This manuscript presents AGPC, a new annual 500-m gridded population dataset for mainland China covering the period 1990–2020. A key strength of the study is that it explicitly incorporates three-dimensional building volume dynamics into the population mapping framework, thereby addressing the density-saturation limitations of conventional two-dimensional population products in vertically developed urban environments. The authors provide comprehensive validation using census and statistical data at multiple administrative levels, and the results demonstrate that AGPC achieves improved spatial representation compared with several widely used population datasets. Overall, the manuscript is well organized, and the proposed dataset will be valuable for studies of population dynamics, urbanization, disaster risk assessment, and sustainable development. I have several comments and concerns that I believe will help further improve the clarity, reproducibility, and interpretation of the manuscript.

1. Section 2.1 indicates that approximately 6,800 townships were used for validation in 2010, whereas only about 3,830 were available for 2020. Please explain the reason for this substantial reduction (e.g., administrative boundary changes or data availability) and discuss whether the smaller validation sample affects the robustness and comparability of the reported validation results.
2. In Section 2.2.1 and the reference list, the GUS-3D dataset is cited as Liu et al. (2025) and Wu et al. (2026), with the latter marked as “Undergoing Review.” Please clarify its current availability, public access protocol, or expected release timeline so that the reproducibility of AGPC can be properly assessed.
3. Please clarify the temporal consistency of the POI dataset. Section 2.2.2 states that the POI data were collected from Gaode Map but does not specify their acquisition year. These POIs are subsequently treated as temporally invariant throughout the study. If they represent recent urban conditions (e.g., 2022–2023), their application to earlier years such as 1990 or 1995 may introduce temporal inconsistency, as also acknowledged in Section 4.1.3. Please specify the acquisition year of the POI dataset and further discuss the potential influence of this temporal mismatch on the early-period population estimates.
4. Section 3.1 states that the GAUD impervious surface dataset was combined with GUS-3D to delineate inhabited areas. However, the cited GAUD product only covers 1985–2015. Please specify how inhabited areas were derived for 2020 and clarify whether the 2015 GAUD layer was directly used as a proxy or whether an alternative dataset or updating strategy was adopted for the post-2015 period.
5. Section 4.4 shows that the SHAP importance of CommercialPOI_den increases from approximately 28% in 2010 to more than 35% in 2020. Since the POI layer is assumed to be temporally invariant, please clarify how this increase should be interpreted. Does the higher SHAP importance in 2020 reflects a stronger dependence of the model on commercial POIs? Please provide a clearer mechanistic explanation to avoid misinterpretation.
6. Figure 3 includes a “District-level Census Data” component under the Population Constraints module, yet district-level census data are rarely mentioned elsewhere in the manuscript. Please clarify whether district-level constraints were incorporated into the baseline population allocation.
7. The word “Grided” appears repeatedly in the title, abstract, and throughout the manuscript (e.g., “500 m Grided Population”). The correct spelling is “Gridded”. Please correct this throughout the manuscript.
8. In the reference list, “China Statics Press” should be corrected to “China Statistics Press”.

Please see the attached file to get my specifc comments.