A Multilayer Perceptron Model for Stochastic Synthesis
Round 1
Reviewer 1 Report
This study proposed a multilayer perceptron network-based stochastic model to generate daily precipitation. This manuscript was originally submitted a few weeks ago. Unfortunately, the resubmitted manuscript is the same as original submission. I regret to recommend rejection.
Comments:
#1 The novelty/significance of this study isn’t compelling enough for journal publication (in the Journal Hydrology)
#2 The superiority of the proposed method should be demonstrated. The results from this method should be compared/contrasted with the results based on the daily precipitation generation methods well established in the literature. See following relevant publications:
https://www.sciencedirect.com/science/article/pii/S0022169410003082
https://rmets.onlinelibrary.wiley.com/doi/full/10.1002/joc.3896
https://www.mdpi.com/2073-4433/12/2/135
Author Response
Regarding the novelty and significance of our method, we have restructured the Introduction section to articulate it more explicitly. Novelty: “This MLP-based approach is novel, because, in contrast to the existing similar approaches, it reproduces the statistical properties of the corresponding historical time series at multiple scales (Hurst effect).” Significance: “… our motivation is to provide a stochastic model that is fairly simple to implement, even in a spreadsheet [21], a tool with which practitioners, and in general stakeholders, are familiar so that the proposed method has good chances of being adopted by that important community.”
*The first and second studies suggested by the reviewer (in our manuscript [24,25]) are based on WeaGETS, a widely used precipitation generation model. Almost every single figure in our study (Figure 2 to Figure 15) displays a comparison between our model and WeaGETS.
**Please see below the reference numbers to which correspond the three studies suggested by the reviewer.
The publications suggested by the reviewer
https://www.sciencedirect.com/science/article/pii/S0022169410003082
https://rmets.onlinelibrary.wiley.com/doi/full/10.1002/joc.3896
https://www.mdpi.com/2073-4433/12/2/135
correspond in our manuscript to:
24. Chen, J.; Brissette, F.P.; Leconte, R. A daily stochastic weather generator for preserving lowfrequency of climate variability. Journal of Hydrology 2010, 388, 480–490. doi:10.1016/j.jhydrol.2010.05.032.
25. Chen, J.; Brissette, F.P. Comparison of five stochastic weather generators in simulating daily precipitation and temperature for the Loess Plateau of China. International Journal of Climatology 2014, 34, 3089–3105, doi:https://doi.org/10.1002/joc.3896.
9. Pan, F.; Nagaoka, L.; Wolverton, S.; Atkinson, S.F.; Kohler, T.A.; O’Neill, M. A Constrained Stochastic Weather Generator for Daily Mean Air Temperature and Precipitation. Atmosphere 2021, 12. doi:10.3390/atmos12020135.
Note: [9,25] were added in manuscript submission id:1166118, following the reviewer’s suggestion.
Reviewer 2 Report
The authors in this study developed a multilayer perceptron network model called MLPS, which properly configured in the input features and the cost function, for the stochastic synthesis of daily rainfall time series.
The specific approach applied in two locations with different climatic conditions in the Hohenpeissenberg in Germany and Gibraltar and was tested versus an established stochastic model. The time series of daily measurements were more than 100 years for the first location and more than 40 years for the second.
I found the manuscript by Rozos et al. very interesting.
The title and abstract are appropriate for the content of the text. The article is well constructed and the objectives were accomplished. The results showed that the MLPS model generated synthetic time series of rainfall at a single location in both case studies.
The results here presented that the MLPS can be easily modified to be applied to other types of hydrologic variables or to support multivariate modelling.
I think that the manuscript, in its current form, can be considered for publication.
Author Response
Thank you for your positive comments.
Reviewer 3 Report
The paper include former models of evaluation of hydrological parameters în comparații WITH New one. The results show that some models can be used even they are including a small number of indicator, but also can be used in the new macine learning hydrology al models. The paper can be published in this form after minor modifications of the languages used.
Author Response
Thank you for your positive comments.
Round 2
Reviewer 1 Report
I have no further comments on this manuscript
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
Please use different methods to compare and use state-of-the art model to improve the result.
Reviewer 2 Report
Good revision. Pleased to be able to recommend it for acceptance.
Reviewer 3 Report
This study proposed a multilayer perceptron network-based stochastic model to generate daily precipitation. I have two major comments:
#1 L82-84: This motivation isn’t compelling enough for journal publication of these kinds of research since programming languages such as “R” are fairly simple and freely available. We can reasonably assume that the readers/groups interested in this study are familiar with R or similar other languages.
#2 The superiority of the proposed method should be demonstrated. The results from this method should be compared/contrasted with the results based on the daily precipitation generation methods well established in the literature. See following relevant publications:
https://www.sciencedirect.com/science/article/pii/S0022169410003082
https://rmets.onlinelibrary.wiley.com/doi/full/10.1002/joc.3896
https://www.mdpi.com/2073-4433/12/2/135
