The MS is an excellent offering in a complex and demanding area of applied science and i compliment the authors on assembling the doc and even more for the background work associated with the project.

On reading the abstract and Introduction, however there are numerous challenges of interpretation of the US English and needs for clarity and clarification. Additionally the grammar and sentence style varied form section to section and really needs skilled attention to enable uniformity of style, especially to minimise the complexity of some the the more detailed sentences for the 'average' reader of your journal.

The attached doc of my comments is not in MS Word 'review' style, but a copy-paste to Word from the original PDF.

The red sections and alterations are my considered suggestions and deductions.  I hope the content of the doc is sufficiently clear for the authors to understand. If not I will gladly clarify by return email (or whatever method is appropriate in these circumstances).

• Some lake water quality observation datasets in the United States are released in the manuscript. These datasets can be used to study the spatial and temporal dynamics of some lakes with varying water quality, ecosystem and landscape as well as the lake water quality prediction.
• Fig. 1 shows 21 lakes. More information of these lakes should be further provided, like depth, area, elevation, etc.
• In situ observations of these lakes should be preprocessed based on some unified standards. What standards are used? How to process these in situ data?
• How to define the benchmark dataset? What characteristics should be satisfied? How to verify the dataset?
• Some applications to ecosystem, landscape and the lake water quality prediction can be further shown.

Review comments

This paper describes a lake water quality baseline dataset called LakeBeD-US, which integrates long-term monitoring data from 21 lakes in the United States to provide a baseline for lake water quality time series and vertical profile studies. This paper describes in detail the construction process of data set, data structure and application cases in machine learning. The construction of this dataset is of great significance for advancing the study of lake water quality, providing a valuable data resource for ecologists and computer scientists. The article is rich in content, clear in logic, rigorous in method, and has high academic value.

1. Advantages

Innovative and valuable dataset: The LakeBeD-US dataset is the first benchmark dataset dedicated to time series and vertical profiles of lake water quality, filling a gap in the field. It contains more than 500 million unique water quality observation data, covering 17 water quality variables, with extremely high data volume and diversity, providing rich data support for relevant research.

Multidisciplinary integration: The dataset is not only suitable for ecological research, but also optimized for computer science, which can promote interdisciplinary cooperation between ecology and computer science, and promote the application of knowledge-guided machine learning (KGML) in lake water quality research.

Data integrity and standardization: The author describes in detail the source of the data, the processing process and the quality control method to ensure the accuracy and reliability of the data. At the same time, the data set follows the FAIR principle, which is easy to find, access, interoperate and reuse.

Practicability of application cases: Taking Lake Mendota as an example, this paper demonstrates how to use LakeBeD-US: Computer Science Edition to make multivariate time series prediction, verifies the feasibility and effectiveness of data set in machine learning tasks, and provides reference for subsequent research.

1. II. Disadvantages

Data representation issues: Although the dataset covers 21 lakes, 13 of them are located in Wisconsin, which may limit the geographic representation of the data to some extent. It is suggested that the authors consider including more lake data from different geographic regions in subsequent work to improve the general applicability of the dataset.

Processing of missing data values: It is mentioned that there are a large number of missing values in the data set, and although this is a common problem in environmental data, the authors adopt the SAITS method in processing missing values, and do not discuss its impact on the results in detail. It is suggested that the authors further analyze the potential impact of missing values on model performance and explore other possible treatment methods.

Adequacy of model validation: In the machine learning application case, the authors used the LSTM-RNN model for prediction, but only reported normalized and unstandardized RMSE metrics. It is recommended to add more model validation indicators, such as coefficient of determination (R²), mean absolute error (MAE), etc., to evaluate the performance of the model more comprehensively.

Generalizability of the conclusions: The conclusions of this paper are mainly based on data from Lake Mendota and may not be fully generalizable to other lakes. It is suggested that the authors further explore the applicability and limitations of the data set and model across different lakes in the discussion.

Data update and maintenance: The timeliness and accuracy of data sets depend on long-term updates and maintenance. It is recommended that the author clarify the plan and mechanism of data update to ensure the continuous availability of the data set.

III. Comprehensive Suggestions

The LakeBeD-US data set proposed in this paper has important scientific value and application prospect, and it is suggested that the author should modify and perfect it according to the above opinions, so as to further improve the quality and influence of the paper. In particular, it is necessary to strengthen the discussion of data representation, model validation and generalization of conclusions, so as to provide readers with more comprehensive and reliable research results.

The following is a more professional review of this paper, which aims to help the research be more scientific and rational:

1. Data quality control

Data cleaning details: It is mentioned that the data set is cleaned and the format is converted, but the specific data cleaning method is not detailed. It is recommended that the author describe the data cleaning process in detail, including how to deal with missing values, outliers, duplicate values, etc., and what algorithms or tools are used to carry out data cleaning to ensure the accuracy and reliability of the data.

Data quality assessment: Authors should provide more detailed data quality assessment indicators and methods, such as data integrity, accuracy, consistency, etc. A data quality assessment tool or framework can be adopted to systematically assess data quality and report the assessment results in a paper.

Data traceability: In order to ensure the traceability of the data, it is recommended that the author record the metadata information such as the source, collection time, collection method, and processing step of the data in detail, and publish it together with the data set, so that other researchers can clearly understand the generation process and quality status of the data.

1. Data set construction method

Data fusion method: The paper mentions the fusion of data from different sources, but does not specify the specific method and algorithm of data fusion. It is recommended that the author describe the process of data fusion in detail, including how to resolve conflicts between different data sources, how to align and integrate data, etc., to ensure the consistency and availability of data sets.

Data sampling strategy: The author should further explain the strategy and method of data sampling, including sampling frequency, time range, spatial coverage, etc. In particular, for the sampling of high and low frequency data, it should be explained how to ensure that the sampling adequately captures the dynamic changes in lake water quality while avoiding data redundancy.

The science of data set construction: It is suggested that the author discuss the science and rationality of data set construction in more depth, such as whether the representation of lakes, randomness and systematization of data are considered. Statistical analysis methods or machine learning techniques can be employed to verify that the construction of data sets meets scientific requirements.

1. Model verification and evaluation

Model selection basis: In the machine learning application case, the author chose the LSTM-RNN model for prediction, but did not specify the basis and advantages of choosing this model. It is recommended that the authors analyze and compare the applicability of different models in the prediction of lake water quality in more depth, explain why the LSTM-RNN model is chosen, and discuss its advantages and disadvantages compared with other models.

Model validation indicators: Currently, only standardized and unstandardized RMSE are used as model validation indicators. It is suggested to add more validation indicators, such as coefficient of determination (R²), mean absolute error (MAE), root mean percentage error (RMSPE), etc., in order to evaluate the performance and generalization ability of the model more comprehensively.

Interpretability of the model: To improve the interpretability of the model, authors are advised to use methods such as feature importance analysis, locally interpretable model-independent exPlanations (LIME), or SHapley Additive exPlanations (SHAP) to explain the model's predictions and key influencing factors.

1. Generalizability of conclusions

Multi-lake validation: The current conclusions are mainly based on the data of Lake Mendota. It is suggested that the author validate and apply the model in more lakes to test the universality of the conclusions and the generality of the model. This can be done in collaboration with other lake research institutions or by leveraging existing lake datasets.

Applicability under different environmental conditions: Considering the differences of lakes in different geographical regions, climatic conditions, ecological types, etc., it is suggested that the author explore the applicability and performance of the model under different environmental conditions, and analyze the limitations and improvement directions of the model.

Long-term forecasting ability: The current forecasting task mainly focuses on short-term forecasting, and it is suggested that the authors further study the performance of the model in long-term forecasting and how to improve the long-term forecasting ability of the model, which is of great significance for lake water quality management and policy making.

The following are the review comments on the language spelling errors in this paper:

1.“LakeBeD-US contains over 500 million unique observations of lake water quality collected by multiple long-term  monitoring organizations across 17 water quality variables in 21 lakes in the United States. "It is suggested to replace" in "with" across ", To more accurately represent data coverage.

2.“Water quality varies across spatial and temporal scales due to a variety of interacting physical and biological processes. "It is suggested that" due to "be changed to" as a result of "to enhance the formality of the sentence.

3.“For example, hypolimnetic anoxia (low oxygen) in lakes decreases habitat for cold-water fish species. "It is suggested to change" decreases "to" reduces ", To express the meaning of reduction more accurately.

4.“Water quality is also a determinant of ecosystem services that lakes provide,  and thus is of great interest to ecologists. "It is suggested to change" is of great interest "to" is highly significant "to enhance the strength of the expression.

5.“To meet this challenge, we need scalable water quality models that are supported by observational data of sufficient spatiotemporal resolution  to recreate key water quality dynamics. "It is suggested that" recreate "be changed to" reproduce "to more accurately express the meaning of the reproduction.