Utilizing LSTM-GRU for IOT-Based Water Level Prediction Using Multi-Variable Rainfall Time Series Data
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsOverall, I think the paper is well prepared, both in structure and presentation.
Please find my detailed comments as attached.
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThis study demonstrates the performance of LSTM, GRU models, and a combination of both for predicting floods in Semarang based on time series data. The manuscript is clearly structured. However, the methods in this study are not significantly innovative compared with previous study. The results do not provide sufficient convincing. Therefore, I would recommend a rejection for this manuscript.
1. “Introduction” Section should provide research gap, but the authors took a lot of part to explain background. The innovativeness of the research is not reflected in the manuscript.
2. Figure 2: This figure is blurry and lacks legends.
3. “Materials and Methods” section does not detail how to couple the LSTM and GPU models.
4. Line 146-147: The specific locations of the data collection should be specified in the figure.
5. Line 183-195, Line 200-201, Line 239-240, Line 253-254: Please check formulas.
6. Line 207: Are Wf, Wc and Wo all weights used in forget gate calculations? Why?
7. Line 210: Line 217 mentions Ct=1, so it may be better to add “tanh (21)=1” in Line 210.
8. Line 237: Capitalize the first letter.
9. Please check formula (14).
10. Section 2.5 only introduces the calculation of accuracy, precision, recall and F1, without clarifying the physical meaning of the variables in the formulas (that is, it is not explained in combination with the research object).
11. Line 317-339: The contents do not match Table 2.
12. It can be seen from Section 3.1 that adding upstream rainfall will worsen model performance, and vice versa in Section 3.2. Why? Which result is more convincing? With different results, can the authors conclude that adding upstream information will improve model performance?
13. Line 398, Line 409-411: Please supplement references to support the conclusions.
14. Line 417-420: This part is not represented in the manuscript.
Comments on the Quality of English LanguageNo
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThis article discusses the use of LSTM-GRU for IOT-based water level prediction using multi-variable rainfall time series data. The article has an interesting topic. There are some issues that need revision.
1- Introduction: There are several families of machine learning and AI-based techniques which have been used in the past 2 decades in hydrological modelling and forecasting applications. The introduction needs some improvements to touch on these methods briefly and provide sufficient justifications for why LSTM and GRU were selected in this study. For example, efforts are made to build adaptive neuro-fuzzy systems that can perform well in water level forecasting (Please see the below references):
- Hong YST, White PA (2009) Hydrological modeling using a dynamic neuro-fuzzy system with on-line and local learning algorithm. Adv Water Resour 32(1):110–119
- Nguyen, P.-K.-T., Chua, L.-H.-C., Talei, A., Chai, Q.H., 2018. Water level forecasting using neuro-fuzzy models with local learning. Neural Comput. Appl. 30, 1877–1887.
2- Introduction: It is necessary to highlight the knowledge gap that the present study aims to address. What are the novelties of this work? These need to be highlighted explicitly.
3- Figure 2: Please add scale and North indicator to your map. Besides, the information on this map needs to be clarified, preferably in the caption. What are the points 1 to 4? What is the difference between blue and orange?
4- Section 2-2:
- More information is needed regarding the data used in this study. For example, the total number of data points at each point. Besides, statistical information (e.g., min, max, ave, ...) of these datasets must be provided. Also, it is not clear how the data is split into training and validation. What is the length of each dataset? What are their statistical differences?
- The explanations provided for scenarios need elaboration. This explanation can be properly cross-referenced to Table 1 for further clarity.
5- Section 2-5: I assume the modelling task in this study is predicting water level time series. If this is correct, using the current performance criteria is not sufficient. Statistical measures such as MAE, RMSE, R2, NSE, etc. are needed to better assess the model performance. This section needs to add those parameters.
6- Equation 14: please use multiplication sign instead of letter x.
7- Section 3: This section starts with a statement saying that "the results show the LSTM model is superior" while no result is presented up to this point. Perhaps restructuring is needed to have a better flow in this section.
8- Figure 3: Error parameters have unit. The vertical axis needs to show that.
9- Section 4: Lines 406-411
Overfitting is an issue that happens in training when the model starts to memorize instead of learning. Data lengths used for training and validation and their statistical properties are important in this case.
10- Section 4: Please add the limitations of your study at the end of this section.
Comments on the Quality of English Language
careful proofreading is necessary.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsThe author's English expression ability still needs to be further improved. Although he can express correctly, he can express too much orally and be more professional, which requires many trainings.
Reviewer 3 Report
Comments and Suggestions for AuthorsAuthors have made commendable efforts to improve the article following the reviewers' comments and suggestions. I will be happy to support this article's publication in its current format.
