Machine Learning Techniques for Estimating Hydraulic Properties of the Topsoil across the Zambezi River Basin

Round 1

Reviewer 1 Report

The manuscript shows relatively simple research for soil hydraulic properties estimation with machine learning methods. The analysis is conducted based on the variables collected from various data sources and trained it with some samples. Different methods are applied to train the relationship. Overall, this study is an empirical study with poor physical background. The innovation is limited in current version. The choose of the variables and the reasonability should be well addressed in the methodology sections. The detailed comments are below:

1. In the abstract, the purpose of this study should be clarified and also the variables used to build the estimation model. Ksat is not explained.
2. The introduction section is too long and it is hard for the readers to follow the real aim of this study. It should be extensively shortened to focus on the background of this study and the key contribution of this study.
3. In the manuscript, there are too many places with (Error! Reference source not found.). It is hard for me to follow the manuscript.
4. The hydraulic properties should be well explained in the manuscript in advance and then for following studies.
5. About the datasets, the training dataset should be introduced first and then followed by the two datasets for validation.
6. About the environmental covariates, why there are so many independent variables? It should be explained also.
7. I remember that the importance is derived from the RF method. Is it appropriate for the determination of the potential covariates for the model construction with other methods.?
8. How to get the spatial autocorrelation analysis in section 3.2. It is not clear in text.
9. The uncertainty of the training samples should be discussed in the manuscript. Meanwhile, the uneven distribution of the samples will influence model accuracy.

Author Response

Please see the attachment

Author Response File: Author Response.doc

Reviewer 2 Report

I think the authors carried out a large amount of work at modeling using ML algorithms which were examined by several predictor variables. The paper is generally well-written in an understandable way. I recommend the manuscript for publication after the following major changes:

1. You used multiple datasets with different sources with different temporal resolutions. How did you merge all those datasets into a common time series, please explain? It is necessary to report how the matching is carried out. This can have a significant impact on the results.
2. Your study area has a high dependency on local climate, topographic complexity. Can you provide detailed climatic information for the selected study areas?
3. How did you integrate SCORPAN model with geostatistical approaches? Could you explain more about these issues?
4. You only showed the importance plot (figure 4) for the Random forest algorithm in terms of Sensitivity analysis. How about the other ML techniques and their sensitivity?
5. Could you show some Partial dependent tests for the most important variables?
6. To optimize ML algorithms, authors need to provide significant hyper-parameter information. Can you provide a table?
7. You didn’t explain about model optimization technique. Could you explain these issues?

Author Response

Please see the attachment

Author Response File: Author Response.doc

Reviewer 3 Report

This study is well-designed. This paper is well-written and well-referenced. However, the following moderate to minor corrections may further enhance the readability of this manuscript:

1. One of the keyword duplicates the same as in the paper title (Machine Learning). Another selections should be re-chosen.
2. The automated messages “Error! Reference source not found.” are very disturbing to readers.
3. Figure 1 has not been referred to in the text (until Lines 39-40).
4. Punctuation error in Line 131.
5. Please use “and” instead of “&” throughout the entire text.
6. Acronym names should be defined only once.
7. In Line 252, “Available Water Capacity” needs not be capitalized.
8. The green points are not obvious in Figure 2. Are they all crowded in one location?
9. Need careful checking on the “Units” column in Table 1.
10. What is the definition for “EVI” in Table 1?
11. What is “AIC” in Line 19?
12. The scale for the horizontal axis is too large in Figure 4 to identify the obvious study results.
13. Difficult to associate the proper legend to the right hydraulic property map in Figure 9.
14. Some of the study implications (beginning of the Discussion section) may be shifted to the Conclusion section.
15. No study results have been demonstrated in this paper on the last sentence of the conclusion remarks.

Author Response

Please see the attachment

Author Response File: Author Response.doc

Round 2

Reviewer 1 Report

Thanks for the revision. The manuscript has a big improvement. However, the discussion section about the impact from the samples should be added. Meanwhile, the qualities of the figures should be improved. The frame is not suggested for each figure but replaced with different numbers. The class name should be provided in Figure 3.  Too many spaces are existing in Figure 6-8.  The legend is better to be provided in each sub-figure. 

Author Response

Please see the attachment

Author Response File: Author Response.doc

Reviewer 2 Report

The authors significantly improved the quality of the paper by addressing most of the previous comments. This research work will be very effective for the Water resources community. I recommend the manuscript for publication after minor updates:

1. can you provide a table for  the default meta-parameters  for the ML models such as tree, mtry, node, loss function etc 
2. As you are focusing on your study over the complex terrain, It will be great to introduce in terms of ML-based evaluation in water resources application. Over complex terrain regions, hydraulic/hydrological estimates can be associated with significant error due to variability and uncertainty introduced by orographic effects (Derin et al. 2016; Mei, et al. 2016, Khan et al. 2021, and so on). The authors should include this aspect in the introduction section.

Khan, et al 2021: Artificial Intelligence-Based Techniques for Rainfall Estimation Integrating Multisource Precipitation Datasets. Atmosphere 2021, 12, 1239. 

Derin, et al . 2016 "Evaluation of multiple satellite-based precipitation products over complex topography." Journal of Hydrometeorology 15.4 (2014): 1498-1516.

Mei, et al. 2016: Evaluating satellite precipitation error propagation in runoff simulations of mountainous basins. J. Hydrometeor., 17, 1407–1423, https://doi.org/10.1175/JHM-D-15-0081.1.

Author Response

Please see the attachment

Author Response File: Author Response.doc