literature, what explains your results, how they can contribute to help fill knowledge gaps in the field, etc.

Lines 413-416: Should not be in the conclusion section. This section should focus on answering your research questions with your findings.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

1. Very good research, very relevant and timely
2. Written in a good style, easy to read and understand.

 

Please see the following comments;

 

3. In your introduction and discussions, it will be nice to relate findings from crop type studies in smallholder focused on the Sentinel family, especially those zooming into complex mountainous terrain like the current article. Issues regarding cloud cover, large image data mining, ML/RF classifiers, indices (VIs) Versus raw bands (SR), complex farming systems and features of importance, as well as crop distributions along elevation and slope gradients can be related with their findings

Please see:

Ibrahim, E. S.; Rufin, P.; Nill, L.; Kamali, B.; Nendel, C.; Hostert, P. Mapping Crop Types and Cropping Systems in Nigeria with Sentinel-2 Imagery. Remote Sens. 2021. 13, 17; DOI: 10.3390/rs13173523.

Mazarire, T.T.; Ratshiedana, P.E.; Nyamugama, A.; Adam, E.; Chirima, G. Exploring machine learning algorithms for mapping crop types in a heterogeneous agriculture landscape using Sentinel-2 data. A case study of Free State Province, South Africa. South Afric. J. Geomat. 2020, 9, doi:10.4314/sajg.v9i2.22.

etc.

4. Figure 2 –Very nice, but no mention of how it helped in building the features/mapping windows. Did you use images for the entire calendar year or a cropping calendar? Means, minimum, maximum, median or all images?

 

5. Line 157- please rephrase “However, it is a pity that …………….”

 

6. 164 - Please rephrase “picked good observations ……………….”

 

9. There is no mention of how crop rotation was mapped and estimated in the method section. Please include
10. Figure 8 – label zoom ins (b and c) in a, same as figure 9 (also no table or statements on the total area of the crop types). The maps are good, but there is a need to relate the estimated area coverage of the maps. Please insert a table for both years.
11. Figure 10 - is difficult to follow, I will suggest having a separate map for stable and another map for rotated fields (changed and unchanged). You also did not report the statistics of the map (same as figure 8 and 9). Which crop class has the most rotation and vice versa? Please insert a table

 

11. Line 352- how does this accuracies compare with studies where only Sentinel 2 is used? And what were the major advantages of integrating Sentinel1? See Ibrahim et al 2021 and Mazazire et al., 2020.

 

12. You reported that you observed significant rotations (please give area coverage), however, did not discuss the reasons for these rotations based on literature or field observations. Or could it be based on the mapping approach? please discuss approaprately
13. This study is for a small holder, but no mention of the field sizes, either based on field observations, literature or satellite base estimations (see Fritz et al., 2016 (satellite based, - deduct for your study area), ibrahim et al 2021 (satellite based and field observation) and Carletto et al., 2016 (questionnaire) ). This is relevant in the context of accuracy, pixel contamination, spectral mixing, etc. Please add in appropriate sections.

 

Fritz, S.; See, L.; McCallum, I.; You, L.; Bun, A.; Moltchanova, E.; Duerauer, M.; Albrecht, F.; Schill, C.; Perger, C.; et al. Mapping global cropland and field size. Glob. Chang. Biol. 2016, 21, 1980–1992.

 

 

Carletto, C.; Jolliffe, D.; Banerjee, R. From Tragedy to Renaissance: Improving Agricultural Data for Better Policies. J. Dev. Stud. 2015, 51, 133–148

 

 

 

 

 

 

 

 

 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The authors adequately addressed the suggestions made, and the article is suitable for publication.

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

General comments

The manuscript evaluates the use of optical (Sentinel 2) and SAR images (sentinel 1) for the generation of crop type maps. Results show that the combination of both sources of information gives better results. Something crucial as how the training and validation data is obtained is not completely clearly described. It is not clear how many of the points came from field measurements, how many came from visual interpretation, and how the samples are distributed among zones. Using visual interpretation for training and validation of crop type maps is risky. A clear description of training and validation data generation should be done before a new version of the manuscript can be evaluated. Description of results is poor. Results should show the differences in accuracies among each crop and zone separately (considering crop - zone interaction effects). In addition, it should be clearly differenced the expected effects in the two zones from the results observed in different crop types. General mention of polarization when talking about Sentinel 1 should be removed. As all the S1 bands have a polarization.

Detailed comments:

Abstract Line 29. Specify which polarization; all S1 bands have a polarization. Or remove the mention to polarization and spectral bands. Lines 66-69. Find more examples of S1 and S2 fusion. You should mention in the introduction, the problems expected of using SAR in mountainous areas (e.g. incidence angle effect due to different slopes).

Line 73. Remove the mention to a rice classification method, because you are talking in general about crops. Line 85 “With the abundant satellite resources” is very similar to line 71 in paragraph above.

Lines 91-93 . Rewrite the sentence. Suggestion: “particularly in the case of mountainous area s and smallhoder systems”.

Lines 93-95. Not very clear a direct relation between crop rotation description and yield. Please describe more in detail.

Lines 96-97. Specify where is located the study area.

Line 101. Are you testing different polarizations in S1?

Line 102. Suggestion to change “describe” instead of “construct”.

Line 104. What do you refer with “classification stability”? Also it is risky to say crop rotation only considering two years. You can say “interannual variability”.

Lines 111-118. Please rewrite these lines. Zones A and B are mentioned before describing the AEZ zonation. Also describe more in detail the differences among the two areas.

Lines 120-121. Include in the description of the figure that the color scale is height in meters (?).

Lines 126. Describe briefly how the phenology varies in the area.

Lines 138-139. Remove “for crop type mapping and rotation monitoring”. Line 140. Specify that these processing was done by Google Earth Engine. The processed data is what is available there. I suggest also the application of a speckle filter to reduce salt and pepper effects mentioned in line 331 (see Dingle Robertson et al, 2021).

Line 163. Remove “The”.

Line 165. “Along with the crop growing season”, same as line 142.

Line 181. Rewrite “To avoid the variability between crop and non-crop”

Line 185. Change to “The product obtained an overall accuracy of more than 94% over China” Line 187. Rewrite “it could also for cropland mask in 2017 and 2018”.

Lines 188-189. No sense to include “After cropland mask, crop could be classified without the interfere of non-crop land cover types.”

Line 192. “We collected ground truth data”

Lines 194-196. Rewrite sentence.

Line 199. Review grammar connectors.

Lines 204-207. It is difficult or no precise to get samples of crop types also using high resolution images.

Lines 207-210. You need to specify where are located the ROIs, how many reference points came from filed surveys and how much came from visual interpretation, and how many points are in each ROI, and how many points for the different crops are available in each zone.

Line 237. Change equation 1, including the accuracies of the different classes. Line 242. Where is the discrimination between zones in Figure 5?

Line 254. Why you mention classification?

Lines 265-267. Please describe the complete results obtained. Accuracy for S1 bands (G3) was higher only in AEZ A in 2018.

Line 270. What do you mean with tendency?

Lines 283-298. Figure 7 should discriminate the results between zones A and B. It is not described with Feature combination results are described.

Line 331. Salt and pepper effects could be removed using a speckle filter for SAR bands.

Line 372. The difference in zones should be mentioned when comparing the results of wheat/rapeseed (mostly located in zone A) and corn/soybean (located mostly in zone B).

Line 385. Replace “was obvious” with “is”.

Line 391. Elevation were corn and soybean is present is not high.

Line 427. Is “obvious”?

Line 438. Which are the optimal features? Please mention it.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors (will be shown to authors)

The manuscript mainly deals with the supervised classification of 4 to 5 annual crops using both the optical and SAR satellites data of Sentinel-1/2 in a mountainous region of north-east China. Some critical aspects are needed authors attentions before accepting for publication. My key point is that I am confused about the title of this manuscript compared to its objectives and results.

3 important words of the title of this manuscript (smallholder, mountainous, rotation) are not really defined and presented in the introduction (what are the problematics and questions), nor in the data/study area part (characterization) and nor in the results and discussion parts. I think that it is usefull and interesting to bring more informations and contextualization, even if some of these 3 words are removed from the title.

Smallholder. Is field size a problematic compared to the spatial resolution of Sentinel ? Do you have any data on field size, length and width ? Problems with plot’s edge (trees…) ? What are the results ?

Mountainous. Where are cultivated the crops ? Aree there some problems with the slope, in relation with SAR data ? What are the results ?

Rotation monitoring. What is expected with a transition matrix between 2 years ? That the fields limits remain stable, whatever crops change or not ? For example, L102-104 is not very clear. What is classification stability according to you ? and why should the classification be stable between years ?

Specific comments:

There are some typos in this paper. The writting of ‘Sentinel’ is frequently wrong (written ‘Sentien’ for example). L79 : why is the word ‘free’, where is the verb ? L167 : cloud and not could !? L187 : where is the verb ? L284 : not AZEs, but AEZs, L380 : may be (and not maybe), L396 : many (and not may)

It’s not clear if the classification of rice is an objective or not. L98, : rice is not indicated for example.

Ligne 20 : ‘is still challenging due to the inter-class variability’. I think on the contrary that a strong inter-class variability helps to perform classification, but a strong intra-class variability is a difficulty to get good classification accuracy. Idem L244.

L107-126. Figure 2 is very interesting. But it would be usefull to add some informations to this ‘study area’ presentation’s part. Especially about moutainous context (until which altitude are cultivated the crops ? in flat or sloppy areas) smallholder farming system (size of the farms, size of the fields…). Is there a problematic of crop rotation and why ? How is cultivated rice : with or without field inondation ? Are there differences between crops area in AEZ A and B , concerning altitude, climatic conditions (mean temperature, rain) and agronomic pratices or crops ?

L109-110 (‘As the cropland is limited in the Inner Mongolia, Hulunbuir city and Hinggan league (115°31′-126°04′E, 44°14′-53°20′N) (Figure 1b),’). Could you be more explicit, in the map or in the explanation (I don’t know or see what or where are Hulunbuir city and Hinggan league in the map).

Parts 2.2.1. and 2.2.2. and figure 3. The number of images is limited to the agronomic season (‘Along with the crop growing season’) : please add this comment to figure 3 legend and precise in the text the concerned period (first and last month for example).

L167 ‘we firstly pick out the good observations with could cover less than 20%’. Please precise the method or tool used (visually photo-interpretated ?).

Table 2. You can improve this table by adding one bibliographic reference par indice, but it’s an optionnal remark.

2.3 Reference data (for crops). I’m not sure to understand what you have really done. Did you use your sample points (is there only one point per field ?) ? If not, it’s not usefull to show them. Or did you have an intermediate procedure mixing sample point and ROI ? If yes, you may add a figure with a zoom in a small area with a sentinel-2 background to illustrate the position of yours points, the final ROI used and the typical size and width of the fields (like figure 8a). You may explain why you get 297 ROI ‘others’ and for what use in this manuscript ?

Figure 2 : it’s difficult to see the color of the points. Could you also enlarge the altitude legend ?

2.4. Crop type classification. Did you realize a negative buffering of your ROI before using them ?
How many classification run did you realize ? Only one ? It’s a pity since repeting several runs allow to calculate a confident interval and check the stability of the classification.

3.1. This part is not clear for me if you present the mean value (and box-plot) of all ROI sample (ground truth) – if yes, the title of 3.1. is not adequate ? or something else ? Be more precise.

L310-311. ‘The crop type distribution was not promiscuous, and each crop type demonstrated an explicit range’. This phrase is very important but it’s not clear enough. What do you mean ? Could you add another adjective to ‘promiscuous’ ?

L354-370 (discussion about training data). Some informations should be in the methodological part and not in the discussion part. On the contrary, the discussion part about training and validation samples need to be more detailed, precised and discussed.

Discussion. It lacks (i) a comparison of this paper classication performance with bibliographic references, for the same kind of crops (ii) why such differences beetween the 2 years – cf figure 6 (iii) are there some regional statistics to compare with the classification results ?

L385-392 : do you suggest that maize have a relatively small size in this region ? Do you have any facts, measurements, in situ references or local bibliography ?

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

General comments about new version of the manuscript

 

This new version of the manuscript doesn’t fulfill the main considerations of the reviewer. Description of results is still poor. Description of rotation results should be much more in deep considering that it is mentioned in the title and objectives. There persist inconsistencies between description of the results and what is showed in Figures. Also the description of figures is still confusing, when the zones are mentioned but not indicated in the figures. I consider that the mention of several words is not adequate. For example, the mention of smallholder farmers as characteristic of the study region is risky. As showed in Figure R2, medium and large fields are also quite relevant, and probably in area are more frequent than small or very small farmers. In addition, “mountainous areas” can be a characteristic of part of the study area (zone A), but not for the whole region. The term “polarization band” in SAR images has no sense unless you mention the different polarizations. In addition, several of the changes in the text, as well as the response to reviewer, have grammar and typing errors; please review carefully the grammar of all the manuscript with a teacher or English translator.

 

Response to reviewer’s comments in original version of the manuscript:

Abstract. Line 29 (original version). My comments were not considered adequately. There is no need to mention SAR polarization bands if you don’t mention to which polarization band are referring. All SAR bands are polarized. In addition, there is no mention to the effect of incidence angle on SAR backscatter. Also the relation of mountainous areas with clouds is not general, there exists arid regions over mountain areas too. It is contradictory that you mention lower precipitation in zone B (lower altitude) than in zone A.

Line 126 (original version). You mention that wheat and rapeseed are sown at the same time, but you don´t specify when it occurs. Also in the mention of harvesting you need to specify earlier than what.

Lines 138-139 (original version). As in other changes in the manuscript there are grammar/writing errors. I still recommend the use of a speckle filter as part of the processing of SAR images.

Line 192 (original version). Again are observed grammar errors  in the changes in the text. Please use “we collected”.

Lines 204-207 (original version). It is still confusing that if you used HRI images only to provide information of crop field boundaries, you mention problems that can’t be solved with this, like the limited number of samples and the spatial distribution of samples. Please clarify this, also in the manuscript. It is not clear also, considering this approach, why the number of ROIs is much lower than the number of points. Please explain this also in the text.

Line 242 (original version). It is not clear to mention the zones while they are not described in the figure. You should mention the crops and that some crops are more relevant in one zone ant the other crops in another zone. In line 131 (new version), you mention that “soybean and corn are mostly in AEZ B”. You are not saying that the other crops (i.e. wheat or rapeseed) can’t be there. You should clarify in the manuscript if for zone A you only selected some classes and for zone B other classes. Also different classifications for each zone can be performed.

Lines 265-267 (original version). The new text also says that G3 showed good performance both years, but in 2017 the performance of G3 was one of the lowest.

Line 270 (original version). I still don’t understand what you refer with tendency, please clarify.

Lines 283-298 (original version). Similar comment than for lines 265-267. If you mention the zones in the results, the figures or groups of data should be separated between zones.

Line 391 (original version). Zone B has lower height and crops are located in the places with less altitude (blue areas in Figure 4. Why are you saying that maize and soybean are in mountainous areas?

Line 427 (original version). I wanted to say in previous review was that when you refer to something obvious, you refer to something that was previously expected (i.e.: something expected before viewing the results of the paper). Why do you expect that differences in soybean and corn should be  larger for Sentinel 1? You should clarify this in the manuscript.

 

Detailed comments about new version of the manuscript

Line 113. Why do you use “and”? Did you separate the zones by altitude? Please describe which characteristics did you use for zonation.

Lines 116-117. It is not expected that in areas with higher altitude you have lower slopes. Please explain more in detail why it happens in these regions. Are you talking about the mean slope values for the whole zone, or the slope in the areas where crops are located?

Line 289. Which combination of bands are you describing in this paragraph and in Figure 7? Please clarify this in the text.

Line 315-316. Add a reference to an object based classification paper.