This work compared the simulation performance of Noah-MP land surface model over Ireland using two global soil property datasets, including a high-resolution SOILGRID data and a coarse-resolution STASGO data. The results showed that, the coarse STATSGO performs as good as the fine-scale SOILGRIDS soil database, although they both have dry biases. Overall, I think comparing the added value of high-resolution soil dataset to land surface modeling is an important and meaningful topic for both data and model developers. However, there are lots of uncertainties caused by other processes, such as the observation and the model physical parameterization, instead of the soil database. Following major issues should be carefully considered before the further consideration of the publication.

1.The innovation. The current innovation is somewhat weak for the HESS journal. The work is very similar to Zhang et al. (2023), although the authors said that they used the SOILGRID dataset. The work compared the difference of soil moisture/temperature simulation and drought processes (2 drought events) over Ireland, and made some conclusions. This makes the work like a technical comparison without in depth analysis of the uncertainties and related reasons. For example, why the high-resolution soil database performs similarly with the coarse-resolution? Is it because the uncertainties from soil database itself or the model structure and physical parameteritzations (for example, the uncertainties of PFT function to accurately derive the soil hydrological properties based on soil texture data)? 

2.The station observation. There are 6 observation stations used in this work, and 4 of them are from a new network (Terrain-AI) using Time Domain Reflectometry (TDR) sensors. I found noteworthy difference between the Terrain-AI based observations and the 2 long-time eddy covariance grass flux sites. For example, most of the Terrain-AI based observations show high values during wet seasons (larger than 0.5 m3/m3)，while observations from 2 eddy covariance grass flux sites are generally lower than 0.5. I wonder whether these high-values in Terrain-AI stations are true or not? This is very important to the conclusion, as the dry bias is mainly due to these stations. If the observation is true, then why there are such large differences considering they are all loam or loam-sand stations?

3.The satellite observation. Why you chose the ASCAT database as a reference? In my practice, the ESA-CCI or SMAP datasets are usually perform better than ASCAT. Is it because the ASCAT has best performance (use the station observation as a reference) or just ASCAT dataset can reproduce a dry bias pattern? In addition, how do you consider the influence of uncertainties of ASCAT dataset on the evaluation results?

4.The soil moisture or soil moisture anomaly. Although SOILGRID seems to show larger negative bias in modeling soil moisture, it improves the correlation coefficient. An important issue is whether the soil moisture absolute value is more important than the soil moisture anomaly (dynamics)? Actually, the observed soil moisture and model simulated soil moisture are physically different. Model simulation is a mean state of a grid box with specific thickness, while observations only represent a point at a fixed depth. 

5.Why the SOILGRID improves the simulation of soil moisture dynamics but increases the dry biases? Some in-depth analysis should be provided. In addition, the differences in soil moisture drought may not simply related to the simulation of soil moisture absolute values because the soil moisture percentiles are used here. I wonder whether the soil hydraulic conductivity or diffusivity is responsible for the difference here. For example, a higher conductivity can cause a faster response of soil moisture to the water deficit.

Overall, I suggest the authors to analyze the issue in-depth and comprehensively consider the uncertainties of model, observation, and soil datasets, so as to provide more novel and impressive results.

see attached PDF.