This research presents a new set of global land surface parameters with a resolution of 1 km for multiple years from 2003 to 2020. This manuscript is well-structured. The figures are well produced. The English is good. The results are clearly presented. However, major issues should be addressed before this manuscript may be reconsidered for publication in the esteemed ESSD.

For a majority of data description papers in ESSD, a solid verification of the new data based on ground truth data and the comparison between the new data set and the existing mainstreaming data set are necessary and always included. Without such information, readers cannot fully understand whether the new data set is reliable and how much this data set has been improved compared with existing data sets. Consequently, the significance of this research cannot be highlighted. Therefore, I strongly recommend the authors to present a quantitative comparison between the new data set and mainstream data sets (e.g. CLM5 and K2012 datasets) based on already existing reference data or manually collected reference data.

Another important issues is about the citation in the text.

The citation should be thoroughly revised. For instance, the list of more than 10 references in a line can provide readers no accurate information and a clear relation between the reference and the mentioned information.

e.g. L49-L50 ... and biogeochemical cycles, as well as land and atmosphere coupling (Giorgi and Avissar, 1997; Chaney et al., 2018; Zhou et al., 2019; Liu et al., 2017; Bou-Zeid et al., 2020; Chen et al., 2020; Nitta et al., 2020; Vrese et al., 2016)…

These references should be clearly cited and explained. Personally, I do not suggest a list of more than 3 references in a line. The citations in the text are poor and all citations throughout the manuscript should be double-checked and revised to the right form.

This manuscript highlights the importance of developing new global land surface parameters with a resolution of 1 km for Earth system models (ESMs) running at the kilometer scale. The study demonstrates that these parameters significantly impact the spatial heterogeneity and information loss in ESM simulations, particularly in relation to soil moisture, latent heat, emitted longwave radiation, and absorbed shortwave radiation. The use of eXplainable Machine Learning methods helps identify the influential factors driving this variability and information loss. The new land surface parameters have implications for advancing our understanding of water, carbon, and energy cycles under global change. The paper is well written and has significant value for high resolution LSM modeling. However, I have several concerns as shown in the following comments for the authors to be considered.

Major comments:

1. The aggregation order problem should be addressed when upscaling the secondary derived parameters including DEM-derived variables, PTF-dervived (Pedotransfer functions) soil parameters such as saturated water content. Previous studies have proved that this order has significant effect on the derived parameters and thus the modeling results. For example, the aggregation after method has been recommended by Shangguan et al. (2014) and (Dai et al., 2019). That is, you should first calculated the derived parameters at the high resolution and then aggregate them into low resolution. This is majorly due to the nonlinear relationship between the original parameters and derived ones. However, the authors chose the aggregate first way according to the description in line 219~222, which is not a good way. At least, you should compare these two methods yourself, evaluate them and choose the better way.

Shangguan, W., Dai, Y., Duan, Q., Liu, B. and Yuan, H., 2014. A Global Soil Dataset for Earth System Modeling. Journal of Advances in Modeling Earth Systems, 6: 249-263, https://doi.org/10.1002/2013MS000293.

Dai, Y.#, Shangguan, W.#, Wei, N., Xin, Q., Yuan, H., Zhang, S., Liu, S., Lu, X., Wang, D., and Yan, F., 2019. A review of the global soil property maps for Earth system models, SOIL, 5, 137-158, https://doi.org/10.5194/soil-5-137-2019.

2. This manuscript only shows how the coarse resolution simulation may lead to spatial information loss but not the improvement in modeling accuracy leading by incorporating the high-resolution parameters. You need to evaluate the results to prove that how much gain in the high-resolution simulation compared to low resolution using some benchmark data such as high-resolution SM observations from remote sensing.

Minor comments:

Table 1: the LAI data is updated for 2000 -2021, see the link: http://globalchange.bnu.edu.cn/research/laiv061

Line 255: γ? You should explain it here.

Figure 4: why choose these four locations as they are located in the arid and semi-sarid regions? It is recommended to choose locations with a representation of various climate zones or PFT.

Line 340~341: the effect of soil texture is very likely amplified by the PTF-derived soil parameters especially soil hydraulic conductivity. Try to investigate this issue. In addition, try to answer questions like is the value of beta related to the clay content itself or other factors like soil heterology?

Figure 6: it shows that L1 has a lower standard deviation of clay, but a more negative beta than L2. This indicate that lower soil heterology does not lead to lower spatial-scale dependence. So, how to explain this? Also, the information loss is higher when the standard deviation is lower. How can it be? Same thing happened in Figure 7.

Figure 8 and 9: what is the link between these two figures? You may discuss this.