Improving the ERA5-Land Temperature Product through a Deep Spatiotemporal Model That Uses Fused Multi-Source Remote Sensing Data

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Please consult attached document

Comments for author File: Comments.pdf

Author Response

Response to review comments

 

We are very grateful for your thorough review and insightful feedback on our manuscript. Your advice on the manuscript title, your suggestion to include the study's contributions and add a comparison with existing research, and your recommendation to elaborate on the study's limitations were extremely helpful. We have carefully considered all your comments and made revisions accordingly. In the following section, we provide our responses to each of your comments.

 

(1) Title: I suggest making some small modifications to the title since it does not faithfully

describe what the authors propose. I suggest including the term ERA5-LAND

Response:

Thank you for your suggestion. According to your suggestion, we revised our manuscript's title to: "Improve the ERA5-Land temperature product through a deep spatiotemporal model that uses the fused multisource remote sensing data "

(2) Introduction: I suggest clearly indicating what the GAP of the research is and the contributions made by the study. These are not clear in the document presented. I also suggest including a final paragraph where the authors describe how the study may affect future research.

Response:

Thank you for the reminder, and I apologize for making the review process an unpleasant experience. We have added our research gap to the Introduction Section, specifically lines 131-133, and we elaborate on it in Section 4.5. In the final paragraph of the Introduction Section, lines 128-131, we clarified our research contributions and how our study will impact future research. In the Conclusion Section, lines 437-445, we further elaborate on our research’s contributions and their impact.

(3) Section 2.2: I suggest including a link to the download of the ERA5 and LULC data as there is with the rest of the data. What ortho-rectification and calibration process have the authors carried out with the ERA5 images?

Response:

Thank you for your suggestion. We have added download links for each dataset in Table 1, below the data source.

ERA5-Land integrates various remote sensing and geographic information sources, with detailed preprocessing and corrections already applied. Thus, no further radiation correction or processing is required. This description has been added to Section 2.2, lines 159-161.

(4) Section 3.2. What software did the authors use with the ERA5 images to apply the Neural Network (CNN) to it?

Response:

Thank you for your question. In the final paragraph of Section 3.3, lines 249-251, we have illustrated that we use PyCharm with the TensorFlow library to build a CNN-BiLSTM model to process the ERA5-Land grid temperature data.

(5) The document presented does not have a section for discussion of the results. This is essential in order to validate the results by comparing them with another research already carried out. Therefore, I suggest that the authors include a section for discussion of the results.

Response:

Thank you for your suggestion. We have added Section 4.5: "Compare with existing research results" to compare our research results with existing research results. The gap in our research mentioned in your second review is also explained in detail in this Section.

(6) Do the LULC data downloaded have a verification or validation process? The authors followed some process to validate these results, for example, a confusion matrix by comparison with satellite images.

Response:

Thank you for your question.

The LULC data used in our research have been verified for accuracy using a confusion matrix by the European Center for Medium-Range Weather Forecasts (ECMWF), eliminating the need for further verification in this study. This description has been added to Section 2.2, lines 163-165.

(7) I suggest including a final section called limitations to the study where the authors establish what the limitations of their research are.

Response:

Thank you for your suggestion. In the final paragraph of the Conclusion Section, we have elaborated on our research limitations. In general, our model does not perform as well in the high-latitude northwest of the Yellow River Basin compared to other areas, and the model may not accurately analyze extreme weather events in the Yellow River Basin. This is discussed in lines 445-453.

 

 

Please allow us to extend our heartfelt thanks once more for your meticulous review of our paper, as well as your invaluable comments and constructive suggestions. Your comments have significantly improved our manuscript, enhanced its scientific value, and provided substantial help for our future research work. Should you have any additional thoughts or suggestions after reviewing the revised manuscript, we would be delighted to hear from you. We look forward to your continued feedback!

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Remote Sensing-3113708 

Improve the temperature reanalysis product through depth spatial-temporal model that use the fuse multisource remote sensing data

 

This study proposes to improve temperature reanalysis products based on deep learning and the fusion of multi-source remote sensing and geographic data. Specifically, convolutional neural networks (CNN) and bidirectional long short-term memory networks (BILSTM) are used to capture the spatial and temporal relationship between temperature and DEM, land cover and population density. A temperature deep spatiotemporal model is established to improve the temperature reanalysis products and obtain temperature products with higher resolution and accuracy. The comparison with the measured temperature at the meteorological station shows that the accuracy of the improved Era5-Land product can be effectively improved (The mean absolute error (MAE) value decreased by 28.7%, and the root mean square error (RMSE) decreased by 25.8%).

 

General comments: The paper is well presented and well analyzed but the grammatical errors in the paper make it difficult to understand. Some of the figures are also not clear due to which the paper needs to be extensively modified. The specific comments are given below.

 

Specific comments

Second paragraph of Introduction: This paragraph content is fine but has a lot of grammatical errors.

 

Line 86-88: Based on the spatial relationship between land cover data and temperature, temperature prediction and temperature products can be realized calibration and spatiotemporal model building [23,24].

What is the meaning of this sentence?

 

Line 92-94: For example, processing massive data with diverse types and complex features to establish accurate models to make accurate classification and predictions[25,26].

This is not a sentence.

 

Line 95-100: The sentences are very badly written.

 

Line 101-105: The sentence is very badly written.

 

Line 122-124: The temperature in the Yellow River Basin is rising overall, and the rate of warming in the midstream and downstream is slowly than in the upstream[34].

This sentence is difficult to understand because of the style of writing. 

 

Line 141-142: It uses radar beams to measure the earth's surface through INSAR on the

space shuttle.

What is the full form of INSAR and which one? Which space shuttle?

 

Line 160-161: Extract data within the research range and unify the spatial scale to 0.05°.

How did the authors unify the spatial scale to 0.05°, since all the datasets are not of the scale 0.05°.

 

Line 181-185: Authors should either use 0.05° x 0.05° or 4.83 km×4.83 km. They should avoid using both.

 

Section 4.1 is well written and explained Section 4.2 is not well written and difficult to understand with the use of Figure 7.

Comments on the Quality of English Language

The quality of English is not good and difficult to understand.

Author Response

Response to review comments

 

We greatly appreciate your professional review of our article. Your detailed comments, especially regarding grammatical errors and issues with the charts, have been invaluable. We made every effort to improve the manuscript based on your feedback and have implemented several changes. We carefully considered all of your comments and revised the manuscript accordingly. Additionally, we scrutinized the manuscript to correct any typos and grammatical errors we found. In the following sections, we provide our responses to each of your comments.

 

(1) Second paragraph of Introduction: This paragraph content is fine but has a lot of grammatical errors.

Response:

Thank you for your comment. I apologize for the poor reading experience caused by my writing. We have carefully modified this part, and the revised content is as follows: "Temperature data from meteorological stations were collected using sensors placed in thermoscreens 1.5 to 2 meters above the ground. However, due to terrain and other factors, the distribution of meteorological stations is uneven, and their coverage is limited, complicating the monitoring and predicting temperature changes. In China, meteorological stations are densely concentrated in the central and eastern regions but are sparse in the western and northwestern areas" In lines 47-52 of our manuscript.

(2) Line 86-88: Based on the spatial relationship between land cover data and temperature, temperature prediction and temperature products can be realized calibration and spatiotemporal model building [23,24].

What is the meaning of this sentence?

Response:

I apologize for the poor reading experience caused by my writing. We have carefully rewritten the sentence to make it more relevant to the context and to clarify its meaning. The revised paragraph is as follows: "Temperature and land cover type have a close spatiotemporal relationship. For example, changes in the proportion of vegetation-covered land in cities will significantly affect urban temperature. By capturing this spatial relationship, it is possible to calibrate temperature products and establish corresponding temperature-land cover type spatiotemporal models [23,24]. " In lines 84-88 of our manuscript

(3) Line 92-94: For example, processing massive data with diverse types and complex features to establish accurate models to make accurate classification and predictions [25,26].

This is not a sentence.

Response:

I apologize for the poor reading experience caused by my writing. We have carefully rewritten this sentence. The revised paragraph is as follows: "For example, deep learning technology can process massive amounts of data with various types and complex features in image recognition and natural language processing, establish accurate models, and make accurate classifications and predictions [25,26]. " In lines 91-94 of our manuscript.

(4) Line 95-100: The sentences are very badly written.
Response:

I sincerely apologize for the poor reading experience caused by my writing. We reconsidered the content, corrected grammatical errors, removed incorrect information, and expanded the text to enhance clarity and better align it with the context. Please refer to lines 95-106 in the manuscript for the revised content.

(5) Line 101-105: The sentence is very badly written.

Response:

I sincerely apologize for the poor reading experience caused by my writing. We have rewritten this paragraph and expanded the content to enhance clarity and better align it with the context. Please refer to lines 107-115 of the manuscript for the revised content.

(6) Line 122-124: The temperature in the Yellow River Basin is rising overall, and the rate of warming in the midstream and downstream is slowly than in the upstream [34].

This sentence is difficult to understand because of the style of writing.

Response:

We rewrote the sentence to make it easier to understand. The revised sentence is as follows: "The overall temperature in the Yellow River Basin exhibits an increasing trend, with the warming rate in the upper reaches being significantly higher than in other regions. [35]. " In lines 145-147 of our manuscript.

(7) Line 141-142: It uses radar beams to measure the earth's surface through INSAR on the space shuttle.

What is the full form of INSAR and which one? Which space shuttle?

Response:

The full form of INSAR is "Interferometric Synthetic Aperture Radar". The type used is Single-Pass INSAR, carried on the Space Shuttle Endeavour. In the 2.2 section, lines 166-171.

(8) Line 160-161: Extract data within the research range and unify the spatial scale to 0.05°.

How did the authors unify the spatial scale to 0.05°, since all the datasets are not of the scale 0.05°

Response:

Using nearest neighbor interpolation to resample the ERA5-Land data, land cover data, and Land Scan Global data to 0.05°. Using bilinear interpolation to resample DEM data to 0.05°.In the 3.1 section, lines 204-207.

(9) Line 181-185: Authors should either use 0.05° x 0.05° or 4.83 km×4.83 km. They should avoid using both

Response:

Thank you for your reminder. We have deleted the "4.83 km×4.83 km", all use the form0.05° x 0.05°.

(10) Section 4.1 is well written and explained Section 4.2 is not well written and difficult to understand with the use of Figure 7.

Response:

I apologize for the poor reading experience caused by my writing. We deleted the original Section 4.2, revised the grammar carefully, and rewrote the section. We also elaborated on Figure 7, now Figure 8,in more detail, including a detailed explanation of its content in Section 4.2 , lines 341-348, and replacing the figure's title with a more understandable one. The new title of the figure is " Left: CNN-BiLSTM MAE, RMSE; Right: The percentage reduction of MAE and RMSE of CNN-BiLSTM compared to the MAE and RMSE values of ERA5-Land; De (%): Decreased (%)".

 

Please allow us to thank you again for your careful reading of our paper, as well as for your very helpful comments and constructive suggestions. Your feedback not only enhanced the scientific quality and readability of our manuscript but also provided us with valuable insights for our future research. If you have any new ideas or suggestions after reviewing the revised manuscript, please feel free to contact us. We look forward to hearing from you!

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This paper introduces an advanced deep spatial-temporal model for temperature fusion by integrating Convolutional Neural Networks (CNN) with Long Short-Term Memory (LSTM) networks. It validates the model using multi-source data, demonstrating superior accuracy compared to the original ERA5-Land dataset. Additionally, the study analyzes the trend of average temperatures during the cold months in the Yellow River Basin from 2015 to 2019. However, while the use of CNN-LSTM or CNN-BiLSTM models is already quite common, and utilizing data such as DEM, Land Cover, and Population Density to analyze temperature or air quality is also typical, the innovative point of the article is not clearly described. Please provide a more detailed description of the novelty. Furthermore, there are several other issues:

1.      The author does not provide a sufficient description of the motivation for using CNN and LSTM in the introduction, and the overview of these methods is too brief.

2.      The abstract describes the experimental method as CNN and BiLSTM, but there is no mention of BiLSTM in the introduction. Additionally, the first point of the research objectives in line 105 describes combining CNN and LSTM, while the experimental results section refers to the method as CNN-BiLSTM. Please clarify the relationship between these terms and ensure consistency between the experimental methods and the experimental results.

3.      In "Figure 5", what is the significance of using ten-fold cross-validation to compare Fold and ERA5-Land? Please provide a necessary explanation. Also, explain why ten-fold cross-validation enables the model to progressively learn all features of the fused data, as discussed in lines 263-265 and 271-273.

4.      The paper does not present the model training process in detail. Please provide some necessary explanations of the model training process.

5.      It is advisable to enhance Figures 6, 8, and 9 with key annotations. For example, labeling “ERA5-Land MAE” on Figure 6(a) and “2015” on Figure 9(a) would improve clarity.

6.      This paper has numerous grammar and language issues, which need to be addressed.

Comments for author File: Comments.pdf

Comments on the Quality of English Language

This paper has numerous grammar and language issues, which need to be addressed.

Author Response

Response to review comments

We sincerely thank you for your thorough review and insightful comments on our manuscript. Your feedback has enhanced the scientific rigor and cutting-edge nature of our work. Your suggestion to elaborate on the novelty of our research has provided us with a clearer direction for improvement. Additionally, your recommendations to further detail the CNN and LSTM methods and to clarify the 10-fold cross-validation process have increased the rigor of our paper. Your comments on modifying our figures have improved the clarity of our research results. Your suggestions will undoubtedly contribute to the overall quality of our manuscript. In the following sections, we list our responses to each of your comments.

 

(1) Additionally, the study analyzes the trend of average temperatures during the cold months in the Yellow River Basin from 2015 to 2019. However, while the use of CNN-LSTM or CNN-BiLSTM models is already quite common, and utilizing data such as DEM, Land Cover, and Population Density to analyze temperature or air quality is also typical, the innovative point of the article is not clearly described. Please provide a more detailed description of the novelty

Response:

Thank you for your detailed comments. In general, our research novelties are: (1) Most existing studies on the relationship between temperature and other geographical elements consider only one factor, such as DEM or land cover. In contrast, this study comprehensively considers multiple factors, including DEM, population density, and land cover, and uses deep learning neural networks to capture the complex spatiotemporal relationships and establish a corresponding model. (2) The results of this study in the upper reaches of the Yellow River show that the model can still obtain high-precision temperature grid data in areas with fewer meteorological stations, addressing the issue of not being able to acquire high-precision temperature datasets with continuous spatiotemporal distribution due to insufficient meteorological stations.

We have added a detailed description of the novelty in Section 1 (Introduction), lines 107-111 and 124-128. Additionally, we have further clarified our research novelty in Section 4.1, lines 329-333, and Section 4.3, lines 379-381.

(2) The author does not provide a sufficient description of the motivation for using CNN and LSTM in the introduction, and the overview of these methods is too brief.

Response:

Thank you for your comment. We added a detailed description of the motivation for using CNN and BiLSTM, and elaborated on their characteristics in Section 1 (Introduction), lines 95-106. Additionally, we added a new reference [42] in Section 3.2, lines 245-248, for further explanation

(3) The abstract describes the experimental method as CNN and BiLSTM, but there is no mention of BiLSTM in the introduction. Additionally, the first point of the research objectives in line 105 describes combining CNN and LSTM, while the experimental results section refers to the method as CNN-BiLSTM. Please clarify the relationship between these terms and ensure consistency between the experimental methods and the experimental results.
    Response:

Thank you for your reminder! I sincerely apologize for the inconvenience caused by the inconsistent names of the model in my research methods and results. Bidirectional Long Short-Term Memory (BiLSTM) is an extension of LSTM. We have unified the names of all models to "CNN-BiLSTM". We have added clarity of BiLSTM in the Introduction Section, lines 102-104.

(4) In "Figure 5", what is the significance of using ten-fold cross-validation to compare Fold and ERA5-Land? Please provide a necessary explanation. Also, explain why ten-fold cross-validation enables the model to progressively learn all features of the fused data, as discussed in lines 263-265 and 271-273.

Response:

Thank you for your detailed comments. By comparing the results of each fold in the ten-fold cross-validation with ERA5-Land, we can assess the model's generalization ability across different data subsets. Additionally, comparing the error fluctuations across different folds helps evaluate the model's stability. We have added these explanations in the Section 4.1, lines 311-315.

The ten-fold cross-validation divides the fused data into ten subsets. In each training process, nine subsets are used as training data, and the remaining subset is used as test data. The training and test sets are input into the model for training. This process is repeated ten times to ensure that each subset is used as a test set once, and the remaining nine subsets are used as training sets. This approach ensures that all features of the fused data are learned and helps avoid bias. We have added these explanations in Section 3.4, lines 283-289.

(5) The paper does not present the model training process in detail. Please provide some necessary explanations of the model training process.
    Response:

Thank you for your reminder. To illustrate the model training process in detail, we added Figure 5. This figure details how the model is divided into training and test subsets through ten-fold cross-validation, how these subsets are imported into the CNN-BiLSTM model for training and evaluation, and lists some of the model's hyperparameters. The title of Figure 5 is "The process of training the CNN-BiLSTM model using ten-fold cross-validation".

(6) It is advisable to enhance Figures 6, 8, and 9 with key annotations. For example, labeling “ERA5-Land MAE” on Figure 6(a) and “2015” on Figure 9(a) would improve clarity.

Response:

Thank you for your reminder. We have updated the labels as follows

Figure 6, now Figure 7: ERA5-Land MAE, ERA5-Land RMSE, CNN-BiLSTM MAE, and CNN-BiLSTM RMSE.

Figure 8, now Figure 9: 0.05° ERA5-Land, Original ERA5-Land, CNN-BiLSTM, and Meteorological Station.

Figure 9, now Figure 10: 2015, 2016, 2017, 2018, and 2019."

(7) This paper has numerous grammar and language issues, which need to be addressed

Response:

Thank you for your comment. We have carefully re-checked the manuscript, corrected grammatical errors, and rewritten unclear sentences.

 

Please allow us to express our gratitude once again. Your detailed and constructive comments and suggestions have greatly improved the overall quality of our manuscript, enhanced the scientific rigor of our research, and provided valuable guidance for my future work. If you have any further thoughts after reviewing the resubmitted manuscript, please feel free to contact us!

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have revised the paper to further emphasize its novelty and clarify the relationships between the relevant terms, ensuring consistency between the experimental methods and results. Additionally, Figure 5 has been added in Section 3.4 to illustrate the model training process. I have no further comments on the paper after these revisions.

Comments for author File: Comments.docx

Author Response

We are deeply grateful for your thorough review and thoughtful comments on our article. Your positive feedback on the revised content is a great encouragement to our research. We would like to express our sincere thanks again for your valuable insights and constructive suggestions. If you have any further ideas or suggestions in the future, please do not hesitate to contact us!

Author Response File: Author Response.pdf