Quantification of Changes in Rice Production for 2003–2019 with MODIS LAI Data in Pursat Province, Cambodia

Round 1

Reviewer 1 Report

The reviews of the paper entitled “Quantification of changes in rice production for 2003-2019 with MODIS LAI data in Pursat Province, Cambodia”. The study is of value to the rice industry. However, I have few comments;

• Be explicit on the scientific contribution of the paper
• How did you select the interviewees?
• How did you validate the results?
• To what extent did you use your interview results in this study, validation or otherwise?

Author Response

Thank you for your valuable comments. We revised the manuscript according to your comments. 

1. Be explicit on the scientific contribution of the paper.

We added these sentences to the conclusion section (L.438 to l.441), “This method provides a novel utilization of consecutive MODIS LAI data every 8 days since 2003. This result implied that a group of small-sized paddy field can be analyzed. This method will be helpful in quantifying crop growth changes in other countries and regions.”

2. How did you select the interviewees?

We added this sentence to the method section (L119 to L120) “Full-time farmers who agreed to this survey were selected as interviewees.”

3. How did you validate the results? And,
4. To what extent did you use your interview results in this study?

Data analyses were independent to the interview results and we just used it as a reference data for comprehension o LAI changes. We added some explanations in section 2.2. (L126 to lL127).

Reviewer 2 Report

The aim of the research was to investigate the changes in rice production during 17 years in Pursat Province (western part of Cambodia) using MODIS LAI data model. The information were obtained from the website of EOSDIS using relationship between the NDVI and LAI (FPAR). Finally the LAI data from MODIS were used to analyse rice production changes. During the research the interviews with rice growing farmers was also involved. From the conclusions is clear that rice production was improved due to changes of cultivars and fertilization approaches. In some areas was implemented double cropping but in some areas is expected stagnation of the improvement in rice production due to water problems. According my opinion the results could contribute to developing strategies for future rice cropping system of farmers in Cambodia. I consider the conclusions in the manuscript are original. I did not find any serious defects in the work or in the presentation or ethical problems. Research in this area is relatively specific (perhaps for this reason, only 24 references are listed in the manuscript). In my opinion, the keywords are consistent with the content of the article. I do not have any serious comments and manuscript can be published in a Journal Remote Sensing.

Comments for author File: Comments.pdf

Author Response

We are very glad with your opinion. With consideration of the comments, we changed the keywords to “Cambodia; dry season cropping; leaf area index; MODIS; remote sensing; rice”.

Reviewer 3 Report

Overall this is a very interesting paper. It is well written and generally easy to follow. I commend the authors for acquiring farmer survey/interview data. Often remote sensing studies lack on-the-ground information and this paper brings in a lot of local knowledge. Please find my detailed comments below.

Major Comments

• What fraction of the MODIS pixels do the fields encompass? Several fields were noted to be in the same MODIS pixel indicating that 1 MODIS pixel encompasses at least 4 fields. Since the authors only have 9 MODIS pixels for the field analysis, it would be interesting to understand the fraction of rice paddy to non-rice paddy within each pixel. Since P8 and P9 seemed to show a non-vegetation pattern, it would be interesting to see what percentage was covered by cropland.
• The authors should validate a few of their field locations using higher-resolution Landsat data. This would be especially useful if the fields only take up a small percentage of a 500m MODIS pixel. 
• Section 2.3.2. Did the authors only use the centroid of each field to determine which MODIS pixel to use? Or did the authors use the field boundary and extract multiple pixels if the field straddled more than 1 MODIS pixel? If so, how did they merge multiple pixels into a time series? 
• There needs more explanation/clarification on why each type of analysis is important. For example, why are the authors looking at point-level analysis of the fields as well as the transect line? Also, only on lines 170 - 173, do the readers understand that the field level analysis is being used as input (training data) into the area analysis. It would be beneficial to make the link between the various parts of the analysis clearer for readers. 
• Table 2: logically it makes sense having clusters 2003 – 2010 and 2011 – 2019 because the rice crops planting/harvesting dates are moving earlier in the year. This makes sense as one year will impact the following year. Can the authors please explain what processes are happening on the ground for the clusters that have random later years in the earlier cluster (e.g., BK or TC6)? Statistically, they might be clustered but does that make sense in reality?
• How many of the 9 time-series were impacted by clouds? How was the moving average impacted?

Minor Comments

• Line 70: change “tens of years” to “two decades”
• Line 118: change “Interviews to” to “Interviews with”
• Line 140: You can delete either “MCD15A2H version 6” or “MCD15A2Hv006”. They imply the same thing.
• Line 152: Clarify that the 46 values are due to the use of the 8-day LAI product.
• Line 155: I believe the authors meant Fig 2 not Fig 1.
• Line 160: Change “Twi” to “Two”
• Lines 158 – 163: Please add additional citations and evidence in this section. Specifically, is using a 40day moving average for LAI acceptable?
• Line 195: Please define “landrace”
• Line 199: Is “IR” a type of rice? When I first read that sentence, I thought the authors were saying infrared. It might be worth specifying that it is a type of rice or write out the full name.
• Line 199: how long is “intermediate maturity”?
• Figures 4 and 5: The DOY values on the x-axis are too close together. Either stretch out the plots, use a similar font size to Fig 3 or rotate your DOY labels 45 degrees.
• Line 240: It would be interesting to see the LAI graph of P8 and P9. Maybe this can be included in a supplementary. 
• Line 242: Have the authors looked through Landsat or Google Earth to see if this transition can be confirmed?
• Table 3: Do the authors know why P6 had no clustering? Also, what is the predominant land cover type in P8 and P9 that would cause it not to cluster considering P1 (forest) clustered?
• Line 315: Change “A Significant” to “A significant”
• Line 365: Change “cannels” to either “channels” or “canals”

Author Response

Thank you for your valuable comments. We think the revision according to your comments obviously improve our manuscript. 

Responses to major comments:

1. What fraction of the MODIS pixels do the fields encompass? Several fields were noted to be in the same MODIS pixel indicating that 1 MODIS pixel encompasses at least 4 fields. Since the authors only have 9 MODIS pixels for the field analysis, it would be interesting to understand the fraction of rice paddy to non-rice paddy within each pixel. Since P8 and P9 seemed to show a non-vegetation pattern, it would be interesting to see what percentage was covered by cropland.

We selected the pixel of the location of interviewed farmer’s house because his fields located nearby. To help the readers understand the situation, we added Fig. 7. The additional 9 points on the traverse line were selected randomly, just to understand if there were geological tendencies and variations from the mountainside to the lakeside. As a result, the pixel of P6 was more housing districts and those of P8 and P9 were in forested area near the lake (Fig. 8 added).

 

2. The authors should validate a few of their field locations using higher-resolution Landsat data. This would be especially useful if the fields only take up a small percentage of a 500m MODIS pixel.

We validate the locations using Google Earth (Figs. 3, 7 and 8).

 

3. Section 2.3.2. Did the authors only use the centroid of each field to determine which MODIS pixel to use? Or did the authors use the field boundary and extract multiple pixels if the field straddled more than 1 MODIS pixel? If so, how did they merge multiple pixels into a time series?

We recorded GPS coordinates at interviewed farmers’ houses because their fields located nearby. We analyzed the pixel including the GPS coordinates.

 

4. There needs more explanation/clarification on why each type of analysis is important. For example, why are the authors looking at point-level analysis of the fields as well as the transect line? Also, only on lines 170 - 173, do the readers understand that the field level analysis is being used as input (training data) into the area analysis. It would be beneficial to make the link between the various parts of the analysis clearer for readers.

We added some explanation about how to classify data into clusters (p.6, section 2.3.2.) and the reasons and purposes of the analysis in section 2.3.3 (p.6). First, point level analysis in section 2.3.2 was conducted to identify temporal changes and the associations with the LAI and the interview results. For 9 points on the transversal line, we geographically evaluated the annual LAI transitions and its temporal changes from the mountain side to the lake side. Based on the point level analysis, we developed dry season rice index (DSRI), difference between the average LAI of January and April to clarify the spread of the dry season rice cultivation.

 

5. Table 2: logically it makes sense having clusters 2003 – 2010 and 2011 – 2019 because the rice crops planting/harvesting dates are moving earlier in the year. This makes sense as one year will impact the following year. Can the authors please explain what processes are happening on the ground for the clusters that have random later years in the earlier cluster (e.g., BK or TC6)? Statistically, they might be clustered but does that make sense in reality?

We added sentences to explain the possible reasons in p.17, L385 to L389: Some later years classified into the earlier cluster and some earlier years classified into the later cluster were likely due to yearly variation in planting time or rice growth, which might be affected by the timing of rain and the availability of water. In addition, each pixel might contain several paddy fields with temporal variations and different growth characteristics in rice cultivation.

 

6. How many of the 9 time-series were impacted by clouds? How was the moving average impacted?

The original LAI data fluctuated probably due to cloud, but moving average made the annual LAI transition clearer (Fig. 4 was added).

 

Responses to minor comments

1. Line 70: change “tens of years” to “two decades”

Modified as pointed out.

2. Line 118: change “Interviews to” to “Interviews with”

Modified as pointed out.

3. Line 140: You can delete either “MCD15A2H version 6” or “MCD15A2Hv006”. They imply the same thing.

Deleted “(MCD15A2Hv006)”.

4. Line 152: Clarify that the 46 values are due to the use of the 8-day LAI product.

Added “(8-day intervals)” in L156 and K165.

5. Line 155: I believe the authors meant Fig 2 not Fig 1.

Changed Fig. 1 to Fig. 2 in L160.

6. Line 160: Change “Twi” to “Two”

Modified as pointed out.

7. Lines 158 – 163: Please add additional citations and evidence in this section. Specifically, is using a 40day moving average for LAI acceptable?

As mentioned previously (Fig. 4 added), we regarded moving averages of 5 values were suitable for this analysis.

8. Line 195: Please define “landrace”

We used the word, with an explanation in the following sentence: “It was a mixture of different but closely related cultivars.”

9. Line 199: Is “IR” a type of rice? When I first read that sentence, I thought the authors were saying infrared. It might be worth specifying that it is a type of rice or write out the full name.

We added “varieties” in p.7, L212 and modified Table 1 and its explanation.

10. Line 199: how long is “intermediate maturity”?

The explanation for “intermediate cultivars” was in L104 to L106.

11. Figures 4 and 5: The DOY values on the x-axis are too close together. Either stretch out the plots, use a similar font size to Fig 3 or rotate your DOY labels 45 degrees.

We modified the size and the ratio of Figure 6 (previously Fig. 4).

12. Line 240: It would be interesting to see the LAI graph of P8 and P9. Maybe this can be included in a supplementary.

We added the transitions of LAI at P8 and P9 in a supplementary.

13. Line 242: Have the authors looked through Landsat or Google Earth to see if this transition can be confirmed?

We confirmed the land use at P1 on Google Earth. Though the availability of the images in earlier years was limited, the location of P1 was near the boundary between mountain and crop land and previously there seemed mountains.

14. Table 3: Do the authors know why P6 had no clustering? Also, what is the predominant land cover type in P8 and P9 that would cause it not to cluster considering P1 (forest) clustered?

The reason of obtaining no clusters at P6 was possibly due to the mixture components of houses or roads (Fig. 8b added). P8 and P9 were the forested area surrounding the lake. The area is often flooded during the wet season; therefore, the land is not suitable for cropping and vegetation is probably affected by rainfall or other climate conditions.

15. Line 315: Change “A Significant” to “A significant”

Modified as pointed out.

16. Line 365: Change “cannels” to either “channels” or “canals”

Modified as pointed out.

Round 2

Reviewer 3 Report

Thank you for responding to my previous comments. The newly added Google Earth figures are very informative and help the reader understand the relative size of the MODIS pixel to the paddy fields and therefore the associated limitations of using MODIS over these small field areas.