Temperature Prediction of Chinese Cities Based on GCN-BiLSTM
Round 1
Reviewer 1 Report
1. Overview and general recommendation:
The research problem was formulated correctly, and the researcher's methodology was also precisely presented. I have no comments on the scientific part.
The presented material corresponds to the profile of the Journal "Applied Sciences". The scientific value of the submitted material qualifies the article for publication in this Journal.
The article may be published after completing and correcting all issues. I recommend that a minor revision is necessary. I made the detailed comments in point 2. I ask that the authors specifically address each of my comments in their response.
2. Minor comments
The layout of the manuscript is incorrect.
Please consider changing the layout of the manuscript by introducing the main points:
1. Introduction
2. Materials and methods
3. Results and discussion
4. Conclusions
To sum up, the "Discussion" and "Conclusions" should be formulated in such a way as to present the key results obtained in effect of the completed research using proprietary methods.
Author Response
[Comment 1] The layout of the manuscript is incorrect.
Please consider changing the layout of the manuscript by introducing the main points:
- Introduction
- Materials and methods
- Results and discussion
- Conclusions
To sum up, the "Discussion" and "Conclusions" should be formulated in such a way as to present the key results obtained in effect of the completed research using proprietary methods.
[Response] Thank you for your precious suggestion. According to your suggestion, we have updated the manuscript structure.
Reviewer 2 Report
Temperature prediction is a complex task because of the spatial and temporal dependence of temperature and the indirect and direct effects of multiple influencing factors. This paper aims to improve the accuracy of existing temperature prediction methods by addressing these challenges. This is worth doing methodologically, but the paper needs a more comprehensive review and comparison to demonstrate its novelty and advantages. The following major comments are encouraged to be included to further improve the manuscript.
1. The review of existing research is incomplete. A lot of work has been done to incorporate the spatial and temporal dependence of variables into the prediction task in many other application domains (Zhao et al., 2019), and has been increasingly done in temperature prediction (Qiao et al., 2022). It is strongly suggested to use a table to compare the performance of existing temperature prediction methods on several methodological issues targeted by this paper (e.g., temporal dependence, spatial dependence, multiple meteorological factors, large scale) in the literature review section. Then, analyze what limitations of existing spatio-temporal prediction models still exist in these aspects.
Zhao, L., Song, Y., Zhang, C., Liu, Y., Wang, P., Lin, T., ... & Li, H. (2019). T-gcn: A temporal graph convolutional network for traffic prediction. IEEE Transactions on Intelligent Transportation Systems, 21(9), 3848-3858.
Qiao, B., Wu, Z., Tang, Z., & Wu, G. (2022, February). Sea surface temperature prediction approach based on 3D CNN and LSTM with attention mechanism. In 2022 24th International Conference on Advanced Communication Technology (ICACT) (pp. 342-347). IEEE.
2. The comparison of methods in the experiments is also unfair. It lacks comparison with existing models that combine both spatial and temporal deep learning networks.
3. In fact, the combination of GCN and BiLSTM is not new (Ma et al., 2021). What is the increment of this work?
Ma, D., Guo, Y., & Ma, S. (2021, March). Short-Term Subway Passenger Flow Prediction Based on GCN-BiLSTM. In IOP Conference Series: Earth and Environmental Science (Vol. 693, No. 1, p. 012005). IOP Publishing.
4. This paper questions the inadequate extraction of spatial features in existing work, but there are no figures, comparisons, and discussion about the spatial distribution of temperature prediction results.
5. This paper also points out that existing deep learning models cannot handle large-scale data. Additional explanation is needed to clarify how the model in this paper solves this problem.
Author Response
[Comment 1] The review of existing research is incomplete. A lot of work has been done to incorporate the spatial and temporal dependence of variables into the prediction task in many other application domains (Zhao et al., 2019), and has been increasingly done in temperature prediction (Qiao et al., 2022). It is strongly suggested to use a table to compare the performance of existing temperature prediction methods on several methodological issues targeted by this paper (e.g., temporal dependence, spatial dependence, multiple meteorological factors, large scale) in the literature review section. Then, analyze what limitations of existing spatio-temporal prediction models still exist in these aspects.
[Response] Thank you for your comment. We have re-written section 2 to review the existing research from line 104 to 164. In the Related Works, we have compared the advantages and disadvantages of the existing temperature prediction methods as shown in Table 1 in Line 164. Also, we have supplemented the literature reviews about the spatial-temporal prediction methods in the field of temperature prediction and other fields. And we have analyzed the limitations of the current spatiotemporal prediction methods.
[Comment 2] The comparison of methods in the experiments is also unfair. It lacks comparison with existing models that combine both spatial and temporal deep learning networks.
[Response] Thank you for your precious suggestion. According to your suggestion, we have carefully referred to the model in the field of temperature prediction combining spatial and temporal prediction. for example, Qiao et al., 2022 and Jeong et al.,2021. We have found that they analyzed image data based on CNN model to obtain spatial features. CNN is good at processing raster and regular graph data, such as images and videos. They belong to Euclidean space and can be transformed into regular matrix for convolution. In this research, we use GCN to process irregular graph structure data (site data). But CNN cannot handle data structures in non-Euclidean spaces. Our experiments are based on non-regularly distributed site data, and we currently do not have corresponding raster data. Therefore, we have not found a suitable spatial-temporal prediction model at present. In our subsequent work, we will continue to explore this issue.
- Qiao, B.; Wu, Z.; Tang, Z.; et al. Sea surface temperature prediction approach based on 3D CNN and LSTM with attention mechanism//2022 24th International Conference on Advanced Communication Technology (ICACT). IEEE, 2022: 342-347.
- Jeong, S.; Park, I.; Kim, H. S.; et al. Temperature prediction based on bidirectional long short-term memory and convolutional neural network combining observed and numerical forecast data. Sensors. 2021, 21(3): 941.
[Comment 3] In fact, the combination of GCN and BiLSTM is not new (Ma et al., 2021). What is the increment of this work?
Ma, D., Guo, Y., & Ma, S. (2021, March). Short-Term Subway Passenger Flow Prediction Based on GCN-BiLSTM. In IOP Conference Series: Earth and Environmental Science (Vol. 693, No. 1, p. 012005). IOP Publishing.
[Response] Thank you for your precious suggestion. Compared with the model in this paper (Ma et al., 2021), the main increments in our work are as follows.
- The process of constructing the GCN network is described in detail in our work, which also includes the processing details of the specific graph network structure. And, it is explained with an example of the geographic distribution of cities in Jiangsu Province. The paper (Ma et al., 2021) did not describe the graph network structure of metro stations and processing method of metro passenger flow, but only illustrated the GCN principle, which lacked detail explanation.
- Our work uses a time sliding window to process data of multiple meteorological factors, including temperature data and other meteorological factors. The GCN network is used to obtain the spatial features of the multi-meteorological factors in the target area and surrounding areas. The paper (Ma et al., 2021) did not consider the influence of other factors on the passenger flow, and only used only one factor of the passenger flow data for modeling.
[Comment 4] This paper questions the inadequate extraction of spatial features in existing work, but there are no figures, comparisons, and discussion about the spatial distribution of temperature prediction results.
[Response] We are very sorry for the absence of this part. We have added this part from line 435 to 444.
[Comment 5] This paper also points out that existing deep learning models cannot handle large-scale data. Additional explanation is needed to clarify how the model in this paper solves this problem.
[Response] Thank you for your comment. We are very sorry for the unclear description. We have re-written this sentence from line 126 to 127.
Reviewer 3 Report
I was asked to review the paper titled 'Temperature Prediction of Chinese Cities Based on GCN-BiLSTM' co-authored by Miao et al. The manuscript discusses on the prediction of the air temperature in the Chinese cities. Following are the comments and suggestions that MUST be addressed prior to publications.
1. The introduction is not sufficient enough to describe the importance of the research in the target study area. Furthermore, instead of reviewing the literature, the authors have mainly discussed in the models and methods. I suggest authors to mainly focus on the gaps in existing research.
2. Methods are well discussed and described, however, the clear description of the used method is missing. The authors have failed to discuss the importance of the used four different methods.
3. Results are surfacely written. The scientific soundness of the results is difficult to capture. I suggest to support the results more through discussion.
4. I am bit surprised why the conclusion is only focused on GCN-BiLSTM not on the other models.
Author Response
[Comment 1] The introduction is not sufficient enough to describe the importance of the research in the target study area. Furthermore, instead of reviewing the literature, the authors have mainly discussed in the models and methods. I suggest authors to mainly focus on the gaps in existing research.
[Response] Thank you for your precious suggestion. We have re-written section 1 to describe the importance of this research in the target study area from line 38 to 71. In the section 1 and section 2, we have added the literatures to review previous studies and discussed the limitations of existing researches.
[Comment 2] Methods are well discussed and described, however, the clear description of the used method is missing. The authors have failed to discuss the importance of the used four different methods.
[Response] Thank you for your comment. We are very sorry for the unclear description. We have added the description from line 351 to 359.
[Comment 3] Results are surfacely written. The scientific soundness of the results is difficult to capture. I suggest to support the results more through discussion.
[Response] Thank you for your precious suggestion. We have added the discussion on the results from line 427 to 458.
[Comment 4] I am bit surprised why the conclusion is only focused on GCN-BiLSTM not on the other models.
[Response] Thank you for your precious suggestion. In Section 4.5, we have added a comparative discussion on GCN-BiLSTM model and other baseline models from line 427 to 458.
Round 2
Reviewer 2 Report
My comments are mostly considered in the revision. Spatial visualization of the temperature prediction results is strongly recommended to be added and analyzed as it is the core of this paper.
Author Response
[Comment 1] My comments are mostly considered in the revision. Spatial visualization of the temperature prediction results is strongly recommended to be added and analyzed as it is the core of this paper.
[Response] Thank you for your precious suggestion. According to your suggestion, we have added this part from line 477 to 491, from Line 520 to 522, and Line 494.
Reviewer 3 Report
Authors have addressed most of the comments. However, I am still not satisfied with the methodology and the conclusion drawn in the manuscript.
My decision is rejection.
Author Response
[Comment 1] Authors have addressed most of the comments. However, I am still not satisfied with the methodology and the conclusion drawn in the manuscript.
[Response] We are very sorry to make you feel unsatisfied. In fact, we have revised the manuscript very carefully one by one in the first round of revisions based on your comments. In 2nd round of revision, we have added spatial visualization of the temperature prediction results and analyzed detailly. We hope you will be satisfied with it. Thank you very much.
