Groundwater Potential Mapping Using an Integrated Ensemble of Three Bivariate Statistical Models with Random Forest and Logistic Model Tree Models
Round 1
Reviewer 1 Report
This manuscript presents a study for groundwater potential mapping by linking wells and some “effective factors” through statistical approaches. Generally, this manuscript is well organized, and the results are clearly presented. The results are also interesting to groundwater resource management communities. However, I have some concerns about the method, and some comments about the results and discussion. I would recommend major revision. My concerns and comments are listed below:
Major:
1. The authors used the wells distribution data for training and validation, and then based on the model results, made conclusions for groundwater potential mapping. This is built on an assumption that the wells used in this study are representative of the groundwater potential. However, this was not validated in the manuscript. In fact, the wells are unnecessarily located at places with highest groundwater potential. In addition, the well locations can be impacted by many other factors, such as population distribution. Without addressing this problem, the results and conclusion of this work are untenable.
2. Based on Table 2, the results for FR and CF are almost the same. In other words, if you rank the classes based on FR or CF values, the results are almost the same. However, the data mining results and conclusion show that CF performed better that FR. Can you add discussion explaining why FR and CF are so similar? Also, can you figure out the reason that why such similar results lead to different performances when combined with RF/LMT methods?
3. It seems that the results for LMT is missing for section 4.2. Can you add the text and associated figure (similar to Figure 4) for LMT?
4. In Eq. (2), what does the gradient of the slope mean? How was it calculated?
5. In Eq. (3), what is Npix(Xj)? Can you add more information for definition of “class” and “criterion”?
6. In Eq. (4), what is “previous likelihood of wells”? How were pps and ppa calculated?
7. Lines 207-208, “The rank of belief and the rank of plausibility are the key part of this concept, so the rank of plausibility is greater than or equal to the rank of belief.” Why?
8. In Eq. (5)-(10), what do T, D, m, i, j and N(…) mean?
9. In lines 269-270, “High altitude areas have higher runoffs, while lower altitude areas are more permeable”, can you explain it?
10. Lines 338-340, “The higher accuracy of EBF and CF models than FR model is due to the uncertainty regarding the occurrence of groundwater in the results.” What does it mean? Can you add more information to explain it?
11. In Figure 6, what do x and y axes stand for? Can you add more information for them, as well as how to interpret the curves?
12. Lines 385-387, “According to the result, the combination of bivariate statistical models with data mining algorithms has increased the accuracy of groundwater potential map”. There is no results to support this statement
Minor:
1. Lines 19 and 85, change “m3/h“ to “m3/h“
2. Line 147 and Figure 3i and 3k, change “km/km2“ to “km/km2 “
3. Please change lines 19-20 to “238 wells (70%) were used for model training, and 101 wells (30%) were used for model validation.”
4. Figure 1, please change “DEM” to “Elevation”.
5. What does TWI mean? Can you add a full name for it?
6. In line 262, why is “This curve consists of two horizontal graphs (x axis)…”
7. Line 286, should the “form” be “from”?
8. Line 293, change “03.0” to “0.03”
9. Line 314, please add a full name for WEKA
10. Line 316, should “slope angle” be “slope aspect”?
11. The following sentences need modification:
Lines 51-52, 127, 170, 192-194, 214-215, 252-254, 329-330, 336-338.
Author Response
Dear Prof. Dr. Arjen Y. Hoekstra
Editor-in-Chief
Water
Special issue: Applications of Remote Sensing and GIS in Hydrology II
July, 2019
Re: water-536182, entitled: “Groundwater Potential Mapping Using Integrated Ensemble of Three Bivariate Statistical Models with Random Forest and Logistic Model Tree Models"
We thank both reviewers for providing us highly constructive and insightful comments to improve our manuscript. We have carefully revised the manuscript, following the reviewers’ suggestions, and have responded in detail to each comment. The next section contains our point-by-point responses (in blue) and changes and referencing to the manuscript (in green), based on the reviewers’ comments (in italic). We believe that our manuscript has substantially improved and is more readable for broader audiences. We would like to thank you, Prof. Dr. Arjen Y. Hoekstra (Editor-in-Chief), related Special Issue Editor-in-Chief and Dr Jinxiu Wang, M.S. (Assistant Editor) for following up this manuscript. We look forward to hearing from you. Also, we would be glad to respond to any further questions and comments that you may have.
Note: The figures cannot be shown here. please see Fig. 1, Fig. 5, Fig 30 and Fig. 31 in the attached file.
Yours Sincerely,
Abolghasem Sadeghi-Niaraki,
Assistant Professor
Department of Geomatics Engineering
Sejong University
Seoul
South Korea
Major:
1. The authors used the wells distribution data for training and validation, and then based on the model results, made conclusions for groundwater potential mapping. This is built on an assumption that the wells used in this study are representative of the groundwater potential. However, this was not validated in the manuscript. In fact, the wells are unnecessarily located at places with highest groundwater potential. In addition, the well locations can be impacted by many other factors, such as population distribution. Without addressing this problem, the results and conclusion of this work are untenable.
Response
We appreciate the reviewer for his/her valuable time for reviewing our manuscript and providing us useful comments; they have certainly improved our article.
In order to provide a groundwater potential map, data from wells in the study area has been used. For this reason, wells for modeling and validation have been used, with a yield higher than ≥11 m3/ h and average of pH and electric conductivity (EC) of 6.9 and 495 μmhos/cm. The overall assumption is that wells with a yield higher than ≥11 m3/ h represent potential areas for groundwater, which are consistent with the research [8,59] that have used well data with ≥11 m3/ h. These items were added to section 3.2.
2. Based on Table 2, the results for FR and CF are almost the same. In other words, if you rank the classes based on FR or CF values, the results are almost the same. However, the data mining results and conclusion show that CF performed better that FR. Can you add discussion explaining why FR and CF are so similar? Also, can you figure out the reason that why such similar results lead to different performances when combined with RF/LMT methods?
Response
We thanks for the reviewer’s comment. The weights obtained from the FR model are greater than zero, while the weights obtained from the CF model have values between -1 and 1. The reason for the different results of these two models in the hybrid model is to consider the negative values in the CF model, so that in this model negative values indicate the negative effect of that parameter on the potential of groundwater, but in the FR model this negative effect not specified precisely. These items were added to discussion.
3. It seems that the results for LMT is missing for section 4.2. Can you add the text and associated figure (similar to Figure 4) for LMT?
Response
Thank you for the reviewer’ suggestions. The LMT model does not get the significance of each parameter as a RF model. But in order to provide results for this model, the output trees of this model were added in the article as follow (Figure 5).
Figure 5: Result of LMT model
4. In Eq. (2), what does the gradient of the slope mean? How was it calculated?
Response
Thank you for good comment. In this equation, gradient of the slope is the slope angle in radians. This parameter is based on a DEM and created in the SAGAGIS software.
5. In Eq. (3), what is Npix(Xj)? Can you add more information for definition of “class” and “criterion”?
Response
We appreciate the reviewer’s comment. In this equation, the order of Npix (Xj) is the number of pixels in each class of each criterion. In this research, 15 parameters affecting groundwater have been used, each of these parameters is called a criterion, and each criterion is divided into different categories, each of which is called a class.
6. In Eq. (4), what is “previous likelihood of wells”? How were pps and ppa calculated?
Response
Thanks for pointing out this comment. The "previous likelihood of wells" means the probability of the wells in the study area. In these relationships, ppa is the ratio of the number of well-pixels in a class to the total pixels of that class, and pps is the ratio of the total number of pixel with wells in the study area to the total pixels of the map. These items were added to the section 3.4.2.
7. Lines 207-208, “The rank of belief and the rank of plausibility are the key part of this concept, so the rank of plausibility is greater than or equal to the rank of belief.” Why?
Response
Thanks for good comment. According to equation 10, the value of the plausibility parameter is the sum of the values of belief (5) and uncertainty (9). If the uncertainty is equal to zero, the value of these two parameters is equal to each other, and if the uncertainty values are greater than zero, the belief value is greater than the plausibility.
8. In Eq. (5)-(10), what do T, D, m, I, j and N(…) mean?
Response
We appreciate the reviewer’s comment. In these relations, m represents the number of criteria considered for modeling, i represents each class of each criterion, and j represents each criterion. In these relations, N (T) and N (D) respectively represent the total number of pixels in the study area and the total number of well-pixels in the study area.
9. In lines 269-270, “High altitude areas have higher runoffs, while lower altitude areas are more permeable”, can you explain it?
Response
Thanks for pointing out this comment. High altitude cause a high slope angle. In areas with high altitudes and slopes, rain water does not accumulate in a region and causes the flow of water and runoff. Also, in areas with lower altitudes and slopes, rain water accumulates in an area on the ground and causes penetration into the earth.
10. Lines 338-340, “The higher accuracy of EBF and CF models than FR model is due to the uncertainty regarding the occurrence of groundwater in the results.” What does it mean? Can you add more information to explain it?
Response
We appreciate the reviewer’s comment. In CF and BF models, areas where there is no wells are considered in modeling, while in the FR model, only the areas in which the well is located is used in modeling.
11. In Figure 6, what do x and y axes stand for? Can you add more information for them, as well as how to interpret the curves?
Response
Thanks for good comment. In this diagram, the X axis represents a sensitivity that expresses the prediction value correct in front of all positive outputs, and also the Y axis represents the specificity that represents the predicted negative value correct in front of all negative outputs. The AUC is between zero and one, values less than 0.5 represent model integrity and for models larger than 0.5 has a higher accuracy.
12. Lines 385-387, “According to the result, the combination of bivariate statistical models with data mining algorithms has increased the accuracy of groundwater potential map”. There is no results to support this statement
Response
According to the previous research, the combination of bivariate statistical models with data mining algorithms has increased the accuracy of groundwater potential map [5,58].
Minor:
1. Lines 19 and 85, change “m3/h“ to “m3/h“
Response
The text was changed to m3/h.
2. Line 147 and Figure 3i and 3k, change “km/km2“ to “km/km2 “
Response
Figure3i and 3l were corrected.
3. Please change lines 19-20 to “238 wells (70%) were used for model training, and 101 wells (30%) were used for model validation.”
Response
Thank you for the reviewer’ suggestions. The lines 19-20 have been modified.
4. Figure 1, please change “DEM” to “Elevation”.
Response
Figure 1 is corrected.
5. What does TWI mean? Can you add a full name for it?
Response
The topographic wetness index (TWI) represents topographic control over hydrological processes.
6. In line 262, why is “This curve consists of two horizontal graphs (x axis)…”
Response
Thank you for the reviewer’ suggestions. This diagram consists of a horizontal axis and a vertical axis. In the text of the article, two horizontal axes were changed to a horizontal axis.
7. Line 286, should the “form” be “from”?
Response
In the paper, the word was changed to from.
8. Line 293, change “03.0” to “0.03”
Response
Corrected in the text of the article.
9. Line 314, please add a full name for WEKA
Response
Waikato Environment for Knowledge Analysis (WEKA) was added in line 314.
10. Line 316, should “slope angle” be “slope aspect”?
Response
In line 316, the word was changed to the slope aspect.
11. The following sentences need modification:
Lines 51-52, 127, 170, 192-194, 214-215, 252-254, 329-330, 336-338.
Response
We thanks for the reviewer’ comment. We have improved the paragraph in the revised version of the manuscript.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The manuscript presents the approach of coupling three bivariate statistical models with two machine learning models in order to obtain groundwater potential map of Booshehr plain. There are many applications of the models in the field of groundwater potential mapping however the idea of ensemble of models is quite original. In my opinion the manuscript requires major revision to be published in Water. The Authors should consider to redesign the structure of the manuscript as follows: move section 2, Study area, into 3 Materials and methods; to group subsections 3.3 to 3.7 into one, Models; to reduce number of maps in Fig. 3 or to place some of them in appendix; to place the whole Tab.2 in appendix and present only summary table. The Methodology section should be rewritten to clearly describe the models used. All the equations should be revised and all parameters explained. What is the spatial resolution applied in equations (pixels, classes)? First part of the Discussion section is too general and not related to the paper outcomes but the last paragraph (L380-390) is too short and should be elaborated. The Authors should think if some references can be removed (‘landslide’ appears 10 times).
There are many grammatical errors, unclear expressions and inappropriate terminology. Therefor the manuscript should be revised by the English native speaker familiar with the water related terminology. In many positions on the reference list geographical names are not starting with capital letters.
To further improve the manuscript the Authors should consider the following specific comments:
1. The statement that ‘is free from pollution’ is not always true.
2. Term ‘groundwater potential’ should be explain (defined) when it appears for the first time (L38).
3. L65 The latitude is not correct. Missing 27.
4. Please verify how many steps were used in the study. In L74 ‘four steps’ but in L79 ‘fifth step’.
5. Please be more specific what are the required data.
6. L107 Table 1 should be after Fig. 3n.
7. L110 The usage of topographic maps to create DEM is quite old technique.
8. L112 TWI should be expanded when mentioned first time.
9. L116, L144 Can word ‘penetration’ be replaced with ‘infiltration’?
10. L118 Please explain how were defined the classes in all maps?
11. L127 There are two definitions of ‘alpha’ parameter.
12. L128 Please place the reference to Fig 3d before Fig. 3 e,f (L125) or rearrange the maps. There are more such situations when referenced map is not in order.
13. L130 ‘nutrition’ or ‘recharge’?
14. L133 Provide number of rain stations used and show their locations on Fig. 3l.
15. L140 What is the difference between ‘beta’ and ‘alpha’ from L127?
16. L149 Please remove ‘permeability’.
17. L161 Please remove ‘production of’ .
18. L162 Rewrite the sentence.
19. Unclear what is ‘Mangro’ L163, ‘event of wells’ L210?
20. L211 Why are the equations 7 to 10 presented if they are not used anywhere?
21. L262 The sentence is unclear. Expand TN, FP, TP, FN.
22. L265 In section 4.1 only 4 out of 15 parameters have comments. The rest is just repetition of Tab. 2. The parameters are referenced not in the same order as they appear in table.
23. L302 Columns Bel, Dis, Unc and Pls (Plc) are not discussed anywhere therefor could be removed from the Table 2.
24. L314 Provide reference to the WEKA software.
25. L316, L326 How ‘importance’ is quantified?
26. L328 Please explain how GPMs were prepared (calculated) and what they represent?
27. L332 Red dots (well validation) are not visible. Please explain what are the classes of GPM? What it means that groundwater potential is very low or high?
28. L335 Title of section 4.3 is the same as 4.2.
29. L338 Table 4 should be referenced not 3.
30. L341 Does the second column contain AUC values?
31. Please rewrite the sentence in L383-384.
32. L396 Model names are reversed.
33. L397 ‘confidence’ or ‘certainty’ factor?
34. L400 Did the Authors consider to eliminate the least important parameters form the approach? 15 parameters is quite a lot to analyze. Maybe less parameters could give the same accuracy.
Author Response
Dear Prof. Dr. Arjen Y. Hoekstra
Editor-in-Chief
Water
Special issue: Applications of Remote Sensing and GIS in Hydrology II
July, 2019
Re: water-536182, entitled: “Groundwater Potential Mapping Using Integrated Ensemble of Three Bivariate Statistical Models with Random Forest and Logistic Model Tree Models"
We thank both reviewers for providing us highly constructive and insightful comments to improve our manuscript. We have carefully revised the manuscript, following the reviewers’ suggestions, and have responded in detail to each comment. The next section contains our point-by-point responses (in blue) and changes and referencing to the manuscript (in green), based on the reviewers’ comments (in italic). We believe that our manuscript has substantially improved and is more readable for broader audiences. We would like to thank you, Prof. Dr. Arjen Y. Hoekstra (Editor-in-Chief), related Special Issue Editor-in-Chief and Dr Jinxiu Wang, M.S. (Assistant Editor) for following up this manuscript. We look forward to hearing from you. Also, we would be glad to respond to any further questions and comments that you may have.
Note: The figures cannot be shown here. please see in the attached file.
Yours Sincerely,
Abolghasem Sadeghi-Niaraki,
Assistant Professor
Department of Geomatics Engineering
Sejong University
Seoul
South Korea
The manuscript presents the approach of coupling three bivariate statistical models with two machine learning models in order to obtain groundwater potential map of Booshehr plain. There are many applications of the models in the field of groundwater potential mapping however the idea of ensemble of models is quite original. In my opinion the manuscript requires major revision to be published in Water. The Authors should consider to redesign the structure of the manuscript as follows: move section 2, Study area, into 3 Materials and methods; to group subsections 3.3 to 3.7 into one, Models; to reduce number of maps in Fig. 3 or to place some of them in appendix; to place the whole Tab.2 in appendix and present only summary table. The Methodology section should be rewritten to clearly describe the models used. All the equations should be revised and all parameters explained. What is the spatial resolution applied in equations (pixels, classes)? First part of the Discussion section is too general and not related to the paper outcomes but the last paragraph (L380-390) is too short and should be elaborated. The Authors should think if some references can be removed (‘landslide’ appears 10 times).
We appreciate the reviewer for his/her valuable time for reviewing our manuscript and providing us useful comments; they have certainly improved our article.
Response
We thanks for the reviewer’ comment. The study area section was transferred to the materials and methods section, and the introduction of models was changed to a section called models.
Table 2 was transferred to the appendix and its results were presented in the text of the paper.
The methodology section was rewritten and all parameters and equations were explained.
The spatial resolution applied in equations is pixels.
First part of the Discussion section was summarized.
Last paragraph (L380-390) of Discussion was rewritten.
The authors completely agree with the reviewer’s comment about references (‘landslide’ appears 10 times). But to compare the advantages and disadvantages of these models, these references are necessary.
There are many grammatical errors, unclear expressions and inappropriate terminology. Therefor the manuscript should be revised by the English native speaker familiar with the water related terminology. In many positions on the reference list geographical names are not starting with capital letters.
Response
We thanks for the reviewer’ comment. We have improved the paragraph in the revised version of the manuscript.
To further improve the manuscript the Authors should consider the following specific comments:
1. The statement that ‘is free from pollution’ is not always true.
Response
We thanks for good comment. This sentence was replaced with the phrase "less impact against pollution".
2. Term ‘groundwater potential’ should be explain (defined) when it appears for the first time (L38).
Response
We thanks for the reviewer’ comment. Groundwater potential was explained in introduction as following.
Groundwater storage potential here relates to the maximum amount of permanent storage in aquifers [60].
3. L65 The latitude is not correct. Missing 27.
Response
In line 65, this item was corrected (27°50').
4. Please verify how many steps were used in the study. In L74 ‘four steps’ but in L79 ‘fifth step’.
Response
The research consists of fifth steps, which is changed in line 79.
5. Please be more specific what are the required data.
Response
We thanks for good comment. Required data were added to Material and Method section as following.
Well distribution map and fifteen factors including: altitude, slope angle, slope aspect, plan curvature, profile curvature, slope length, TWI, rainfall, distance from river, distance from fault, drainage density, fault density, lithology, land use and soil were chosen and ready for modeling.
6. L107 Table 1 should be after Fig. 3n.
Response
Corrected in the text of the paper.
7. L110 The usage of topographic maps to create DEM is quite old technique.
Response
The authors completely agree with the reviewer’s comment. The ASTER digital elevation model (DEM) was downloaded in a spatial resolution of 30 30 m (https://gdex.cr.usgs.gov/gdex/).
8. L112 TWI should be expanded when mentioned first time.
Response
In line 112, the TWI's full name was added. (Topographic wetness index (TWI)).
9. L116, L144 Can word ‘penetration’ be replaced with ‘infiltration’?
Response
Thank you for the reviewer’ suggestions. The word ' penetration ' was replaced in the paper with ' infiltration’.
1. L118 Please explain how were defined the classes in all maps?
Response
We thanks for the reviewer’ comment. The classification of all maps was based on the natural break technique as well as the characteristics of the region and previous research. This item was added in line 113-114.
2. L127 There are two definitions of ‘alpha’ parameter.
Response
The main definition of this parameter (α is the slope angle) was corrected in this line.
12. L128 Please place the reference to Fig 3d before Fig. 3 e,f (L125) or rearrange the maps. There are more such situations when referenced map is not in order.
Response
Thank you for the reviewer’ suggestions. The order of the figures was corrected in the text.
13. L130 ‘nutrition’ or ‘recharge’?
Response
This word was changed to 'recharge'.
14. L133 Provide number of rain stations used and show their locations on Fig. 3l.
Response
We thanks for good comment. In the study area, five stations were used to prepare the rainfall map, which was corrected in the text.
15. L140 What is the difference between ‘beta’ and ‘alpha’ from L127?
Response
Thank you for the reviewer’ suggestions. Both parameters represent the slope angle. In order to be identical, the alpha parameter was used in two relationships instead of the beta parameter.
16. L149 Please remove ‘permeability’.
Response
This word was removed in line 149.
17. L161 Please remove ‘production of’ .
Response
This word was removed in line 161.
18. L162 Rewrite the sentence.
Response
Thank you for the reviewer’ suggestions. This sentence was rewritten as following.
The land use map was prepared on a scale of 1:100,000 from the Natural Resources Organization of Booshehr Province.
19. Unclear what is ‘Mangro’ L163, ‘event of wells’ L210?
Response
The term ‘Mangro’ is mangrove forest, which was corrected in line 163. The meaning of ‘event of wells’ is the existence of a well in the region that was altered to ‘occurrence of wells’ in the text.
20. L211 Why are the equations 7 to 10 presented if they are not used anywhere?
Response
These equations were used to calculate the parameters of Unc, Dis, and Pls (Table 2). But in this research, only the Bel weights were used for hybrid models, so the equations 7 to 10 were deleted.
21. L262 The sentence is unclear. Expand TN, FP, TP, FN.
Response
We thanks for good comment. When the actual output is positive and the prediction value is positive, this state is called TP (True positive), and also FN (False negative) represents the state where the actual output is negative and the prediction value is also negative. The TN (True negative) represents the state where the actual output is positive and the prediction value is negative, and the FP (False positive) is the state where the actual output is negative and the prediction value is positive. These indices are derived from the confusion matrix and the ROC curve is calculated on this basis.
22. L265 In section 4.1 only 4 out of 15 parameters have comments. The rest is just repetition of Tab. 2. The parameters are referenced not in the same order as they appear in table.
Response
Thank you for the reviewer’ suggestions.
The class of (-0.4-0.8) had the maximum weight (FR = 1.23, Bel = 0.611 and CF = 0.189) based on the plan curvature factor and had the greatest effect on groundwater incidence. This category of the plan curvature holds more water over a long period of time [6].
The distance from fault and fault density can control the water exchange between the ground and the basement, so the distance closer to the fault can have a positive effect on the occurrence of groundwater [6].
The TWI shows the impact of topography on the location and size of runoff saturation areas [29].
As the slope length increases, the weight of its classes decreases, which indicates that the lower values of this parameter have a greater effect on groundwater.
With regard to the results, less distances to rivers have had a greater impact on groundwater [6]. Drainage density represents the lithology structure of an area and has a significant impact on the identification of groundwater resources [29].
Considering the fact that the study area has a small annual precipitation, more rainfall classes indicate more groundwater occurrence.
According to the lithology factor, the Qft2 class has the highest weight (FR = 1.95, Bel = 0.857 and CF = 0.489). This is because most of the area is comprised of the Qft2 unit
23. L302 Columns Bel, Dis, Unc and Pls (Plc) are not discussed anywhere therefor could be removed from the Table 2.
Response
Thank you for the reviewer’ suggestions. The Bel parameter has been used to create a hybrid model, and its results are presented in Section 4.1, but other parameters such as Dis, Unc, and Pls are ineffective in modeling, and therefore the results of these models were deleted in table 2.
24. L314 Provide reference to the WEKA software.
Response
The following references have been added to the paper.
61. Aburub, F.; Hadi, W. Predicting groundwater areas using data mining techniques: Groundwater in jordan as case study. International Journal of Computer, Electrical, Automation, Control and In 62. Faridi, M.; Verma, S.; Mukherjee, S. Integration of gis, spatial data mining, and fuzzy logic for agricultural intelligence. In Soft computing: Theories and applications, Springer: 2018; pp 171-183.formation Engineering 2016, 10, 1475-1478.
25. L316, L326 How ‘importance’ is quantified?
Response
We thanks for the reviewer’ comment. The RF model measures the importance of a feature by looking at the number of tree nodes that it uses, and reduces impurities throughout the forest tree. The tool scales the score automatically for each feature after training, and measures the results, so the sum of all important values is 1.
26. L328 Please explain how GPMs were prepared (calculated) and what they represent?
Response
We thanks for good comment. After modeling in the WEKA software, the model was generalized to the total pixels of the study area and was calculated in ArcGIS 10.3 software for each pixel, which represents the effect of that pixel on groundwater. Five potential classes (very low, low, moderate, high and very high) were categorized using a natural break technique. A very high potential class that indicates the likelihood that more groundwater will occur in these areas.
27. L332 Red dots (well validation) are not visible. Please explain what are the classes of GPM? What it means that groundwater potential is very low or high?
Response
Thanks for pointing out this comment. The figures were corrected and the validation points were displayed more clearly on the maps.
After generalizing the model to the total pixels in the area, each pixel has a weight that indicates the occurrence of groundwater. In order to describe the numerical data, the natural breaks classifier was used. In this technique, the results were classified as rising in very low potential, low potential, moderate potential, high potential, and very high potential. Very low and low potential classes show a low probability of occurrence of groundwater in these areas, and very high and high potential classes also indicate the probable occurrence of groundwater in these areas.
28. L335 Title of section 4.3 is the same as 4.2.
Response
Title of section 4.3 was changed to Validation of models.
29. L338 Table 4 should be referenced not 3.
Response
Corrected in the text of the article.
30. L341 Does the second column contain AUC values?
Response
The second column of Table 3 represents the AUC values, which were corrected in Table 3.
31. Please rewrite the sentence in L383-384.
Response
Thank you for the reviewer’ suggestions. L383-384 were corrected and rewritten as follows.
EBF and CF models are able to combine the confidence of different sources and are flexible against uncertainty
32. L396 Model names are reversed.
Response
In line 396, these cases were corrected (EBF-RF and FR-RF).
33. L397 ‘confidence’ or ‘certainty’ factor?
Response
The correct expression is the certainty factor. Corrected in line 397.
34. L400 Did the Authors consider to eliminate the least important parameters form the approach? 15 parameters is quite a lot to analyze. Maybe less parameters could give the same accuracy.
Response
Thanks for pointing out this comment.
In the previous research, all these parameters have been considered as an effective parameter. Also, in order to evaluate the parameters, the VIF and TOL indices were used, which showed that all parameters can participate in modeling.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors have addressed most of my comments. I am still not convinced by the responses to comments 7 and 9:
Comment 7: based on the authors' response, there is no causal relationship between the first and second halves of this sentence. It does not make sense. Modification and clarification are needed
Comment 9: the authors' statement in the response is not always true. The runoff is affected by many factors, more than slope, such as land cover/land use, soil types. The statement is misleading. Modification and clarification are needed.
In addition, for Comment 5, did the authors add some information clarifying “class” and “criterion”? If not, please add this info to the text.
Author Response
Dear Prof. Dr. Arjen Y. Hoekstra
Editor-in-Chief
Water
July, 2019
Re: water-536182, entitled: “Groundwater Potential Mapping Using Integrated Ensemble of Three Bivariate Statistical Models with Random Forest and Logistic Model Tree Models"
Special Issue "Applications of Remote Sensing and GIS in Hydrology II"
We would like to express our special thanks to the reviewer#1’ efforts for re-evaluating our manuscript and offering us extremely valuable comments for Round 2. Based on the reviewer#2’ comments, we have again revised the manuscript. The next section contains our point-by-point responses (in blue) and changes and referencing to the manuscript (in green), based on the reviewers’ comments (in italic). We believe that our manuscript has substantially improved and is more readable for broader audiences. We would like to thank Editor-in-Chief (Prof. Dr. Arjen Y. Hoekstra), Section Editor-in-Chief, Special Issue Editors (Dr. Frédéric Frappart, Dr. Nicolas Baghdadi, Dr. Pere Quintana, and Dr. Mehrez Zribi) and Managing Editor (Dr. Evelyn Ning) for following up this manuscript. We look forward to hearing from you. Also, we would be glad to respond to any further questions and comments that you may have.
Note: Please also check the revised manuscript of the Round 2(track change version) in the attached file.
Best Regards,
Abolghasem Sadeghi-Niaraki,
################################################
Dr. Eng. Abolghasem Sadeghi-Niaraki
Assistant Professor
Faculty of Geodesy and Geomatics Engineering, GIS Dept.
K.N.Toosi University of Technology
Sejong University, South Korea
Tel | (+9821) 8887-7070
Fax | (+9821) 8878-6213
Mobile (Korea) |+8210 4253 5313
Official Email | [email protected]
Personal Email | [email protected]
Web | www.ubgi.ir
################################################
Reviewer 1:
Comment 7: based on the authors' response, there is no causal relationship between the first and second halves of this sentence. It does not make sense. Modification and clarification are needed.
Response
We thanks for the reviewer’s comment. In EBF model, Bel parameter takes into account the pessimistic mode and low probability and Pls parameter considers the optimistic mode and high probability state, so the value of Bel parameter is smaller or equal than Pls parameter, and difference between these two parameters is called Unc. This text was added to section 2.4.3.
Comment 9: the authors' statement in the response is not always true. The runoff is affected by many factors, more than slope, such as land cover/land use, soil types. The statement is misleading. Modification and clarification are needed.
Response
Thank you for good comment. Areas with low altitude have more permeable against runoff. This text was added to section 3.1.
In addition, for Comment 5, did the authors add some information clarifying “class” and “criterion”? If not, please add this info to the text.
Response
Thank you for the reviewer’ suggestions. This text was added to section 3.1.
Author Response File: Author Response.pdf
Reviewer 2 Report
The authors have made great effort to improve the manuscript. They addressed satisfactory the comments of my first review but by doing this some new editorial errors were introduced which are:
L40 Inserting new reference requires renumbering all subsequent references.
L65 Section Material and Methods should have number 2 not 3. Please renumbered all following sections appropriately.
L66 Replace ‘fifth’ with ‘five’.
L75 Referencing figures should start with 1. It means Figures 1 and 2 should be replaced with each other.
L112 TWI is defined as hydrological parameter in L138 so it should not be listed as topographic parameter.
L178 Some maps on Figure 3 are duplicated: a, b, e, f, g, h.
L206 There is no ‘t’ in equation 3.
L454 What the authors mean by 'hyper parameters'?
The English of the manuscript still needs polishing.
Author Response
Dear Prof. Dr. Arjen Y. Hoekstra
Editor-in-Chief
Water
July, 2019
Re: water-536182, entitled: “Groundwater Potential Mapping Using Integrated Ensemble of Three Bivariate Statistical Models with Random Forest and Logistic Model Tree Models"
Special Issue "Applications of Remote Sensing and GIS in Hydrology II"
We would like to express our special thanks to the reviewer#1’ efforts for re-evaluating our manuscript and offering us extremely valuable comments for Round 2. Based on the reviewer#2’ comments, we have again revised the manuscript. The next section contains our point-by-point responses (in blue) and changes and referencing to the manuscript (in green), based on the reviewers’ comments (in italic). We believe that our manuscript has substantially improved and is more readable for broader audiences. We would like to thank Editor-in-Chief (Prof. Dr. Arjen Y. Hoekstra), Section Editor-in-Chief, Special Issue Editors (Dr. Frédéric Frappart, Dr. Nicolas Baghdadi, Dr. Pere Quintana, and Dr. Mehrez Zribi) and Managing Editor (Dr. Evelyn Ning) for following up this manuscript. We look forward to hearing from you. Also, we would be glad to respond to any further questions and comments that you may have.
Note: Please also check the revised manuscript of the Round 2(track change version) in the attached file.
Best Regards,
Abolghasem Sadeghi-Niaraki,
################################################
Dr. Eng. Abolghasem Sadeghi-Niaraki
Assistant Professor
Faculty of Geodesy and Geomatics Engineering, GIS Dept.
K.N.Toosi University of Technology
Sejong University, South Korea
Tel | (+9821) 8887-7070
Fax | (+9821) 8878-6213
Mobile (Korea) |+8210 4253 5313
Official Email | [email protected]
Personal Email | [email protected]
Web | www.ubgi.ir
################################################
Reviewer 2:
L65 Section Material and Methods should have number 2 not 3. Please renumbered all following sections appropriately.
Response
Thank you for good comment. Sections numbering was corrected in the text.
L66 Replace ‘fifth’ with ‘five’.
Response
In line 66, the word five was replaced the word fifth.
L75 Referencing figures should start with 1. It means Figures 1 and 2 should be replaced with each other.
Response
Thank you for the reviewer’ suggestions. Figure 1 and figure 2 were corrected in the text of the paper.
L112 TWI is defined as hydrological parameter in L138 so it should not be listed as topographic parameter.
Response
In line 112, this case was deleted.
L178 Some maps on Figure 3 are duplicated: a, b, e, f, g, h.
Response
Duplicated maps have been removed (due to the use of track changes, the deletion of figures is not shown).
L206 There is no ‘t’ in equation 3.
Response
In section 2.4.1, parameter j was replaced with parameter t.
L454 What the authors mean by 'hyper parameters'?
Response
Thank you for good comment. In machine learning, the hyper parameter is called parameters that can be adjustable by the user (before training model).
Author Response File: Author Response.pdf
