Hybrid Methodology for the Estimation of Crop Coefficients Based on Satellite Imagery and Ground-Based Measurements
Round 1
Reviewer 1 Report
Two main suggestions: 1. Would be helpful to discuss the footprint difference of spectroscopy and landsat 30X30 2. Maybe add validation of metric ET with field ET 3. Some discussions on whether the findings are site dependent which in turn will impact the application See other comments in the pdf
Comments for author File: 
Comments.pdf
Author Response
Response to reviewer#1 comments
Manuscript ID: water-489250
Type of manuscript: Article
Title: Hybrid methodology for the estimation of crop coefficients based on satellite imagery and ground based measurements
Authors: Marios Spiliotopoulos *, Athanasios Loukas
The authors thank the reviewer for his/her constructive and useful comments. All comments made by the reviewer have been addressed. Additions have been done in the revised annotated paper in read and our replies to reviewer’s comments are presented in detail in the next paragraphs.
REVIEWER’S COMMENTS AND SUGGESTIONS FOR AUTHORS
1. Would be helpful to discuss the footprint difference of spectroscopy and Landsat 30X30.
AUTHORS REPLY:
The comment of the reviewer has been addressed (see lines 183-187 of the revised paper). According to GER 1500, Spectra Vista Corporation, there are 512 spectral bands from 350-1050 nm regarding the spectroradiometer. It is reminded that to compute vegetation indices with Landsat ETM+, band 1 (0.441-0.514 μm), band 2 (519-601 μm),band 3 (0.631-0.692 μm) and band 4 (0.792-0.878 μm) are mainly utilized. The spectral resolution on broadband Landsat TM/ETM+ is indicated by the relative spectral response (RSR) of the spectral bands. The goal of the described conversion is the calculation of surface reflectance values which are equivalent to Landsat ETM+ bands 1, 2, 3 and 4. The filtering of the data through the RSR values of Landsat ETM+, is made through the interpolation of the initial GER1500 reflectance values (Fig.3 at the manuscript). The result is the generation of reflectance values at the incremental wavelength of the RSR (at 450, 451, 452 nm etc.). This procedure (Figure 1) is necessary since the GER1500 reflectance values are given at a different incremental wavelength scale (e.g. 449.81, 451.48, 453.15 nm) (Papadavid et al., 2012).
Figure 1. Interpolation of the average initial reflectance values.
As described in the manuscript, an average of twenty (20) separate in-situ measurements for each experiment day was implemented for each crop. Those in-situ measurements are point measurements and are assigned to specific 30mx30m pixels recorded from ETM+, which are representative of 100% coverage of a specific crop. In every other case the information taken from those pixels would have a “mixed” behavior and that is not desirable. GER 1500 experimental data are finally filtered then through the RSR functions and averaged within the limits of the first four TM/ETM+ bands, to obtain the in-band reflectance values.
2. Maybe add validation of metric ET with field ET.
AUTHORS REPLY:
The comment of the reviewer has been addressed in the revised version of the paper (Page 13-14:lines 292-343 and Figures 9,10 and 11). Validation of METRIC with field ET remains one of the main shortcomings of the manuscript. Unfortunately, there are no reliable field ET measurements (lysimeters etc.) at the region. For that reason, a validation between METRIC ETrF and the proposed model ETrF has been implemented. Specifically, we used a portion of those in-situ spectroradiometer data for the development of the relationships and, then, the remaining data as validation data for the purpose of model evaluation, with satisfactory results. More internal validation has been addressed at the text (Pages 14 and 15).
3. Some discussions on whether the findings are site dependent which in turn will impact the application.
AUTHORS REPLY:
The comment of the reviewer has been addressed in the section Conclusions of the revised paper (lines 399-407). The results confirm that the methodology developed for the estimation of crop coefficients from field (in situ) measurements with spectro-radiometer is simple, validated, and reliable for the assessment of crop water requirements at a typical Mediterranean agricultural region. Furthermore, the developed equations between ETrF and VIs, and especially NDVI, may be used in other areas as well. Then, the proposed developed methodology, which is a Kc – NDVI approach, may be applied elsewhere, along with satellite derived NDVI and assuming crop homogeneity. On the other hand, METRIC methodology, using the “hot” and “cold” pixels for finding the anchor values, is a site dependent methodology, because those pixels are generally different for every image and location.
OTHER COMMENTS
Define ETa
The comment of the reviewer has been addressed in the revised version of the paper (lines 90-93 and 144-145). Now the definition of ETa is clarified and properly presented.
Would be helpful to show the weather station on the map in Figure 2
The comment of the reviewer has been addressed in the revised version of the paper (Page 5, Figure 2). The location of the weather station is shown in the revised Figure 2.
Would be helpful to show the plots in addition to Table 2
The comment of the reviewer has been addressed in the revised version of the paper. The plots of the relationships are now presented in Figure 7 (Page 12).
Some discussions on whether the findings are site dependent which in turn will impact further application
This comment is the same with main comment #3 which has been discussed above. The comment of the reviewer has been addressed in the revised version of Conclusions part (lines 399-407).
Author Response File: 
Author Response.docx
Reviewer 2 Report
In this paper, the authors have proposed a model for estimation of crops Evapotranspiration from vegetation indexes. They have used both in-situ measurements and remote sensing data for this modelling. A reasonable effort on data collection and analysis have been accomplished. The results have been relatively well analyzed and presented. Nonetheless, some critical issues in the manuscript should be considered and revised.
̶ Using a simple regression model which has only a single input is a drawback for this study and paper, which is a great model for this research. There are more efficient and flexible algorithms in the literature compared to the used simple regression in this research. For example, machine learning and artificially based approaches such as Random Forest and Support Vector regressions, and neural network modelling have been demonstrated very successfully for crops parameter modelling.
̶ I would recommend adding more descriptions and explanations about “crop coefficients” from the beginning of the paper. As far as I know, it is not a common term in this literature.
̶ What were the reason and motivation of the authors to use only these three vegetation indexes? Pimpale et al. (2015) have used some other VIs for wheat.
Pimpale, A.R., Rajankar, P.B., Wadatkar, S.B., Wanjari, S.S., Ramteke, I.K., 2015. Estimation of water requirement of wheat using multispectral vegetation indices 208–212.
̶ As the authors have mentioned, they did not use any in-situ observations to validate this modelling. Accordingly, these results are not statistically reliable. I would recommend using a portion of those in-situ samples as training data, and the rest as testing data for the purpose of model assessment and evaluation.
̶ Since the proposed modelling is based on the results from the METRIC model, the results of your simple linear modelling are also affected by the METRIC model. I think it would be more logic to make a statistical comparison in the “Validation of the Methodology” for the evaluation of the METRIC and the proposed models.
̶ The modelling of crop parameters has been done based on the in-situ spectral measurements. However, the ultimate goal of such modelling is to use remote sensing data. As a result, you would need to compare the Vis extracted from Landsat data and used in the METRIC model, with the Vis extracted from in-situ measurements by applying the temporal interpolation between RS and in-situ observations.
̶ I think more statistical criteria, such as RMSE, MAE, CV, would help better to analyze the results.
̶ In lines 279-290, the authors have stated that they have used the CROPWAT model to calculate the average water needs or irrigation requirements for crops. In this model, the crop coefficients were used for the evaluation propose. However, the other CROPWAT model’s inputs were not introduced. Accordingly, I think these inputs should be introduced and explained in the manuscript.
̶ The paper needs final proofreading for its English; there are some minor grammatical issues in the manuscript.
Based on my evaluations and the above comments, I think this paper can be accepted after addressing these modifications and providing the answer to the questions.
Author Response
Response to reviewer#2 comments
Manuscript ID: water-489250
Type of manuscript: Article
Title: Hybrid methodology for the estimation of crop coefficients based on satellite imagery and ground based measurements
Authors: Marios Spiliotopoulos *, Athanasios Loukas
The authors thank the reviewer for his/her constructive and useful comments. All comments made by the reviewer have been addressed. Additions have been done in the revised annotated paper in read and our replies to reviewer’s comments are presented in detail in the next paragraphs.
REVIEWER’S COMMENTS AND SUGGESTIONS FOR AUTHORS
In this paper, the authors have proposed a model for estimation of crops Evapotranspiration from vegetation indexes. They have used both in-situ measurements and remote sensing data for this modelling. A reasonable effort on data collection and analysis have been accomplished. The results have been relatively well analyzed and presented. Nonetheless, some critical issues in the manuscript should be considered and revised.
1. Using a simple regression model which has only a single input is a drawback for this study and paper, which is a great model for this research. There are more efficient and flexible algorithms in the literature compared to the used simple regression in this research. For example, machine learning and artificially based approaches such as Random Forest and Support Vector regressions, and neural network modelling have been demonstrated very successfully for crops parameter modelling.
AUTHORS REPLY:
Thank you very much for the comment. Our scope was to offer an alternative way of estimation of crop coefficients combining in-situ spectroradiometers and simple VI relationships. For that reason, there was an approach to compare our methodology with simple linear regression methodologies applied earlier in the literature (Jackson et al., 1980; Sellers, 1985; Choudhury et al., 1994; Jayanthi and Neale 2000; Irmak and Kamble 2009). However, we have tested other non-linear formulas (e.g. exponential, polynomial etc.) but their results were not as satisfactory as the results of linear relationships. Nevertheless, we are planning to work with Random Forest and/or Support Vector regressions or NNM in the near future.
2. I would recommend adding more descriptions and explanations about “crop coefficients” from the beginning of the paper. As far as I know, it is not a common term in this literature.
AUTHORS REPLY:
The comment of the reviewer has been addressed in the revised version of the paper (Page 2, lines 46-55).
3. What were the reason and motivation of the authors to use only these three vegetation indexes? Pimpale et al. (2015) have used some other VIs for wheat.
Pimpale, A.R., Rajankar, P.B., Wadatkar, S.B., Wanjari, S.S., Ramteke, I.K., 2015. Estimation of water requirement of wheat using multispectral vegetation indices 208–212.
AUTHORS REPLY:
The idea for choosing the particular three (3) Vegetation Indices for our work brought out from the literature as well from previous experience working with those indices. We have already mentioned the study of Pimpale et al., 2015 as [18] in our paper. Indeed, Pimpale et al. tested RVI, NDVI, TNDVI, SAVI and MSAVI2 for wheat and found that NDVI-Kc model gave more accurate results than the other Kc-VI models. That was another reason for choosing NDVI as the best VI for processing in our work.
Pimpale et al. also suggest linear regression analysis which is in agreement with our study (we have replied to a previous comment).
4. As the authors have mentioned, they did not use any in-situ observations to validate this modelling. Accordingly, these results are not statistically reliable. I would recommend using a portion of those in-situ samples as training data, and the rest as testing data for the purpose of model assessment and evaluation.
AUTHORS REPLY:
Thank you for that comment. The comment of the reviewer has been addressed in the revised version of the paper (Pages 13-14: lines 292-317 and Figures 9 and 10). We have used a part of in-situ samples as training data and the remaining data as testing data as suggested. That procedure is described in the revised version of “Validation of the model” part of the paper and the results are very satisfactory.
5. Since the proposed modelling is based on the results from the METRIC model, the results of your simple linear modelling are also affected by the METRIC model. I think it would be more logic to make a statistical comparison in the “Validation of the Methodology” for the evaluation of the METRIC and the proposed models.
AUTHORS REPLY:
Thank you for that comment. The comment of the reviewer has been addressed in the revised version of the paper (Page 14: lines 310-317 and Figure 11). We have now added a statistical comparison between METRIC ETrF and the respective ETrF values estimated by the proposed methodology at the section “Validation of The Methodology” of the revised paper (Figure 11) as suggested by the reviewer. No, significant correlation has been found (R2=0.14), but the average values between the two cases are encouraging (METRIC ETrF average value=0.66 vs Modeled ETrF average value=0.63). The other statistics criteria are also relatively low (RMSE=0.24, MAE=0.16, CV=0.35).
6. The modelling of crop parameters has been done based on the in-situ spectral measurements. However, the ultimate goal of such modelling is to use remote sensing data. As a result, you would need to compare the VIs extracted from Landsat data and used in the METRIC model, with the Vis extracted from in-situ measurements by applying the temporal interpolation between RS and in-situ observations.
AUTHORS REPLY:
Thank you for that comment. The comment of the reviewer has been addressed in the revised version of the paper (Page 12: lines 280-285 and Figure 8). According to your comment, after interpolation, a comparison of Landsat based NDVI and in-situ based NDVI values has been implemented at the beginning of revised version of “Validation of The Methodology” Part. No significant differences have been found between the two sources (R² = 0.82, RMSE=0.09, MAE=0.07, CV=0.23).
7. I think more statistical criteria, such as RMSE, MAE, CV, would help better to analyze the results.
AUTHORS REPLY:
The comment of the reviewer has been addressed in the revised version of the paper. Table 2 is fully reconstructed and additional statistical criteria (e.g. RMSE, MAE and CV) have been used as suggested. The statistical validity of the results has been proven and for the additional statistical criteria. The specific set of criteria has been used throughout the text.
8. In lines 279-290, the authors have stated that they have used the CROPWAT model to calculate the average water needs or irrigation requirements for crops. In this model, the crop coefficients were used for the evaluation propose. However, the other CROPWAT model’s inputs were not introduced. Accordingly, I think these inputs should be introduced and explained in the manuscript.
AUTHORS REPLY:
The comment of the reviewer has been addressed in the revised version of the paper (Page 16-17, lines 358-379). CROPWAT model is now properly introduced and the basic parameters and requirements are introduced and presented in the text.
9. The paper needs final proofreading for its English; there are some minor grammatical issues in the manuscript.
AUTHORS REPLY:
The comment of the reviewer has been addressed in the revised version of the paper. English language has been extensively revised throughout the text according to the comment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
The authors revised the manuscript and answered my questions. It is now in good shape for publication.
Reviewer 2 Report
As I reviewed the revised manuscript and the answers of the authors to my questions and comments, I saw they have done a good job on the improvement of the quality of their manuscript. As a result, I recommend the publication of this manuscript in the Water journal.
