Learning Global Evapotranspiration Dataset Corrections from a Water Cycle Closure Supervision

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper is probably the best paper I have reviewed over many years. It's an excellent contribution in improving global E remote sensing estimates especially in areas with systematic errors like sub-tropical and semi arid-regions. I'm confident that it will be a valuable contribution in the years to come. I have only minor observations and I recommend it for publication.

Lines 27-29. You may refer global remote sensing products such as (Yuan et al. 2010)

Line 36-39: Correct observation on the difficult to PM and the necessity of temperature-base model. An example on how this approach can be combined with remote monthly temperature to provide Global PET is presented by Tegos et al. 2022.

Line 43-48: Can you please clarify the reasons of systematic errors in the existing global E remote sensing models.

Line 50: Can you clarify what is the WB?

Chapters 2.1 and 2.1. I would add a discussion of the reliability of the selected  remote sensing products used within the study such as GLEAM etc. There is a discussion/analysis in chapter 3.2 however it will benefit the paper if you include an analysis herein.

Lines 486-448: Fully agree on this observation on how the parameterization of the PET can impact the E in sub-tropical areas.

General comment for consideration: Would the authors provide free access to the improved datasets?

References

1. Yuan, W.; Liu, S.; Yu, G.; Bonnefond, J.-M.; Chen, J.; Davis, K.J.; Desai, A.R.; Goldstein, A.H.; Gianelle, D.; Rossi, F.; et al. Global estimates of evapotranspiration and gross primary production based on MODIS and global meteorology data. Remote Sens. Environ. 2010, 114, 1416–1431. 
2. Tegos, A.; Malamos, N.; Koutsoyiannis, D. RASPOTION—A New Global PET Dataset by Means of Remote Monthly Temperature Data and Parametric Modelling. Hydrology 2022, 9, 32. https://doi.org/10.3390/hydrology9020032
3.  

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript was well organized, and the data processing process and method construction description were well-written and easily accessible. As such, I am recommending it for publication.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

In this manuscript, the research perspective is novel, with the goal of correcting the basin E value after OI processing, Machine Learning method is used to learn the pixel scale errors and aggregate them to the basin scale, solving the problem of no target value at the pixel scale. But there are several issues and doubts:

 

1.       There are many unclear or problematic expressions, such as:

Line 7: About the abbreviation “ET”, It needs to give the full name, when the abbreviation first appears in the text.

Line 94: the abbreviation “WC”.

Line 222: Unification of Variable Terminology: Tskin and Ts.

Line 270: “but this does not mean that a systematic bias is present for the individual basins”. I think the expression is reversed.

Line 280: The expression of G = [1-1-1-1]?

It is needed to refine and modify the above sentences but not limited to these.

 

2.       Questions about datasets：

There is a question: many Precipitation、Water storage and Discharge datasets were used, what are their roles, and what is their processing process?

 

3.       It is suggested that further improve and correct the Figures and Tables of this manuscript, such as:

Table 1: The auxiliary information used in the ML-correction model is inconsistent with the variables in Line 222.

Figure 1: Would it be more appropriate to change positions for 3) and 4)? The OI operator and machine learning operator are both yellow dots, please distinguish them by color.

Most of the Figures require more refined modifications.

 

4.       The scale problem of Machine Learning Operator:

Machine learning method performed well at the basin scale, but its performance at the pixel scale is not clear. Due to the fact that network parameters are trained at the basin scale, the summation process may compensate for some of the problems of Machine Learning at the pixel scale. If there are no issues, a correction map at the pixel scale can be displayed; If there is a problem, why not use Machine Learning directly at the basin scale, instead aggregating the pixel corrections to the basin scale?

Comments on the Quality of English Language

There are many unclear or problematic expressions.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

The authors addressed all my comments, and I recommend for publication at current form.