Mapping a Paddy Rice Area in a Cloudy and Rainy Region Using Spatiotemporal Data Fusion and a Phenology-Based Algorithm

Round 1

Reviewer 1 Report

The authors address the topic of paddy rice mapping using optical satellite imagery. They employ a data fusion method to combine Sentinel-2, Landsat-8 and MODIS data and employ a phenology-based approach for detecting rice cropping regions in Central China. Results were validated against local in-situ observations as well as administrative statistics.

The topic is relevant in the context of this journal, however, the approach presented here is not really a new methodology but a combination of existing ones. It comprises of multiple independently developed techniques that are combined to a workflow. This is further illustrated by the multitude of different software and programming languages mentioned. 

Before it can be considered for publication, the manuscript needs significant improvements. This includes structure and methodology description but also language and style. I recommend the manuscript to be carefully reviewed regarding language and phrasing, possibly by a native speaker. Although the introduction provides a quite comprehensive overview of the literature background and some of the methods descriptions are quite detailed, a majority of the text lacks sufficient clarity.

Below, I list my main comments followed by minor points that mostly refer to aspects of style or language. 

Parts taken from the manuscript are italicized.

 

1. There are multiple sentences and paragraphs that need rephrasing to improve clarity and understanding. These include ll. 58f, 64ff, 219ff, 295ff, 315f, 333ff, 349f, 356f, 439f, 443ff, 455f, 561f.
2. Ll. 103ff: this whole paragraph should be improved to increase clarity. It is currently difficult to understand.
3. The entire section 2 needs significant improvements in terms of language ans structure. At the moment, major portions of the text are difficult to understand and partly read more like an instructions manual than a scientific description (especially section 2.2.1). Similarly, section 3 needs significant work as well.
4. Ll. 122ff: this part provides duplicate information to the descriptions directly before it. It is also not clear why heat or solar energy resources are of relevance in this context.
5. L. 127: are you referring to weather forecasts or weather measurements? If it is the former, then why not use the actual measurements if you are looking at past events?
6. L. 132: what do you mean by  “promotion and demonstration significance”?
7. Ll. 142ff: atmospheric correction using Sen2Cor is mentioned but it is not clarified if it was actually used in this study.
8. Section 2.2.2 and Figure 3: why do you describe all regions of Chongqing, including cropping practices, if you are later (as I understand it) only working on the western part?
9. Section 2.2.3 and 2.2.4: the descriptions of the mapping process and obtaining the auxiliary data are too detailed. Please shorten and potentially merge them.
10. L. 194: it is unclear if land use data is the only data used or just one out of multiple datasets.
11. L. 204: please explain what kind of mask was established.
12. Eq. 1: this seems to be conceptually quite similar to a bilateral filter, although with a different objective. Can you please elaborate a bit more on the concept?
13. L. 259: what is meant by “realize the batch cloud removal function”?
14. Section 3.2: at the beginning of this section, it is not obvious what the purpose of this method actually is. Please restructure the paragraph to better guide the reader.
15. L. 268: is the statement on ISODATA a general part of this step or a comment on the particular implementation used in this study?
16. L. 290: you write that t2 is also considered but not in what way.
17. L. 319: what previous study are you referring to?
18. L. 327: please choose a different, more specific title for this subsection.
19. Section 3.4.2 is difficult to follow. Please improve it. It is also not necessary to explain the concept of a confusion matrix here.
20. L. 383: you say MNSPI is generally better but it is not clear what you are referring to (better than what?).
21. In section 4.1 (and the remaining manuscript) it is not clarified properly if RMSE and r are calculated just on the restored pixels or over the whole image (patch). This would have significant impact on the obtained values. If the metrics are calculated over a whole image patch where only a small portion was masked out and restored, it is logical that the resulting error would be fairly low since the majority of pixels are identical in both images.
22. Ll. 399ff: you say that NDVI reconstruction works much better than LSWI reconstruction but you make no effort of explaining it. Could this be connected to the different resolution of SWIR bands (20 m)? Please elaborate.
23. L. 436: please explain what you mean by “triple convolution method” or refer to a source.
24. Ll. 446ff: this should be provided in a table, not as a text. It is also basically the same information as Figure 10 if I understand it correctly. It is not necessary to have both.
25. L. 460f: this sentence should be rephrased. It is better to actually describe the landscape instead of describing the distribution of elevation or slope values in the dataset.
26. L. 468: have you considered any postprocessing to reduce the salt and pepper effect?
27. Ll. 493ff: the beginning of this paragraph reads like a part of the introduction and is not really needed for the discussion.
28. Ll. 507ff: the advantages presented here are very vague and lack any explanation. Please elaborate. It is also not clear what is meant by “very friendly to people”? You mean user-friendly?
29. Ll. 520f: it is not clear what this sentence is supposed to convey.
30. Ll. 534ff: you mention that small rice fields form mixed pixels with other land cover. What are the effects of this? How does it affect your results?
31. Ll. 555ff: this part again reads like an instruction text.

 

Figures and tables

• Figure 3: this should be a table, not a figure. And please move the description (“Note”) to the actual table description.
• Figure 5: it would be good to add the numbers used in the paragraph above to the flowchart to help the reader follow along the descriptions. In that case, please add the explanation of the steps to the figure description.
• Figure 6: it is not obvious why you display so many different land covers if you are only interested in paddy rice. Also, please consider making this figure smaller (condensing the plots, reducing white space).
• Figure 9: I would suggest choosing a different image arrangement or better explaining the current arrangement as it is not immediately clear to the reader.
• Figure 12: please add an explanation of the three images to the description.
• Table 5: the table should be rearranged to help understanding the context. I would recommend avoiding wrapped columns like these. It leads to confusion regarding the “Total area” row since it is not located below the actual data it summarizes.
• Figure 13 should be smaller.

 

Language and style

• There are multiple occasions of unclear terms being used. Please explain or replace:
  • 78: “long interval”
  • 79: “recovery effect”
  • 533: “smashed”
  • 122: “remarkable”
• l. 17: “…high-quality remotely sensed paddy rice-growing area maps is limited due to frequent cloud cover and rainfall…”
• l. 19: “…combining a spatiotemporal fusion algorithm with a phenology-based technique.”
• l. 23: please remove the comma after “then”.
• l. 27: “Our results were validated by the field survey data and showed a high…”
• l. 43: “Time series of remote sensing data, …”
• l. 51: “paddy rice fields.”
• ll. 61f: please use the term “temporal resolution” rather than “time resolution”.
• l. 71: “Removing the thick clouds from the image…”
• l. 75: “…to reconstruct spectral values of the cloud areas in Landsat images.” Mistakes like these occur many times throughout the manuscript, e.g. ll. 78, 102, 296, 552.
• l. 77: “…cloud pollution from satellite images over land.”
• l. 216: “Removal of thick clouds”
• l. 242: “…the system is a time series of images, …”
• l. 243: “…and the output is a time series images but without missing pixels…”
• l. 515: “…the correlation may decrease, and the prediction accuracy may decrease.”
• ll. 511ff: since the individual points are quite long, it would be better to write “First, … Second, …” instead of using numbers.
• ll. 519f: “The longer the time interval…”
• l. 523: “In this research, MODIS images are used as input.”
• ll. 525f: “One problem common to the two methods above…”

 

Miscellaneous

• L. 95: do you mean crop type detection here?
• L. 136ff: please be careful about the phrasing. The Sentinel-2 system comprises two satellites, but it is not divided into two. Further, revisit periods are always given at the equator and change with increasing latitude.
• L. 142 & 145 & 155: I do not think that the URLs are necessary here. Explaining about data formats of downloaded files is also not necessary.
• Ll. 281f: please use a clearer time format.
• Eq. 2 & 3: what is the purpose of the ‘ in the formula (above “swir”)?
• L. 358: the correlation coefficient (Pearson) is usually denoted by a lower-case “r”. Please use this notation throughout the manuscript.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Minor comments:

line 580: remove "Please add"

Figure 11: create insert maps so that the distribution of paddy rice can be clearly seen. Similar to figure 4

line 240: Zhu et al. [add a reference number]

line 195: re-write: We used existing land-use data ….

Add more information about how official statistical data was generated. This can be done under section 2

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

How to map crops in cloudy and rainy region is important and interesting topic.The authors constructed the NDVI and LSWI time series using the fused MODIS, Landsat-8 and Sentinel-2 data, and mapped the paddy rice based on the obtained vegetation indices time series and the physical characteristics of paddy rice. Different from previous paddy rice mapping studies, cloud removal and multi-source data fusion were emphasized in this study.

However, the description of this paper cannot highlight the innovation of data processing methods or paddy rice mapping algorithm. Moreover, there are some questions that may be addressed:

1. In the process of multi-source data fusion, the TOA data of Sentinel-2, the SR data of Landsat-8 and MODIS are selected. How do you consider the differences between different data types? It is recommended to select Sentinel-2 SR data, which can be obtained free of charge on many data processing cloud platforms such as GEE.
2. Why only fuse Sentinel-2 data and MODIS data instead of Sentinel-2, Landsat-8 and MODIS data?
3. Has the vegetation index time series used to identify paddy rice been further processed (such as interpolation, composing and smoothing)?
4. This study used LSWI + 0.1 ≥ NDVI to identify the transplanting flood period. What was the specific time window to meet this rule? Moreover, after identifying this signal, how did this study observe the follow-up changes of NDVI time series?
5. This study claimed that the NDVI time series of paddy rice reaches its peak in about 100 days, which is inconsistent with Figure 6a.
6. In the process of generating different masks, I noticed that the author used LSWI < NDVI during the growing season as the only condition for identifying crops and forests in non-flooded period. How was the “growing season” defined? Most importantly, previous studies have shown that paddy rice pixels also meet LSWI < NDVI in the whole growing season(Wang et al. 2020).
7. In the period of frequent rainy season, the LSWI of built area cannot always be less than 0. Moreover, the LSWI > NDVI may not be satisfied for rivers in dry season.
8. The study area is large and there are too few sample points for verification.

In general, I think the experimental process of this paper needs to be greatly improved.

Wang, J., Xiao, X., Liu, L., Wu, X., Qin, Y., Steiner, J.L., & Dong, J. (2020). Mapping sugarcane plantation dynamics in Guangxi, China, by time series Sentinel-1, Sentinel-2 and Landsat images. Remote Sensing of Environment, 24710.1016/j.rse.2020.111951

 

Specific comments:

L35 rice guarantees clothing…?

L46 Phenology-based algorithms have been developed. Need add reference, e.g.?

Figure13, Statistic Data in axis is not required uppercase.

L497 “the classification of” revised to ”map” the paddy rice

L529 after the following sentence “it is recommended to use a high performance computer or parallel computing to increase the calculation speed.”, add. e.g. Pei Engine, Google Earth Engine (Pan, L., Xia, H., Zhao, X., Guo, Y., Qin, Y., 2021. Mapping Winter Crops Using a Phenology Algorithm, Time-Series Sentinel-2 and Landsat-7/8 Images, and Google Earth Engine. Remote Sensing 13, 2510.)

L569-574 rewrite these sentences, including fusion of optical and radar data, clouding computing platform, object-based methods.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The authors addressed all the main points and significantly improved the manuscript.

Reviewer 3 Report

Accept in present form