Medium-Term Rainfall Forecasts Using Artificial Neural Networks with Monte-Carlo Cross-Validation and Aggregation for the Han River Basin, Korea
Round 1
Reviewer 1 Report
To whom it may concern,
the paper presents a study about application of artificial neural network (ANN) models to predict the rainfall amounts in the months of May and June, which are prone to water shortage for the Han River basin, in South Korea, using the lagged global climate indices and historical rainfall data.
The study is rather interesting and timely however in the reviewer opinion still many open questions remain:
1) lines 122-124:
"The choice of appropriate input variables is crucial in prediction modeling. Generally, the selection of input variables is based on a priori knowledge of causal variables [19]; however, any a priori knowledge could not be available to be used in the study area"
It seems that quite some studies exist about the climatology of the Han river area (see below) therefore it is not clear why a priori knowledge was not available a priori.
Kim, J. S., Jain, S., & Yoon, S. K. (2012). Warm season streamflow variability in the Korean Han River Basin: links with atmospheric teleconnections. International Journal of Climatology, 32(4), 635-640.
Kim, B. S., Kim, B. K., & Kwon, H. H. (2011). Assessment of the impact of climate change on the flow regime of the Han River basin using indicators of hydrologic alteration. Hydrological Processes, 25(5), 691-704.
2) The authors selected 14 climate indices as input for their ANN models but the choice is not adequately motivated and thus it remains unclear if other climate indices could eventually play a role.
3) The authors do not make any reference of the possible impacts of climate change on the reliability of the ANN models results for monthly rainfall over the Han River Basin. Are the results of tables 2 and 3 influenced by possible changes of rainfall climatology characteristics during the target period 1966-2018?
Kim, S. J., Kim, B. S., Jun, H. D., & Kim, H. S. (2010). The evaluation of climate change impacts on the water scarcity of the Han River Basin in South Korea using high resolution RCM data. Journal of Korea Water Resources Association, 43(3), 295-308.
4) This statement (lines 381-382) is very generic
"Further research will be required to improve the performance of ANN models to capture extremely high and low rainfall characteristics."
and it does not clarify how the authors are planning to tackle this shortcoming emerging from their ANN models results
Author Response
Thank you very much for your support and for taking the time to review our manuscript. We greatly appreciate your helpful comments and suggestions. All of your comments and suggestions have been incorporated into the revised manuscript.
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Editorial
The article uses 'We' in multiple locations. This is a style issue, but 'The authors' is always a better choice.
Comments; Line - Comment
116 - Check with the editor if the date of retrieval would be needed for internet addresses.
126 - 'Finding the most significant input variables using trial-and-error can be time-consuming; thus,' this isn't quite necessary. Please consider removing this.
141 - Please explain at least a brief description of the major nature of each index. Also, explain why these are selected. Finally, the inter-correlation between the selected index. There should be already published articles about this but I can't name them top off my head. This is important since some of the indexes are redundant due to high correlations.
197 - Any guideline or suggested equation to chose the number of a training run and set?
231 - Explain the box plot for the general audience. Not everyone knows how to read a box plot, believe it or not. One short sentence is enough.
247 - Fig 2. Suggest vertically align (a) and (b) instead of horizontally as is now. It'll be much easier to see the difference between the models. Also, the methods are labeled with (-), different from the writing.
258 - Some key statistics are given for Fig 2, but not for Table 2. Please mark the median values in Table 2 (box, bold type, etc).
275 - It would be great to look into this with the variable inter-correlation mentioned for line 141.
321 - '...produced a reasonable uncertainty band...' The entire section 3.3 is good, but there are better indicators out there such as skill scores. Please consider.
326 - In Fig 5, use 'Forecast (95% range)' instead of 'Lower Upper'
Author Response
Thank you very much for your support and for taking the time to review our manuscript. We greatly appreciate your helpful comments and suggestions. All of your comments and suggestions have been incorporated into the revised manuscript.
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The presented paper entitled »Medium-term rainfall forecasts using artificial neural networks with Monte-Carlo cross validation and aggregation for the Han River basin, Korea« describes the development of a model for prediction of rainfall amount in an area of South Korea, which is prone to drought. The artificial neural networks were used to construct the model, taking into account various variables and testing different variations. The paper is in general well written, the text is connected and the reading flows nicely. The descriptions are not too complicated and the explanations gives sufficient information.
I don’t have many comments but the two major ones in my opinion present an important completion:
Chapter 2.1: Please supplement the chapter with detailed description of the used data. Explain more about monthly areal rainfall data – probably this are measured data? From how many stations/locations did you collect them? Did you use the average values from all the stations or separately values from each station? What was the period of these data – also 1965 – 2018? Why did you select this period? Which are the other climate indices (L117) that you obtain? Were all collected for the same period (see L120 “data for a lot of climate indices” – a lot is not equal all)? Clearly state which input variables were chosen and list all of them, which were tested in correlation analysis (see L131: “the highly correlated” – this are not all of them). Here you should also explain all of the abbreviations used for the variables.
Chapters 3.2 and 3.3: Please add a discussion about the results, comparing them with the results of previous studies, other research work and other ANN models for predicting the rainfall. For example L303: why are the results for testing set the poorest? Was this also observed by other researchers and in case of other models? Compare the results. What about the value of the standard deviation in other studies? Description of figures 5 and 6: add comparison of the results from other studies – how successful were other models in predicting the rainfall amounts? What was the difference between yours and the other models, which performed better/worse? What about the prediction of peaks by other models? Did they also have the same problem?
Minor comments
L116 – L119: the links to web pages should be also cited among the references and in text marked with corresponding number. See the Instructions for authors: (Websites: 9. Title of Site. Available online: URL (accessed on Day Month Year). Unlike published works, websites may change over time or disappear, so we encourage you create an archive of the cited website using a service such as WebCite. Archived websites should be cited using the link provided as follows: 10. Title of Site. URL (archived on Day Month Year).
L201: Add: which software was used for the construction of the ANN and for all the analyses?
L228: Please add the description on RI in the methods section and explain what it actually means and what represents its number.
Table 2: Variable QBO(8), line 10 – was this variable again included in the testing and why or is this a mistake?
Author Response
Thank you very much for your support and for taking the time to review our manuscript. We greatly appreciate your helpful comments and suggestions. All of your comments and suggestions have been incorporated into the revised manuscript.
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
Thank you for providing such a detailed response to my comments. Everything has been taken into account. I only have two more suggestions:
- The Table 1 really provides an insight into extensive (pre)work. But it is really a big one and would maybe suite best in appendix.
- Response in the scope of Point 2: Explanation to my questions are written really nicely and it is too bad that you didn't include them in the manuscript. Reconsider to insert this in the Results and Discussion section ...
