In this work, the authors aim to develop a long-term global dataset encompassing multiple types of livestock and poultry to describe long-term changes in animal populations and spatial distributions. Undoubtedly, this is a significant effort. Such a dataset would advance our understanding of global livestock system transformations, support evaluations of environmental impacts of livestock production, and contribute meaningfully to sustainable agricultural practices and integrated ecosystem management.

The introduction of this manuscript is logically rigorous, and the description of the research findings is clear. However, concerns arise regarding the methodology and discussion sections, as outlined below:

1. The authors rely on FAOSTAT’s national-level livestock statistics as the primary data source for mapping. While these data span a long temporal range (1961–2021), their spatial resolution is generally coarse. Deriving gridded datasets primarily based on these national statistics may introduce substantial spatial uncertainty, as livestock distributions exhibit strong intra-national heterogeneity (https://doi.org/10.1016/j.oneear.2023.08.012; https://doi.org/10.1016/j.rse.2019.111301). And this issue could be particularly pronounced in large, transhumant livestock nations such as the United States, China, Brazil, and India.

2. As noted in the discussion (Lines 281–290, Figure 8), the authors indicate that adopting finer-scale livestock statistics (e.g., municipal or county-level) is one of the most effective methods to reduce uncertainties. In fact, numerous studies have already leveraged such high-resolution data to develop regional spatial datasets, such as https://www.nature.com/articles/s41597-024-03072-y；https://doi.org/10.5194/essd-13-515-2021. A recent study even compiled over 50,000 fine-scale records for global livestock mapping (https://doi.org/10.21203/rs.3.rs-6201916/v1). Compared to these efforts, what advantages does this study offer in uncertainty control?

3. The authors mention using GLW4 to downscale FAOSTAT’s national statistics to municipal (city) scales (Lines 117–119), yet the specific methodology remains unclear. Is the process based on calculating municipal proportions from GLW4 data and then scaling national totals by these proportions? If so, this approach may inherit significant uncertainties, as municipal proportions can vary substantially over time.

4 . Based on the difference in feeding systems, authors categorize animals into “grazing livestock” (e.g., buffalo, cattle, goats, horses, sheep) and “captive livestock” (e.g., chickens, ducks, pigs), and assume grazing species inhabit grasslands while captive species are confined to impervious surfaces (Lines 83–85). This assertion appears questionable, as intensively raised animals often occupy peri-urban or rural agricultural lands (https://doi.org/10.1016/j.oneear.2023.08.012).

5. The discussion is not very adequate. For instance, the claim that vegetation omission minimally impacts predictions (Lines 280–281) is counterintuitive. What underlying reasons justify this assertion? Have other studies observed similar patterns? Is it premised on the assumption that grasslands or impervious surfaces serve as “theoretical suitable masks” for livestock distribution (Lines 83–85)? Additionally, Figure 4 shows marked disparities in prediction accuracy across species (notably lower for cattle and higher for horses). What factors explain these variations?

The manuscript by Du et al. presents a well-structured and timely study that reconstructs annual livestock density at a global scale from 1961 to 2021. Addressing this issue is crucial, as accurate and consistent time-series data on livestock distributions are essential for quantifying environmental impacts, supporting food security assessments, and informing policy decisions related to climate change and land use. I recommend a minor revision. My major comments are as follows:

1. The methods section lacks clarity in certain areas, particularly regarding the stratified sampling approach. The manuscript does not clearly describe how stratified sampling was implemented (L140-L146). This information is critical, as it directly influences the composition of the training dataset and consequently affects the accuracy and reliability of the global predictions. I recommend that the authors provide a more detailed explanation of the sampling procedure, including the criteria for stratification and how the strata were defined and selected.
2. The causal relationships between the predictors and the response variable warrant further clarification. In this study, the authors used a range of environmental and anthropogenic factors to predict livestock density (Fig 1). For predictors with limited historical data, such as population, the authors applied year-2000 values to years before 2000 and found that population had little influence. This conclusion seems counterintuitive. Unlike wildlife, livestock is more likely to be influenced by human management. Therefore, one would expect population density to be an important predictor. However, in this study, soil and climate variables were found to be more influential (fig 7). This may reflect correlations rather than causal mechanisms. A comparison between the spatial patterns of cattle or sheep and population density (https://hub-worldpop.opendata.arcgis.com/content/WorldPop::global-1km-population-total-grid-2000-2020/about) suggests that a strong spatial association likely exists. I think that the lack of observed influence in the model may be due to two reasons: (1) errors or bias introduced during stratified sampling (as noted in comment 1); and (2) potential multicollinearity among predictors. If population is indeed an important factor, I think the authors to revisit its treatment carefully. In addition, I strongly recommend including partial dependence plots or similar visualizations to show how each predictor relates to the response variable.
3. The meaning of the dots in some figures (e.g., figs 4-7) should be clarified in the figure captions

Shuai Ren