General comments

This article uses a practical method based on multimodal data to construct the first global scale 3D information dataset of buildings. The data set provided by this study fills the gap in fine-grained building height data globally, which is of great significance for urban morphology research and climate change analysis. The model validation results are comprehensive and promising; however, a more detailed explanation of the technical methods would enhance the paper's clarity (see specific comments). Overall, the paper is well-structured, and the dataset is valuable to urban studies.

Specific comments

1. The article uses multiple sources of data for analysis, and it is recommended to add ablation experiments between different data to demonstrate the effectiveness of using the data.
2. Why did you choose to use XGboost instead of random forest or support vector machine? Please provide additional experiments or explanations.
3. Please explain the specific operation of manual measurement in section 4.2 and the basis for the authenticity of manual height measurement values.
4. What are the advantages of building scale 3D data over coursescale 3D data set?
5. Please explain in detail how to aggregate building scale height data to coarse resolution scalesfor validation.
6. What are the contributions of statistical values in the model?

This paper presents the world's first three-dimensional building footprint dataset, 3D-GloBFP, which integrates multi-source remote sensing data and various reference building height data. By employing machine learning methods, it generates high-precision global building height data. This dataset holds significant importance across multiple application domains, including urban planning, environmental monitoring, disaster management, and energy consumption analysis. The research demonstrates substantial promise and value, providing a crucial foundation for the acquisition and application of global 3D building data. I believe that 3D-GloBFP is an indispensable foundational dataset for urban research. During the review process, I identified several areas that require further clarification and improvement.

1. In Section 3.2.1 "Division of Subregions," the information of training and testing samples (e.g., the total amount) for each sub-region should be explicitly provided. Clearly specifying the selection criteria and distribution of these samples will help readers better understand the process of model training and validation.
2. I noticed that there are some missing tiles in Ghana and incomplete regions in Guangdong, Please ensure that the dataset is complete globally.
3. In China, reference building heights used as training samples were mostly concentrated in city centers. Please clarify the accuracy of the model's estimates for building heights in different urban areas (e.g., urban fringe). This is crucial for validating the model's applicability in diverse environments.
4. For high-rise buildings, especially super tall buildings, what are the potential reasons for height underestimation? It is recommended to include a detailed error analysis and explanation.

The study by Che et al. represents a significant advancement in the field of urban geography and Earth observations. The development of the 3D-GloBFP dataset is a groundbreaking achievement that fills a critical gap in the availability of global, high-resolution, and accurate building height information. The methodology employed is innovative and rigorous, resulting in a dataset with exceptional performance and reliability. The implications and applications of the 3D-GloBFP dataset are vast, spanning from climate modeling to sustainable development policies. Overall, this study deserves high praise for its contributions to the scientific community and beyond. However, to further strengthen the research, I would suggest addressing the following minor issues:

The division of the 33 regions mentioned in the paper is not particularly clear and requires a brief elaboration or a reference to the specific figure where they are illustrated.

The first sentence of several paragraphs in the result section introduces the methodology, but it is recommended to revise them to summarize the findings of the current paragraph instead. You may place the corresponding methodology in the method section, or write it after the figure caption.

While most of the results presented in this paper appropriately utilize the present simple tense, there are some instances where the past tense has been used inappropriately, see for example, line 149.

Expanding the discussion of challenges and future work would provide valuable insights into the dataset's limitations and potential for growth. Identifying gaps in current knowledge, discussing opportunities for integrating additional data sources, and outlining plans for updating and maintaining the dataset over time would demonstrate the authors' commitment to ongoing improvement and research.

By addressing these points, the authors can enhance the readability and presentation of their work, thereby ensuring that the 3D-GloBFP dataset becomes an invaluable asset to the scientific community.