A Novel Hybrid Extreme Learning Machine Approach Improved by K Nearest Neighbor Method and Fireworks Algorithm for Flood Forecasting in Medium and Small Watershed of Loess Region
Round 1
Reviewer 1 Report
The title seems too long and a shorter title of the title is recommended. “Simulation effect” is the result of the simulation? What is ELM model should be shortly explained in the abstract such as it is a rainfall-runoff model or hydraulic model or AI model? Figure 1 is not clear, please replace a higher resolution picture with it. Figure 6 is also not clear. Any reason to choose this area for study? Line 206 stated terraced area is 98.48 km2 and those artificial terraces are cultivated lands? In 2012, the area of the cultivated lands is 84.27 km2 which is larger than that of terraced area. What is the unit of the vertical axis of Fig. 9? CMS? Some typo or grammar errors are found (such as Line 310, “It basic principle is assume….”), please check it up. Line 391 mentioned “lag time”, what does it mean? If it refers the time lag between the rainfall and the discharge, why the maximum deltaT is slightly larger than the average lag time of floods? Line 317, k or K nearest neighbor? Table 3 shows the results of ELM, what is the criteria of qualification? In the previous sections, it stated that some floods are used for training set and some are used for testing. It is recommended to show those results separately. Line 462 stated that the peak flood error decreases from HSP1 to HSP3. The improvement is not significant and it may improve if you train it a little more. As for the average peak current difference, it doesn’t mean “that the peaks of all the simulated floods occur early”. It can be told from Table 3 that some events such as event 19970729 the peak current difference is positive but on the average it is. Line 529, it stated “Table 3 show that ELM model has a lower great flood simulation qualified rate”. It seems we don’t know which event is “great flood” in Table 3. If it is true, then it implies that ELM is less accurate for great floods than that of KNN-FWA-ELM. In the following, it stated the RMSE is sensitive to extraordinarily large or small deviation in the simulation process. Then the authors got the conclusion, the RMSE of ELM are smaller than that of KNN-FWA-ELM. If the authors conclude that when dealing with large event floods, ELM is better than that of KNN-FWA-ELM. Then, “Table 3 show that ELM model has a lower great flood simulation qualified rate” should be false. In short, ELM should have higher simulation qualified rate for great floods. In addition, Table 7 lists the results for great floods. For RMSE of ELM, the 19670822 event has RMSE of 85.43, the average RMSE of ELM should be 40.32. As for NS of ELM, the two NSs are about 0.6 and 0.7 and the other two is about 0.83. How does it make the average value of 0.85. I recommend to check all the data listed in all tables. Table 8, the average of ELM is all wrong. The authors should check all the data. Line 570, how KNN conducts continuous forecasting if the observed discharge data are absent? If the discharge is absent, how to calculate the accuracy of the model since there is no observed discharge then. How is the completeness of the data set?Author Response
I would like to thank you for your time and effort to provide the helpful and informative feedback on our paper “A Novel Hybrid Extreme Learning Machine Approach Improved by K Nearest Neighbor Method and Fireworks Algorithm for Flood Forecasting in Medium and Small Watershed of Loess Region” submitted to Water. We have responded to each of the comments and updated correspondingly in the revised manuscript. The detailed responses are appended to the cover letter.
(1) The title seems too long and a shorter title of the title is recommended.
Reply:
Thank you very much for pointing out this question. In the revised version, we have revised the title. (Pg.1 line 2-5)
(2) “Simulation effect” is the result of the simulation?
Reply:
Thank you very much for pointing out this question. In the revised version, we have revised the “simulation effect” to “simulation performance”. Simulation performance is a description and summary of the simulation results (Pg.1 line 26, 29; Pg.18 line 503; Pg.20 line 537; Pg.27 line 631, 641, 651-652; Pg.28 line 664, 675-676; Pg.29 line 680)
(3) What is ELM model should be shortly explained in the abstract such as it is a rainfall-runoff model or hydraulic model or AI model?
Reply:
Thank you very much for pointing out this question. The full name of “ELM model” is “Extreme Learning Machine model”. The ELM model is one kind of the AI model, in order to correspond to the title, we have modified “ELM model” into “Extreme Learning Machine model”. (Pg.1 line 24)
(4) Figure 1 is not clear, please replace a higher resolution picture with it. Figure 6 is also not clear.
Reply:
Revision has been made as suggested by the reviewer. In the revised version, the figure 1 and figure 6 have been replaced with a higher resolution picture. (Pg.4 line 143; Pg.6 line 191)
(5) Any reason to choose this area for study?
Reply:
Thank you very much for pointing out this question. In the selection of the basin, the Gedong basin is a typical loess region, and it is also a typical medium and small watershed. In the loess region, the rainfall period is mainly from July to September and the high of intensity rainfall causes the flood to be of rapid flow production, high flood peaks, short duration, rapid flow rate and destructive power. In recent decades, the conditions of the underlying surface of the loess region have changed under the influences of natural and human activities. Sudden floods in medium and small watershed caused by sudden rainstrom and local heavy rainfall often lead to flash floods, which in turn results in the disasters and enormous property damage and casualties. Therefore, it is of great practical and theoretical significance to explore appropriate flood forecasting models for medium and small watershed for flood control and disaster reduction in the loess region under the condition of underlying surface changes and carrying out flood simulation and flood warning. And the Gedong basin has a long sequence of hydrological and meteorological data. In view of the above reasons, we choose the Gedong basin as a study area.
(6) Line 206 stated terraced area is 98.48 km2 and those artificial terraces are cultivated lands? In 2012, the area of the cultivated lands is 84.27 km2 which is larger than that of terraced area.
Reply:
Thank you very much for pointing out this question. We are very sorry for our negligence of the data validity. In the revised version, we have checked and modified these data. The terraced area is 76.75km2, accounting for 10.6% of the total area of the basin. (Pg.6 line 202)
(7) What is the unit of the vertical axis of Fig. 9? CMS?
Reply:
Thank you very much for pointing out this question. The unit of the vertical axis of figure 9 is “m3/s”. In the revised version, we have added the name and unit of the vertical axis. (Pg.9 line 247)
(8) Some typo or grammar errors are found (such as Line 310, “It basic principle is assume….”), please check it up.
Reply:
Thank you very much for pointing out this question. In the revised version, the whole manuscript has been double-checked and the language and grammar have been improved throughout the whole manuscript. (Pg.1 line 36-37, 39-40, 41-42; Pg.2 line 77, 84, 88; Pg.3 line 118-119, 123-124, 128, 134; Pg.4 line 149-157; Pg.5 line 166-173, 190; Pg.6 line 194-196, 200-202; Pg.7 line 205, 207-208, 210-211, 214; Pg.9 line 258; Pg.10 line 267, 285, 292, 294-295; Pg.11 line 308; Pg.12 line 317, 320, 324, 335, 338, 349, 361; Pg.13 line 371-372, 376, 386-387, 389; Pg.14 line 405-407, 415, 437-439; Pg.15 line 441; Pg.16 line 473; Pg.20 line 511-512, 524, 543; Pg.21 line 565-567; Pg.26 line 592-593, 597-604, 606-607, 610, 613-614; Pg.27 line 620-636, 644; Pg.28 line 655; Pg.29 line 699-724)
(9) Line 391 mentioned “lag time”, what does it mean? If it refers the time lag between the rainfall and the discharge, why the maximum deltaT is slightly larger than the average lag time of floods?
Reply:
Thank you very much for pointing out this question. In the manuscript, “lag time” is refer to the time lag between the rainfall and discharge, “ΔT” is the period by which the rainfall and discharge are shifted to a certain early time as the input data of the model and the period by which the discharge is shifted to a certain late time as the output data of the KNN-FWA-ELM model. If the maximum of ΔT is smaller than the average lag time, it may result in limited information obtained by the KNN-FWA-ELM model, making the optimal value ofΔT cannot be determined. Therefore, in order to determine the optimalΔT, the maximum of ΔT is slightly larger than the average lag time of floods.
(10) Line 317, k or K nearest neighbor?
Reply:
Revision has been made as suggested by the reviewer. It is the “K nearest neighbor”. (Pg.11 line 315)
(11) Table 3 shows the results of ELM, what is the criteria of qualification?
Reply:
Thank you very much for pointing out this question. In the revised version, we have added the evaluation indexes for forecasting performance (chapter 3.5). (Pg.15 line 445-469)
To evaluate the performance of the hybrid KNN-FWA-ELM model, the peak flood error (ΔQ), the peak current difference (Δh), the coefficient of determination (R2), Nash-Sutcliffe efficiency coefficient (NS), root mean squared error (RMSE), mean squared relative error (MSRE) and mean absolute relative error (MARE) are used as the evaluation indicators in this study, which have been widely and commonly used for evaluating the performance of flood forecasting and hydrological simulation. The formulas are defined as follows equations 1, 2, 3, 4, 5, 6, 7 respectively.
(1)
Where is the peak flood error (%); is the simulated discharge (m3/s); is the observed discharge (m3/s).
(2)
Where Δh is the peak current difference (h); hspeak is the simulated peak current time (h); hopeak is the observed peak current time (h).
(3)
(4)
Where is the simulated discharge at time t (m3/s); is the observed discharge at time t (m3/s); is the average of simulated discharge (m3/s); is the average of observed discharge (m3/s).
(5)
(6)
(7)
Where is the ith simulated discharge (m3/s); is the ith observed discharge (m3/s).
(12) In the previous sections, it stated that some floods are used for training set and some are used for testing. It is recommended to show those results separately.
Reply:
Revision has been made as suggested by the reviewer. (Pg.16 table 3; Pg.18 table 4)
(13) Line 462 stated that the peak flood error decreases from HSP1 to HSP3. The improvement is not significant and it may improve if you train it a little more.
Reply:
Thank you very much for pointing out this question. Revision has been made as suggested by the reviewer, and the simulated results are improved. In the revised version, we have tried our best to collect the flood events data from 2013 to 2018, but unfortunately, we only collected the data of four flood events from 2013 to 2016. The data of 2017 and 2018 is unavailable, so we have added four flood events in our paper. The simulated results are given in table 3-9. (Pg.16 table 3; Pg.18 table 4; Pg.21 table 5; Pg.26 table 6; Pg.27 table 7; Pg.28 table 8, table 9)
(14) As for the average peak current difference, it doesn’t mean “that the peaks of all the simulated floods occur early”. It can be told from Table 3 that some events such as event 19970729 the peak current difference is positive but on the average it is.
Reply:
Thank you very much for pointing out this question. Revision has been made as suggested by the reviewer. (Pg.18 line 490; Pg.20 line 524)
(15) Line 529, it stated “Table 3 show that ELM model has a lower great flood simulation qualified rate”. It seems we don’t know which event is “great flood” in Table 3. If it is true, then it implies that ELM is less accurate for great floods than that of KNN-FWA-ELM. In the following, it stated the RMSE is sensitive to extraordinarily large or small deviation in the simulation process. Then the authors got the conclusion, the RMSE of ELM are smaller than that of KNN-FWA-ELM. If the authors conclude that when dealing with large event floods, ELM is better than that of KNN-FWA-ELM. Then, “Table 3 show that ELM model has a lower great flood simulation qualified rate” should be false. In short, ELM should have higher simulation qualified rate for great floods.
Reply:
Thank you very much for pointing out this question. In the revised version, we have made correction and supplementary instruction according to the reviewer’s comments. The qualified rate and the RMSE are the two evaluation indexes for flood forecasting of KNN-FWA-ELM model. The simulation results and accuracy of the two models is determined by the values of eight statistical evaluation indexes (qualified rate, ΔQ, Δh, NS, R2, RMSE, MSRE and MARE). In the manuscript, we conclude that KNN-FWA-ELM model has better simulation performance for great, moderate and small floods than ELM model. (Pg.21 line 554-558)
(16) In addition, Table 7 lists the results for great floods. For RMSE of ELM, the 19670822 event has RMSE of 85.43, the average RMSE of ELM should be 40.32. As for NS of ELM, the two NSs are about 0.6 and 0.7 and the other two is about 0.83. How does it make the average value of 0.85. I recommend to check all the data listed in all tables. Table 8, the average of ELM is all wrong. The authors should check all the data.
Reply:
Thank you very much for pointing out this question. In the revised version, the data in table 7 and table 8 of the manuscript has been checked. In the manuscript, the average value of the ΔQ, Δh, NS, R2, RMSE, MSRE and MARE is obtained from the qualified flood events, and all the evaluation indexes of the not qualified flood events are not considerable. (Pg.27 table 7 line 643; Pg.28 table 8 line 654)
(17) Line 570, how KNN conducts continuous forecasting if the observed discharge data are absent? If the discharge is absent, how to calculate the accuracy of the model since there is no observed discharge then. How is the completeness of the data set?
Reply:
Thank you very much for pointing out this question. In the manuscript, the modeling of KNN method is as follows:
Where Es(t+) is the simulated discharge error at time t+; Qs(t+) is the simulated discharge at time t+; P is the rainfall at time t and that at a certain early time; Q is the discharge corresponding to the rainfall at the early time.
In the initial calculation of KNN, the observed discharge data is used as the early-stage discharge of model input; as the calculation continues, the simulated discharge can be used as early-stage discharge of model input in the subsequent calculation. By performing this loop simulation, the information of the early-stage discharge can be effectively used, and the simulated value of early-stage discharge value, instead of the observed value, can serve as the input data of the model, which is favorable to the continuous flood forecasting. The KNN method is applied to predict the error of the discharge at the runoff gauging station and to achieve continuous flood forecasting when the observed data is absent and to correct the simulation results when the observed data is complete, which improves the simulation and prediction accuracy. In the prediction of the KNN, the training and testing data is employed to determine the prediction precision of the model on the basis of evaluation indexes. And then, the KNN method is used to predict the discharge error. Even if the observed data is absent, we can assure the prediction precision is reliable and don’t need to calculate the accuracy of the model with the observed discharge data. In our paper, the data set of the Gedong basin is complete.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This study proposes a hybrid model combining k-nearest neighbor method, fireworks algorithm, and extreme learning machine for improving flood forecasting for different surface conditions and flood grades. This paper deals with an interesting topic from the hydrological point of view and is consistent with the scope of this journal. However, there are problems that detract from the quality of this manuscript. Thus, I cannot recommend the acceptance of this study. Comments for improvement are presented below:
[1] Above all things, there are so many incorrect parts in view of grammar and writing style throughout the manuscript. Further, the manuscript includes many misspellings. The manuscript should be thoroughly corrected through the review of native speakers.
[2] Abbreviations should be defined in full words when they are used at first, and thereafter only the abbreviations should be used.
[3] The authors use five performance indices (peak error, peak flow difference, NS, R2, and RMSE) for evaluating the model performance. R2 can be biased towards higher values although they evaluate the level of overall agreement between observed and modeled values. NS can be used as the alternatives of R2. However, NS also has some limitations. Due to the use of squared differences, NS is oversensitive to outliers and thus biased towards extreme events. NS is sensitive to peak values, whereas insensitive to lower values. In addition, NS is insensitive to systematic errors. On the other hand, RMSE is biased towards higher magnitude events since it is based on squared differences. The index is also insensitive to lower magnitude events.
Since each statistical index has its own disadvantages, various indices should be used for evaluating the model performance effectively. In addition, I suggest that statistical indices suitable for low, medium, and high data values should be used in order to assess the model performance effectively. For example, R2, NS, and index of agreement (IOA) can be used for evaluating the level of overall agreement between observed and modeled values. RMSE is a good efficiency index for high data values, whereas MARE (mean absolute relative error) is for low data values. MAE (mean absolute error) evaluates all deviations from observed values and MSRE (mean squared relative error) can be used as a good efficiency index for moderate data values. AARE (average absolute relative error) and TS (threshold statistics) give appropriate weights for low, moderate, and high data values, and it has been reported that they can provide better performance evaluation. More information on the evaluation of model performance can be found in Dawson et al. (2001), Jain and Srinivasulu (2004), and Dawson et al. (2007).
<References>
Dawson CW, Wilby RL (2001). Hydrological modelling using artificial neural networks. Prog. Phys. Geogr., 25:80-108.
Jain A, Srinivasulu S (2004). Development of effective and efficient rainfall-runoff models using integration of deterministic, real-coded genetic algorithms and artificial neural network techniques. Water Resour. Res., 40(4).
Dawson CW, Abrahart RJ, See LM (2007). HydroTest: A web-based toolbox of evaluation metrics for the standardized assessment of hydrological forecasts. Environ. Mode. Softw., 22:1034-1052.
[4] Section 4: The authors describe only the results obtained from this study and do not discuss the results deeply. Please provide the discussion based on the results. This can include the meaning of the results, the comparison with previous studies, the theoretical description of why the proposed method provides better results, the limitations and improvements of the proposed method, etc.
[5] Section 5: Please add the limitations and future tasks of this study at the end of the section.
[6] Figures 2-4: In the legends, units are missing.
[7] Figure 9: Please add the name of the vertical axis.
[8] Section 4: The unit of RMSE is missing.
[9] Section 3: Please add the description of performance indices (NS, R2, etc.), which are used in section 4, to section 3.
[10] Section 4: Generally, “R2” refers to the coefficient of determination. However, the author uses the term, “correlation coefficient”, in the text. This needs to be corrected clearly.
[11] Figure 13: The abbreviation “RMSE” is not defined.
[12] Table 3: The abbreviations, “NS” and R2” are not defined.
Author Response
I would like to thank you for your time and effort to provide the helpful and informative feedback on our paper “A Novel Hybrid Extreme Learning Machine Approach Improved by K Nearest Neighbor Method and Fireworks Algorithm for Flood Forecasting in Medium and Small Watershed of Loess Region” submitted to Water. We have responded to each of the comments and updated correspondingly in the revised manuscript. The detailed responses are appended to the cover letter.
This study proposes a hybrid model combining k-nearest neighbor method, fireworks algorithm, and extreme learning machine for improving flood forecasting for different surface conditions and flood grades. This paper deals with an interesting topic from the hydrological point of view and is consistent with the scope of this journal. However, there are problems that detract from the quality of this manuscript. Thus, I cannot recommend the acceptance of this study. Comments for improvement are presented below:
[1] Above all things, there are so many incorrect parts in view of grammar and writing style throughout the manuscript. Further, the manuscript includes many misspellings. The manuscript should be thoroughly corrected through the review of native speakers.
Reply:
Thank you very much for pointing out this question. Revision has been made as suggested by the reviewer. In the revised version, the whole manuscript has been double-checked and the language and grammar have been improved throughout the whole manuscript. (Pg.1 line 36-37, 39-40, 41-42; Pg.2 line 77, 84, 88; Pg.3 line 118-119, 123-124, 128, 134; Pg.4 line 149-157; Pg.5 line 166-173, 190; Pg.6 line 194-196, 200-202; Pg.7 line 205, 207-208, 210-211, 214; Pg.9 line 258; Pg.10 line 267, 285, 292, 294-295; Pg.11 line 308; Pg.12 line 317, 320, 324, 335, 338, 349, 361; Pg.13 line 371-372, 376, 386-387, 389; Pg.14 line 405-407, 415, 437-439; Pg.15 line 441; Pg.16 line 473; Pg.20 line 511-512, 524, 543; Pg.21 line 565-567; Pg.26 line 592-593, 597-604, 606-607, 610, 613-614; Pg.27 line 620-636, 644; Pg.28 line 655; Pg.29 line 699-724)
[2] Abbreviations should be defined in full words when they are used at first, and thereafter only the abbreviations should be used.
Reply:
Thank you very much for pointing out this question. Revision has been made as suggested by the reviewer. The abbreviations have been defined in full words when they are used at first, and thereafter the abbreviations are used in the manuscript. (Pg.15 line 449-451; Pg.18 line 476-483, 487-491, 493, 495, 498-499; Pg.20 line 511-517, 521-525, 527, 529, 532-533, 541; Pg.21 line 548-551, 553-554, 560, 562; Pg.26 line 612, 613, 616; Pg.27 line 621-623, 632, 645-648; Pg.28 line 657, 659-660, 668-671; Pg.29 line 699, 702-704, 706)
[3] The authors use five performance indices (peak error, peak flow difference, NS, R2, and RMSE) for evaluating the model performance. R2 can be biased towards higher values although they evaluate the level of overall agreement between observed and modeled values. NS can be used as the alternatives of R2. However, NS also has some limitations. Due to the use of squared differences, NS is oversensitive to outliers and thus biased towards extreme events. NS is sensitive to peak values, whereas insensitive to lower values. In addition, NS is insensitive to systematic errors. On the other hand, RMSE is biased towards higher magnitude events since it is based on squared differences. The index is also insensitive to lower magnitude events.
Since each statistical index has its own disadvantages, various indices should be used for evaluating the model performance effectively. In addition, I suggest that statistical indices suitable for low, medium, and high data values should be used in order to assess the model performance effectively. For example, R2, NS, and index of agreement (IOA) can be used for evaluating the level of overall agreement between observed and modeled values. RMSE is a good efficiency index for high data values, whereas MARE (mean absolute relative error) is for low data values. MAE (mean absolute error) evaluates all deviations from observed values and MSRE (mean squared relative error) can be used as a good efficiency index for moderate data values. AARE (average absolute relative error) and TS (threshold statistics) give appropriate weights for low, moderate, and high data values, and it has been reported that they can provide better performance evaluation. More information on the evaluation of model performance can be found in Dawson et al. (2001), Jain and Srinivasulu (2004), and Dawson et al. (2007).
<References>
Dawson CW, Wilby RL (2001). Hydrological modelling using artificial neural networks. Prog. Phys. Geogr., 25:80-108.
Jain A, Srinivasulu S (2004). Development of effective and efficient rainfall-runoff models using integration of deterministic, real-coded genetic algorithms and artificial neural network techniques. Water Resour. Res., 40(4).
Dawson CW, Abrahart RJ, See LM (2007). HydroTest: A web-based toolbox of evaluation metrics for the standardized assessment of hydrological forecasts. Environ. Mode. Softw., 22:1034-1052.
Reply:
Thank you very much for pointing out this question. Your suggestion is of great significance to the revision and improvement of our paper. We have learned more about the statistical evaluation indexes in your suggestion and literatures. In the manuscript, we have added the evaluation indexes for forecasting performance (chapter 3.5) (Pg.15 line 445-469). In this section, we define seven statistical indexes and refer to the evaluation of model performance in the literature (Dawson CW, Wilby RL, 2001; Jain A, Srinivasulu S, 2004; Dawson CW, Abrahart RJ, See LM, 2007) (Pg.31 line 831-832; Pg.32 line 833-837). In the manuscript, we add the value of MSRE and MARE in the results and discussion (Pg.16 table 3; Pg.18 table 4; Pg.21 table 5; Pg.26 table 6; Pg.27 table 7; Pg.28 table 8, table 9). In the future, we will adopt more statistical indexes for further research. Special thanks to you for your good comments.
References
[47]Dawson, C. W.; Abrahart, R. J.; See, L. M. HydroTest: A web-based toolbox of evaluation metrics for the standardized assessment of hydrological forecasts. Environ. Mode. Softw. 2007, 22, 1034-1052.
[48]Jain, A.; Srinivasulu, S. Development of effective and efficient rainfall-runoff models using integration of deterministic, real-coded genetic algorithms and artificial neural network techniques. Water Resour. Res. 2004, 40, 1-12.
[49]Dawson, C.W.; Wilby, R.L. Hydrological modelling using artificial neural networks. Prog. Phys. Geogr. 2001, 25, 80-108.
[4] Section 4: The authors describe only the results obtained from this study and do not discuss the results deeply. Please provide the discussion based on the results. This can include the meaning of the results, the comparison with previous studies, the theoretical description of why the proposed method provides better results, the limitations and improvements of the proposed method, etc.
Reply:
Thank you very much for pointing out this question. Revision has been made as suggested by the reviewer. We have added the discussion in the section 4. (Pg.26 line 597-604; Pg.27 line 631-636; Pg.28 line 674-681)
[5] Section 5: Please add the limitations and future tasks of this study at the end of the section.
Reply:
Thank you very much for pointing out this question. Revision has been made as suggested by the reviewer. We have added the limitations and future tasks of this study at the end of section 5. (Pg.29 line 716-724)
[6] Figures 2-4: In the legends, units are missing.
Reply:
Revision has been made as suggested by the reviewer. In the figures, we have added the units in the legends. (Pg.4 figure 2; Pg.5 figure 4, 5; the positions of the figure 3 and figure 5 are changed)
[7] Figure 9: Please add the name of the vertical axis.
Reply:
Revision has been made as suggested by the reviewer. In figure 9, we have added the name of vertical axis. (Pg.9 figure 9)
[8] Section 4: The unit of RMSE is missing.
Reply:
Revision has been made as suggested by the reviewer. We have added the unit of RMSE in section 4. (Pg.16 table 3; Pg.18 table 4; Pg.21 table 5; Pg.26 table 6; Pg.27 table 7; Pg.28 table 8, table 9)
[9] Section 3: Please add the description of performance indices (NS, R2, etc.), which are used in section 4, to section 3.
Reply:
Revision has been made as suggested by the reviewer. In the revised version, the description of evaluation indexes for forecasting performance have been supplemented (chapter 3.5). (Pg.15 line 445-469)
To evaluate the performance of the hybrid KNN-FWA-ELM model, the peak flood error (ΔQ), the peak current difference (Δh), the coefficient of determination (R2), Nash-Sutcliffe efficiency coefficient (NS), root mean squared error (RMSE), mean squared relative error (MSRE) and mean absolute relative error (MARE) are used as the evaluation indicators in this study, which have been widely and commonly used for evaluating the performance of flood forecasting and hydrological simulation. The formulas are defined as follows equations 1, 2, 3, 4, 5, 6, 7 respectively.
(1)
Where is the peak flood error (%); is the simulated discharge (m3/s); is the observed discharge (m3/s).
(2)
Where Δh is the peak current difference (h); hspeak is the simulated peak current time (h); hopeak is the observed peak current time (h).
(3)
(4)
Where is the simulated discharge at time t (m3/s); is the observed discharge at time t (m3/s); is the average of simulated discharge (m3/s); is the average of observed discharge (m3/s).
(5)
(6)
(7)
Where is the ith simulated discharge (m3/s); is the ith observed discharge (m3/s).
[10] Section 4: Generally, “R2” refers to the coefficient of determination. However, the author uses the term, “correlation coefficient”, in the text. This needs to be corrected clearly.
Reply:
Thank you very much for pointing out this question. In the revised version, we have corrected the phrase about R2. The R2 refer to the coefficient of determination. (Pg.15 line 449)
[11] Figure 13: The abbreviation “RMSE” is not defined.
Reply:
Revision has been made as suggested by the reviewer. In the revised version, we have defined the full words of RMSE. (Pg.12 line 362)
[12] Table 3: The abbreviations, “NS” and R2” are not defined.
Reply:
Revision has been made as suggested by the reviewer. In the revised version, we have defined the full words of NS and R2 in section 3.5. (Pg.15 line 449-450)
Author Response File: Author Response.pdf
Reviewer 3 Report
Lines 41, 80, 151, 173
In scientific papers, expressions such as “stage gave birth”, “bright application prospects”, “extremely dissected” or “causes the terrain to be broken” are preferred to be avoided.
Line 60
Put the correct citation number at „Lima et al. [9]”
Line 153
The first part of the phrase should be rephrased.
Figure 4
The slope aspect map – values can be classified according to coordinates, in order for them to be easier to understand. (https://gisgeography.com/slope-aspect-microclimate-south-facing/)
Line 188
Misspelled word „land use”
Figura 6.
The text in the legend is unreadable.
Tabelul 2 is not cited
A phrase/paragrah about Figure 8 can be added, if it is cited and introduced inside the text.
Why does the analysis stop in 2012? Is there not data from 2017 or 2018 available?
All maps, from 1, to 9 must have higher print quality, the scale, and north arrow of similar sizes, also the legend text should be in the same format and properly readable.
Conclusions must be written more to the point, with focus on the results and the findings.
After reading the article, I have noticed an increase in error count, for the three study periods, in the case of both models. This can lead to another conclusion, implying that the models are better suited for intensly modified terrain, by contrast to those covered in natural vegetation.
Author Response
I would like to thank you for your time and effort to provide the helpful and informative feedback on our paper “A Novel Hybrid Extreme Learning Machine Approach Improved by K Nearest Neighbor Method and Fireworks Algorithm for Flood Forecasting in Medium and Small Watershed of Loess Region” submitted to Water. We have responded to each of the comments and updated correspondingly in the revised manuscript. The detailed responses are appended to the cover letter.
1.Lines 41, 80, 151, 173
In scientific papers, expressions such as “stage gave birth”, “bright application prospects”, “extremely dissected” or “causes the terrain to be broken” are preferred to be avoided.
Reply:
Thank you very much for pointing out this question. In the revised version, the whole manuscript has been double-checked and the expressions, language and grammar have been improved throughout the whole manuscript. (Pg.1 line 36-37, 39-40, 41-42; Pg.2 line 77, 84, 88; Pg.3 line 118-119, 123-124, 128, 134; Pg.4 line 149-157; Pg.5 line 166-173, 190; Pg.6 line 194-196, 200-202; Pg.7 line 205, 207-208, 210-211, 214; Pg.9 line 258; Pg.10 line 267, 285, 292, 294-295; Pg.11 line 308; Pg.12 line 317, 320, 324, 335, 338, 349, 361; Pg.13 line 371-372, 376, 386-387, 389; Pg.14 line 405-407, 415, 437-439; Pg.15 line 441; Pg.16 line 473; Pg.20 line 511-512, 524, 543; Pg.21 line 565-567; Pg.26 line 592-593, 597-604, 606-607, 610, 613-614; Pg.27 line 620-636, 644; Pg.28 line 655; Pg.29 line 699-724)
2.Line 60
Put the correct citation number at „Lima et al. [9]”
Reply:
Revision has been made as suggested by the reviewer. (Pg.2 line 58)
3.Line 153
The first part of the phrase should be rephrased.
Reply:
Revision has been made as suggested by the reviewer. (Pg.4 line 151-155)
4.Figure 4
The slope aspect map – values can be classified according to coordinates, in order for them to be easier to understand.
(https://gisgeography.com/slope-aspect-microclimate-south-facing/)
Reply:
Thank you very much for pointing out this question. Your suggestion is of great significance to the revision of our paper. In the revised version, we have classified the slope aspect according to coordinates. (Pg.5 figure 4)
5.Line 188
Misspelled word „land use”
Reply:
Revision has been made as suggested by the reviewer. (Pg.5 line 184)
6.Figure 6.
The text in the legend is unreadable.
Reply:
Thank you very much for pointing out this question. In the revised version, the figure 6 has been replaced with a higher resolution picture and the legend is readable. (Pg.6 line 191)
7.Table 2 is not cited
Reply:
Revision has been made as suggested by the reviewer. (Pg.8 line 239-240)
8.A phrase/paragraph about Figure 8 can be added, if it is cited and introduced inside the text.
Reply:
Revision has been made as suggested by the reviewer. (Pg.7 line 227-229)
9.Why does the analysis stop in 2012? Is there not data from 2017 or 2018 available?
Reply:
Thank you very much for pointing out this question. In the revised version, we have tried our best to collect the flood events data from 2013 to 2018, but unfortunately, we only collected the data of four flood events from 2013 to 2016. The data of 2017 and 2018 is unavailable. The simulated results of the added four flood events are given in the revised version. (Pg.16 table 3; Pg.18 table 4; Pg.21 table 5; Pg.26 table 6; Pg.27 table 7; Pg.28 table 8, table 9)
10.All maps, from 1, to 9 must have higher print quality, the scale, and north arrow of similar sizes, also the legend text should be in the same format and properly readable.
Reply:
Revision has been made as suggested by the reviewer. The all figures (from 1 to 9) have been replaced with a higher print quality picture. The scale and north arrow are of the similar sizes. The legend is set as the same format and it can be properly readable. (Pg.4 figure 1, 2, 3; Pg.5 figure 4, 5; Pg.6 figure 6; Pg.7 figure 7; Pg.8 figure 8; Pg.9 figure 9)
11.Conclusions must be written more to the point, with focus on the results and the findings.
Reply:
Revision has been made as suggested by the reviewer. (Pg.29 line 699-724)
12.After reading the article, I have noticed an increase in error count, for the three study periods, in the case of both models. This can lead to another conclusion, implying that the models are better suited for intensly modified terrain, by contrast to those covered in natural vegetation.
Reply:
Thank you very much for pointing out this question. Special thanks to you for your good comments about the conclusion. Your suggestion is of great significance to the improvement of our paper. After further consideration, we think that the research content of our paper is not enough to support this conclusion, and we will conduct further research in order to reach this conclusion in the future. Once again, thank you very much for your comments and suggestions.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Some figures are still not clear such as Fig. 1. The character is very fuzzy in the PDF.
Reviewer 2 Report
Overall, the manuscript has been revised appropriately and the quality of the manuscript has also been greatly improved. Therefore, I recommend the acceptance and publication of the manuscript.
