Research and Application of Intelligent Weather Push Model Based on Travel Forecast and 5G Message

Round 1

Reviewer 1 Report

In this study, the authors analyzed a large-scale user travel data and developed an application for predicting users' future locations and times of arrival. They utilized this application to send weather information using a 5G messaging platform. Though the authors' motive is scientifically strong, I am not sure how this study will fit into atmospheric research journals, as the authors are focused on predicting the location and time of arrival of users and just sending weather information based on the predicted location. Since the current analysis is limited to location prediction and time of arrival models and does not focus on weather prediction-related aspects, I advise the authors to include the weather prediction aspect to align with the objectives of the journal "Atmosphere." My major comments are provided below. 

1.      In this study, it is not clear which model-based weather information the authors have considered for providing to users. There are no weather model details provided in this manuscript.

2.      Figure 1: Is this a model predicted data, if so there is no validation of the provided weather information with observed measurements.

3.      Table 3 and 4: From the caption of tables, I thing authors are showing Shanghai meteorological observations against the forecasts, if so the two table times are not matching.

4.      In my view, personalized or customized weather data for users denotes customized weather parameters (temperature, rh, wind, rainfall, chill index etc.,) alone but not the location based all the weather information.   

5.      The authors did not make an attempt to forecast weather parameters using AI/ML-based models. They should incorporate this aspect and translate the text using more meteorological terminology to cater to a wider readership in this journal.  

6.      The authors frequently used the term "5G messaging platform," but they did not provide any justification for why 5G service was chosen for this purpose and why 4G couldn't be employed instead.

7.      There are significant constraints in the practical application of this type of location prediction models, such as privacy concerns related to user movement tracking. However, the authors did not include any limitations of this study in this manuscript. 

Author Response

Dear Reviewer,

Thank you very much for your thorough review and valuable feedback on our manuscript. We have carefully considered each of your suggestions and made detailed revisions to better align with the requirements of the journal. Below are our specific responses:

1. We have added a detailed description of the weather models used, including their working principles and the data sources utilized, in Section 6. This addition aims to provide readers with a better understanding of the weather information provided.

2. We have refined Figure 1 to include both predicted weather data from the model and actual observed data, ensuring the accuracy and reliability of the model's weather service predictions.

3. We have aligned the time frames for the Shanghai meteorological observations and forecasts in Tables 3 and 4, ensuring consistency and accuracy in the data comparison.

4. We have further defined and clarified the personalized or customized weather data to specifically refer to the personalization of certain weather parameters such as temperature, relative humidity, wind speed, precipitation, and chill index.

5. We have standardized the use of meteorological terminology throughout the manuscript to enhance the professionalism and cater to a broader readership.

6. In the introduction section, we have provided a detailed explanation for choosing the 5G messaging platform over 4G, emphasizing the specific advantages and capabilities of 5G technology in delivering the proposed services.

7. In the final section of the manuscript, we have included a discussion highlighting the limitations of the practical application of the location prediction model, particularly concerning privacy issues related to user movement tracking. This addition aims to provide readers with a better understanding of the practical constraints and limitations of the research.

Once again, we appreciate your review and guidance. We value your input greatly, and we believe that these modifications will significantly improve our manuscript. Please feel free to reach out to us if you have any further suggestions or comments.

Best regards,

Fengchen Fu

Reviewer 2 Report

1- Abstract: 

In summary, the abstract introduces an intriguing approach, but could benefit from a more in-depth discussion on the context, data utilized, limitations, and practical implications of the study. Additionally, comparisons with existing methods and a clearer geographical contextualization could enhance the robustness of the proposed approach.

Practical Impact: While the model appears statistically sound, the abstract does not specify how this could have a practical impact for end-users.

Geographical Scope of the Study: The abstract does not specify whether the study is limited to Chengdu or if the results can be generalized to other regions or cities.

2- Introdction : 

Your introduction provides a precise and detailed overview of the issues related to personalized weather forecasting and emphasizes its significance in individuals' daily lives. You clearly outline the current problems regarding the dissemination of weather forecasts and highlight the positive impact a personalized approach could have. However, here are some areas that could be improved:

- The introduction is quite lengthy and contains a lot of information. Consider breaking it into shorter paragraphs to enhance readability.

- While you articulate the problems very clearly, it might be helpful to add a brief literature review or mention previous studies that have addressed similar issues. This could help contextualize your research within the current landscape.

- Although you briefly mention the aim of your research, you could further specify it. What are the specific questions you are seeking to address?

-Consider adding a sentence or two to briefly introduce the methodology you will use to achieve your objectives.

Overall, your introduction lays a solid foundation for your research. By clarifying certain points and improving the flow of the text, you can further strengthen your argumentation.

3. Related Work: Methodology: While you mention the use of the location prediction model based on the XGBoost algorithm, providing a brief justification for this choice would be beneficial. This could include discussing the algorithm's proven effectiveness in similar applications or its ability to handle complex data patterns, which align with the objectives of your research. Providing this rationale will help establish confidence in the chosen methodology.

The section on dataset and data processing is well-detailed and clearly explains the choices made for analyzing user travel patterns.

The use of bike-sharing data as a proxy to extract individual mobility patterns is an interesting and well-justified approach.

The discussion on integrating meteorological data is relevant and shows how these data can be combined to enrich the analysis.

Data preparation is crucial and the steps of cleaning, transformation, and normalization are clearly explained, demonstrating careful attention to data quality.

The explanation of using the XGBoost algorithm for predicting users' future locations is clear and well-justified.

The analysis of features of shared bike data, such as distance and duration of rides, provides a good understanding of user behavior.

The correlation analysis of variables related to time and location is well-presented, highlighting important relationships for prediction.

The analysis of feature importance for prediction highlights the most influential factors, which is crucial for understanding the model.

The comparison of performance across different models is a key element and is well-executed, providing a quantitative evaluation of predictions.

The section concludes with a discussion on model hyperparameters, showing careful consideration of model parameters.

Overall, this section is well-structured, informatively rich, and demonstrates a methodical approach in data processing and model construction.

While the conclusion provides a thorough summary of the research findings and their implications, it could benefit from a more explicit mention of any potential limitations or areas for future research. This would add depth to the conclusion and offer a more holistic view of the study.

It would be valuable to include a brief discussion on how the proposed personalized weather prediction model compares to existing methods or technologies in terms of accuracy, personalization, and practicality. This would provide context for the significance of the developed model.

While the integration of 5G messaging platform technology and machine learning algorithms is a promising approach, it would be helpful to include a brief consideration of potential challenges or considerations related to the implementation of this technology, such as scalability, user adoption, or privacy concerns. This would provide a more balanced discussion of the proposed solution.

Author Response

Reviewer 2

Dear Reviewer,

Thank you for your insightful feedback on our manuscript. We have made the necessary revisions and adjustments in response to your comments. Below are the specific improvements we have implemented for each of your suggestions:

1- Abstract:

- We have supplemented the abstract with a more comprehensive discussion of the study's background, data utilized, research limitations, and practical implications. Additionally, we have included comparisons with existing methods and provided a clearer geographical context to enhance the robustness of the proposed approach.

- In section 6.4, we have elaborated on the specific practical applications, elucidating the actual impact of the model on end-users.

2- Introduction:

- We have reorganized the introduction into shorter paragraphs to improve readability.

- While maintaining clarity on the issues, we have added a brief review of related research to contextualize our study within the current research landscape. We simplified the introduction, concluding with a clearer statement of the research objectives and the problems we aim to address.

- We further clarified the specific objectives of the research and the methodology employed.

3- Related Work and Methodology:

- We provided a detailed explanation and justification for the choice of the XGBoost algorithm, highlighting its effectiveness in similar applications and its capability to handle complex data patterns.

- We emphasized the systematic approach to data processing and model construction.

- In the conclusion, we included explicit mention of the research limitations and areas for future research, further enhancing the conclusion's depth.

- We added a discussion comparing the proposed personalized weather prediction model with existing methods or technologies in terms of accuracy, personalization, and practicality to provide a more profound understanding of the research significance.

- We included a discussion on potential challenges or considerations related to the implementation of 5G messaging platform technology and machine learning algorithms, encompassing aspects such as scalability, user adoption, and privacy concerns, for a more comprehensive discussion of the proposed solution.

We highly value your valuable insights and guidance, and we believe that these revisions will significantly enhance the quality of our manuscript. Please feel free to reach out to us if you have any further suggestions or comments.

Best regards,

Fengchen Fu

Reviewer 3 Report

The authors report on deploying a system of predicting people’s future locations using XGBoost and delivering weather reports which are customized to the user’s future locations.   Several machine learning methods are tested using shared bicycle data to learn future locations, of which XGBoost out-performed others leading to its selection.   The paper seems to lack certain details about the time scale of the learning, does it know where cyclist A went yesterday, for example, when we’re making a prediction for cyclist A today, or does the database essentially record popular routes and learns the destination from the typical starting point and initial direction for each route?  

The report was lacking in details on the type of weather information delivered and the time scale of the weather, current or future conditions is not specified; the time scale of prediction of user location prediction seems to be only 1 hour.  For very short term (0-3h) one could use a time series of weather observations, time-of-day to extrapolate using ML to predict the weather, but observations are not that closely spaced, generally 10-20 km apart or more.  As one goes further out in time you’d require use of a gridded numerical weather prediction model; the weather model will not be able to forecast with any accuracy weather that will differ over such small distances;  it was not clear that improved forecasts or weather reports could be delivered given the travel distances on the order of 5-km whereas the most precise numerical weather prediction models available operationally today are made on grids with spacing on the order of 1- to 3-km, but are, realistically, only able to resolve weather at scales 6-8 times the grid spacing.  

These methods would be improved if the weather forecasts could be more accurately calibrated for specific locations to include micro-climates, especially where steep terrain and/or urban effects may be present, and if learning of longer commutes and other routine travel behavior could be done using past user cell phone GPS positions, etc, or more simply just ask “where are you going today?”, with a menu of destinations based on past behavior plus an option for custom destination.

Specific Comments

11)      Line 511, Table lacks units for MAE and RMSE (latitude and longitude degrees?, km?)

22)      Line 535, Table lacks units for MAE and RMSE

33)      Line 552 and ff, in this section there is a decided lack of information.  Is the weather information about current conditions, short term forecasts, or something else?  What is the source of the and scale of the input weather information?  Has it been downscaled using historical data to this spatial scale?  See general comments above.

44)      Line 572 and ff, Figure 6.  Understand this is a screenshot of a Chinese app, but being an English language publication a caption with some more detail would be helpful to the English-language reader.

Writing acceptable.   Figure 6 caption needs more detail in English.

Author Response

Dear Reviewer,

Thank you for your insightful feedback on our manuscript. We have made the necessary revisions and adjustments in response to your comments. Below are the specific improvements we have implemented for each of your suggestions:

1. Regarding the specifics of the learning time scale and weather information delivery:

- We have provided additional explanations regarding the learning time scale, including specific details on the database recording method.

- For the time scale of weather information delivery, we have offered a more detailed explanation, including the types of weather information, time scales, and data sources.

2. Improvements on the accuracy of weather forecasting and location specificity:

- We have further discussed the precision of weather forecasting, particularly when considering microclimates and terrain factors, proposing more refined forecasting methods.

- To address the spatial scale issue in the weather model, we have proposed more precise spatial scaling methods to better accommodate the meteorological characteristics of different regions.

3. Refinements on the manuscript details:

- We have added specific unit descriptions for MAE and RMSE in the tables, including degrees of latitude and longitude as well as kilometers.

- For the weather information section, we have included detailed descriptions of the weather data sources and data scales to enhance readers' understanding of the data sources and processing methods.

- Regarding the English description of Figure 6, we have added more detailed explanations to help English-speaking readers better understand the content of the figure.

In the revised Discussion and Limitations section (Section 8) of our manuscript, we have addressed the constraints imposed by the shared bicycle dataset on the prediction of location distances and times. We have highlighted how the utilization of spatial-temporal data obtained from 5G terminal feedback can overcome the limitations of the prediction time scale and spatial positions, facilitating the generation of personalized weather reports synchronized with the corresponding spatial-temporal data for user dissemination.

We highly value your insightful comments and suggestions, and we believe that these revisions significantly enhance the quality of our manuscript. Please feel free to reach out to us if you have any further suggestions or comments.

Best regards,

Fengchen Fu

Round 2

Reviewer 1 Report

Authors have addressed all comments raised during the first review. I therefore recommend for the publication of the manuscript.

Reviewer 2 Report

The authors have provided responses to all my comments