A Remote-Sensing-Assisted Estimation of Water Use in Rice Paddy Fields: A Study on Lis Valley, Portugal

Round 1

Reviewer 1 Report

This article deals a very interesting subject, the possibility of estimating ET_a in rice paddies from series of satellite NDVI images, obtaining a process that can be applied extensively to improve irrigation and crop management. The study was conducted in Lis Valley Irrigation District (LVID), Portugal, for three growing station, from 2019 to 2021. The experiments focused on a paddy field (P1), and three additional rice plots (P2, P3 and P4), adjacent to P1, were used for validation.

Two Remote Sensing (RS) methods were used for estimating ET_a in rice: 1.-METRIC and 2.-The RS-assisted FAO56 method, a method calculating rice ET_a using the single crop coefficient (K_c) (ET_a = K_c x ET_o; FAO), which is obtained with the onset of K_c(b)-NDVI. Then, the results of both methods were compared with ET_a Field Measurements. Several Landsat 8 and Sentinel 2 images were used for calculating the NVDI for each station and year, from May to October, and the platform SPIDER was used for the management of satellite images datasets.

Average seasonal ET_a (FAO56) resulted 586 ± 22.6 mm and Water Productivity (WP) 0.47 ± 0.03 kg m^-3. Good correlations were found between K_c proposed by FAO and the NDVI evolution in control rice fields. The results obtained from both RS methods show an average estimation error of ± 0.8 mm d^-1 with negligible bias, and R²=0.56. NDVI obtained from METRIC reflects a slight overestimation when compared to SPIDER and field measurements. The conclusion is the potential of the RS-assisted method to monitor ET_a and WP, capturing the local and seasonal variability in rice growing, being a free tool available to farmers.

A lot of work has been made in the field and with the images processing, and the materials, methods and results are presented correctly. Also, the results obtained from the RS-assisted method are similar to those obtained from more complex methods, as that of METRIC. For all it, I recommend this article for publication.

However, I think that some changes and the inclusion of some figure more would improve the article. These minor changes are:

1.   The coefficient of determination is R², not r², please, change this in line 20, Table 6, line 612, lines 481-482, etc. 

2.   I do not know if the second part of the Equation (2) is joined to the first part and some is loss, or if both parts are separated. Please, clarify. What is 0.3? Do you mean x^0.3? And what is 0. In Equation (3)? Note also the size of the number of the Equation (3).

3.   I think that the satellites images and its processing require an own section, maybe, Section 2.3, before to the RS methods. You say that there are 21 Landsat-8 images, and how many Sentinel-2 images? But, after, it seems that the cloud-free images Landsat-8 (and Sentinel-2?) are reduced. Please, explain this in detail. Also, it will be useful to include a table with the dates of acquisition of the images and the NDVI obtained from each one of them. Some very important to clarify is the correction applied on the images, specially the atmospheric correction. So, it should be indicated if the reflectance obtained of the images is the BOA (Bottom of Atmosphere) reflectance (Level 2A in Sentinel 2 images). It is necessary to correct the images for an adequate calculus of the NDVI and for comparing NDVI-Landsat and MDVI-Sentinel2.

4.   Line 400: In P1, this decrease was not visible…? I see that the decreasing of P1 is even higher than that of the others plots….

5.   Figure 4: I would locate the 3 figures in vertical position, to do the figures bigger than now.   

6.   Lines 475-482: I would like that the correction K_c-NDVI were shown graphically in a figure.

7.   Figure 6. Idem that comment 5.

8.   Line 470: Have you used the acronym DAS before in the text?

Author Response

1. The coefficient of determination is R², not r², please, change this in line 20, Table 6, line 612, lines 481-482, etc. Modified, as suggested.
2. I do not know if the second part of the Equation (2) is joined to the first part and some is loss, or if both parts are separated. Please, clarify. What is 0.3? Do you mean x^0.3? And what is 0. In Equation (3)? Note also the size of the number of the Equation (3). Yes, the equation syntax should be together and not separated. Fixed, as well as font size.
3. I think that the satellites images and its processing require an own section, maybe, Section 2.3, before to the RS methods. You say that there are 21 Landsat-8 images, and how many Sentinel-2 images? But, after, it seems that the cloud-free images Landsat-8 (and Sentinel-2?) are reduced. Please, explain this in detail. Also, it will be useful to include a table with the dates of acquisition of the images and the NDVI obtained from each one of them. Some very important to clarify is the correction applied on the images, specially the atmospheric correction. So, it should be indicated if the reflectance obtained of the images is the BOA (Bottom of Atmosphere) reflectance (Level 2A in Sentinel 2 images). It is necessary to correct the images for an adequate calculus of the NDVI and for comparing NDVI-Landsat and MDVI-Sentinel2. Thanks to this referee comment, a new Table 4 was inserted, with the acquisition dates of L8 images + season crop growth. Regarding the atmospheric correction, we inserted some considerations in Section 2.4, as suggested: "Orthorectified Surface Reflectance (Bottom-Of-Atmosphere: BOA) imagery were considered, i.e., the images are atmospherically corrected in order to eliminate or compensate the effects of atmospheric elements on the image thus obtaining a comparable surface signal for areas and different acquisition dates".
4. Line 400: In P1, this decrease was not visible…? I see that the decreasing of P1 is even higher than that of the others plots…. Text revised
5. Figure 4: I would locate the 3 figures in vertical position, to do the figures bigger than now. Modified, as suggested.
6. Lines 475-482: I would like that the correction K_c-NDVI were shown graphically in a figure. We understand and appreciate this referee suggestion, but the authors think there are already too many figures in this paper, and these Kc-NDVI plots are superfluous beyond the regression parameters listed in Table 6. Honestly, we prefer to maintain this Table.
7. Figure 6. Idem that comment 5. Modified, as suggested
8. Line 470: Have you used the acronym DAS before in the text? Yes, in Table 3 (above Line 255)

Reviewer 2 Report

Title: Remote Sensing assisted estimation of water use in rice paddy fields. Study on Lis Valley, Portugal

I have read the document and found the following general and specific comments

In the abstract section: line 19. The short form “WP” and “Kc” used for the first time without stating the full term.

Keywords could be written in alphabetical order

Introduction: requires improvement the problem is not clearly identified, the ideas are not written in coherence.

Study area: figure 1 needs grids/coordinates

Line 202: 5 sampling points and 0.5m2 sample size were used for your study; how is the sample size specified, there should be a clear method of selecting the number of sample points.

The date of acquisition, resolution and path and rows of the data need to be specified in the methods section that I couldn’t find it.

For plot level studies Landsat (30m resolution) is not appropriate. Instead of 30m Landsat imagery and sentinel (10m resolution), why don’t you used freely available more fine resolution such as Planet scope (3m resolution)

In line 430; it is indicated that only few cloud free images are available for the study area, so how you used the images without pre-processing such as atmospheric correction.

The calibration is done which is good, but the method of calibration should be discussed in the methods section. On the other hand, only coefficient of determination (r^2) was used as performance evaluation criteria during calibration which only indicates the pattern instead variations in magnitude can be expressed using additional parameters such as RMSE, PBias, and others.

The conclusion is not well written.

Author Response

1. In the abstract section: line 19. The short form “WP” and “Kc” used for the first time without stating the full term. The full term was introduced, as suggested
2. Keywords could be written in alphabetical order. Modified, as suggested.
3. Introduction: requires improvement the problem is not clearly identified, the ideas are not written in coherence. The last paragraph of introduction was reformulated, in order to identify the problem clearly and the novelty of this work: "The main objective of this study was to derive an approach to determine ET_a in rice paddies from temporal series of satellite NDVI images, provided by SPIDERwebGIS©, from University of Castilla-La Mancha, Spain, making possible to obtain a process that can be applied extensively to improve irrigation and crop management. An assessment of this approach was conducted comparing with ET_a results from METRIC in the study site, on Lis Valley. To our knowledge, this is the first report of a study in which the ET_a provided by the METRIC platform is compared with the ET_a calculated by FAO56 methodology, concluding by the success of this tool in rice. Therefore, this paper also intends to demonstrate that both tools are reliable and allow to consult data provided from RS, in an accessible and user-friendly way".
4. Study area: figure 1 needs grids/coordinates. Modified, as suggested.
5. Line 202: 5 sampling points and 0.5m² sample size were used for your study; how is the sample size specified, there should be a clear method of selecting the number of sample points. A reference was added (Gonçalves et al. (2022), to explain/justify the chosen method, and the sentence was reformulated: “To obtain rice yield, 5 sampling points were chosen to spatially represent the entire plot P1, crosswise, from which a 0.5 m² (100 cm x 50 cm) sample was taken, following the methodology used by Gonçalves et al. [50] (Figure 2c), in a procedure that is commonly used in rice experimentation in Portugal”
6. The date of acquisition, resolution and path and rows of the data need to be specified in the methods section that I couldn’t find it. A new Table 4 was inserted, with the acquisition date of L8 images + season crop growth. Honestly, the authors do not consider necessary to list all dates for the Sentinel images since this is redundant with information plotted in Figure 4. The resolution is mentioned in Sections 2.3 and 2.4 for L-8 and S2, respectively. A new sentence was added in 2.4: “Path/row for Landsat is 204/32 and for Sentinel-2 is “TILE29TNE”.
7. For plot level studies Landsat (30m resolution) is not appropriate. Instead of 30m Landsat imagery and sentinel (10m resolution), why don’t you used freely available more fine resolution such as Planet scope (3m resolution). We understand this referee concern, and we do agree spatial resolution provided by Planet scope might improve the potential not only of this technique, but also of many other remote sensing approaches for agricultural applications. However, there are several reasons driving the selection of Sentinel+Landsat imagery in this work. First, the pixel size of both Sentinel 2 and Landsat is sufficient to monitor our rice fields of several hectares; note rice fields are typically large enough to fit several sentinel-2 or Landsat pixels with no trouble. Secondly, EEFLUX is based on Landsat imagery since thermal information is required to run METRIC code and Planet scope lacks thermal bands. Also, surface reflectance information provided by Sentinel-Landsat constellation has been long tested, calibrated and validated, and has a robustness Planet scope still needs to demonstrate. Moreover, Landsat legacy allows to monitor our fields decades back in time. To sum up, we do agree integration of higher resolution satellites, such as Planet scope, could improve this paper, and will be explored in further works. However, our proposal of using Sentinel-Landsat capacities is perfectly valid in the framework of the aims of the present study.

1. In line 430; it is indicated that only few cloud free images are available for the study area, so how you used the images without pre-processing such as atmospheric correction.  This has been clarified in Section 2.4: “Orthorectified Surface Reflectance (Bottom-Of-Atmosphere: BOA) imagery were considered, i.e., the images are atmospherically corrected in order to eliminate or compensate the effects of atmospheric elements on the image thus obtaining a comparable surface signal for areas and different acquisition dates.”.

9. The calibration is done which is good, but the method of calibration should be discussed in the methods section. On the other hand, only coefficient of determination (r^2) was used as performance evaluation criteria during calibration which only indicates the pattern instead variations in magnitude can be expressed using additional parameters such as RMSE, PBias, and others.
10. The conclusion is not well written. Some considerations were added in conclusion section: “Some final considerations about this study may guide further work, namely: 1) the characteristic cloudiness in the study site limited the acquisition of satellite images. Also, depending on the field size, Landsat spatial resolution might not be appropriate. Further studies should integrate higher resolution platforms, such as Planet scope; 2) NDVI saturates for dense full vegetated covers. Further studies may explore the use of additional VI’s; 3) it will be interesting to expand the assessment of the calibrated Kc = Kc(NDVI) equation to other rice production areas with a similar agronomic management”.

Reviewer 3 Report

Manuscript, entitled (Remote Sensing assisted estimation of water use in rice paddy fields. Study on Lis Valley, Portugal.  The results showed that good correlations were found between Kc proposed by FAO and the NDVI evolution in control rice fields, ranging r² between 0.71 and 0.82 for stages II+III, and between 0.76 and 0.82 for stage IV. Results from the derived RS-assisted method were compared to ET a values obtained from the surface energy balance model METRIC, showing an average estimation error of ±0.8 mm d⁻¹ with negligible bias. There are some comments that should be included in order to enhance the manuscript for the readers.

Title

·         The title should be improved. Two sentences should be combined as one sentence.

Abstract

·         Line 19. Please write the full name of WP?

·         Line 20. r² should be modified to coefficient of determination (R²).

·         Line 20. Please write the name of growth stages of II+III and IV

·         What is the reason for using NDVI since several spectral indices could be used.  The disadvantage of NDVI is that the NDVI has saturation affected with big amount of biomass and also is affected by soil reflectance.

Keywords: should be arranged alphabetic

Introduction

·         Line 45. More citations should be added at the end of the sentence.

·         Line 52. Citations should be added at the end of the sentence.

·         Please highlight in introduction, what is the novelty (originality) of the work? And what is new in your work that makes a difference in the body of knowledge?

Minor editing of English language required

Author Response

Title

1. The title should be improved. Two sentences should be combined as one sentence. Combined, as suggested.

Abstract

2. Line 19. Please write the full name of WP? Revised, as suggested.
3. Line 20. r² should be modified to coefficient of determination (R²). Modified, as suggested.
4. Line 20. Please write the name of growth stages of II+III and IV. Introduced, as suggested.
5. What is the reason for using NDVI since several spectral indices could be used.  The disadvantage of NDVI is that the NDVI has saturation affected with big amount of biomass and also is affected by soil reflectance. The referee is right, the NDVI may have some limitations to capture variability under very dense and vegetated coverages compared to other vegetation indices (for instances this is an issue in forested areas), however, note we aim to monitor all the phenological phases of the crop and then cover a wide range of NDVI values (0.2-0.9). Under these conditions, the NDVI has previously shown a good performance to characterize crop coefficients including crops such a maize, cereals, alfalfa, etc. The main reason to choose the NDVI is then to give continuity to previous works in other crops reported in the literature. But also, both tools used in this work, EEFLUX and SPIDERwebGIS, are based and provide temporal series of NDVI data. However, these critical aspects are mentioned in the introduction, where we add some new references to reinforce NDVI limitation: Xing et al. (2020), Guan et al. (2019) and González-Betancourt et al. (2018).

Keywords: should be arranged alphabetic. Modified, as suggested.

Introduction

5. Line 45. More citations should be added at the end of the sentence. Following this referee comment, additional citations have been added as suggested (Kaspary et al. (2020); Chen et al (2013); De Bauw et al. (2019)).
6. Line 52. Citations should be added at the end of the sentence. Following this referee comment, additional citations have been added as suggested (Zampieri et al. (2023); Ringler et al. (2015))
7. Line 95. Please do not use the pronouns in the scientific writing such as (we also share)? Pronouns were supressed, as suggested
8. Line 106 to line 118. More citations should be added. Following this referee comment, additional citations have been added as suggested (Xue et al. (2017); Imran et al. (2020); Zheng et al. (2018); Rabatel et al. (2014)).
9. Please highlight in introduction, what is the novelty (originality) of the work? And what is new in your work that makes a difference in the body of knowledge? The last paragraph of introduction was reformulated, in order to highlight the novelty of this work: “The main objective of this study was to derive an approach to determine ETa in rice paddies from temporal series of satellite NDVI images, provided by SPIDERwebGIS©, from University of Castilla-La Mancha, Spain, making possible to obtain a process that can be applied extensively to improve irrigation and crop management. An assessment of this approach was conducted comparing with ETa results from METRIC in the study site, on Lis Valley. To our knowledge, this is the first report of a study in which the ETa provided by the METRIC platform is compared with the ETa calculated by FAO56 methodology, concluding by the success of this tool in rice. Therefore, this paper also intends to demonstrate that both tools are reliable and allow to consult data provided from RS, in an accessible and user-friendly way”. This idea was also reinforced in the text body, and in the conclusion.
10. Materials and methods   
11. Line 158. Please write the full name of PI? Added, as suggested
12. Please write the field capacity and wilting point of the soil?  Thank you for this suggestion. This information was added in section 2.1: “Field capacity is 0.385 cm³ cm^-3 and wilting point is 0.204 cm³ cm^-3”.

Results

13. The resolution of Figure 4 should be improved. Figures have been resized to improve resolution, as suggested.       
14. Please add the significant levels of R2 of table 6?  We are not sure what the referee means. R2 ranges between 0 and 1, depending on the data scatter, but we think this is superfluous information. Could you please explain what do you exactly refer to? ·Thank you.         
15. The resolution of Figure 6 should be improved.   Figures have been resized to improve resolution, as suggested.       

Discussion

16. The discussion is well presented.  Thank you.

Conclusions     

 Please write about the limitations of this work in details in conclusion section. Some considerations were added in conclusion section: “Some final considerations about this study may guide further work, namely: 1) the characteristic cloudiness in the study site limited the acquisition of satellite images. Also, depending on the field size, Landsat spatial resolution might not be appropriate. Further studies should integrate higher resolution platforms, such as Planet scope; 2) NDVI saturates for dense full vegetated covers. Further studies may explore the use of additional VI’s; 3) it will be interesting to expand the assessment of the calibrated Kc = Kc(NDVI) equation to other rice production areas with a similar agronomic management”.

Round 2

Reviewer 3 Report

The authors did significant improvment acording to my comments