The paper is very well written. 

Hyper-resolution global water resource modeling is important for provide locally relevant information (e.g., water stress) for the public and policy makers. However, the lack of precise data (e.g., forcing, flow directions with the consideration of facilities, local scale water consumption, local scale dam operation rules and etc.). Hanasaki et al., provide a successful example of high resolution water resource modeling by using the global water resource model. The paper is generally well written, but I still have some major comments before its publication in HESS journal.

Major Comments:

1. The added value of high resolution meteorological forcing, high resolution hydrological and geographic data (e.g., flow direction, agricultural area, domestic water withdraw) and model improvements (e.g., improved crop classification, new aqueducts scheme) are not clear currently. Although the author discussed this issue in the results and discussion, for example “We speculate that the difference in performance between the two simulations is due to the quality of precipitation data”in L367. I suggest to separate the contributions from these three different factors to make it more clear to the reader that which is the most important for the improvement in high resolution modeling. Sensitive experiments such as “only use high resolution meteorological forcing”, “use both high resolution forcing and hydrological data” can be conducted to address the above issue.
2. The inter-grid-cell connection for groundwater. The author give a comprehensive discussion on the challenges of high-resolution modeling and have noticed the importance of inter-grid-cell connection (e.g., the aqueducts and water supply). However, the subsurface or groundwater lateral flow which is important at high resolution is not discussed (Ji et al., 2017). I am wondering whether the H08 model has considered the lateral transport of groundwater from the adjacent grids to the gird which experiences extensive groundwater plumping? If not, whether this process, that is important in hydrology science, lead to uncertainties to the water resource assessment?

Ji, P., Yuan, X., & Liang, X.-Z. (2017). Do Lateral Flows Matter for the Hyperresolution Land Surface Modeling? Journal of Geophysical Research: Atmospheres, 122(22), 12,077-012,092. doi:10.1002/2017jd027366

3. The author review the Wada’s work which found “ water stress is unrealistically concentrated in grid cells containing the downtown areas of the largest cities (e.g., Paris, New York) and therefore recommended assessment across larger spatial domains, such as sub-basins and counties.”While current work wants to assess the water stress at 2 km resolution or even higher resolution in the future. How do you consider or address this unrealistic phenomenon shown by Wada et al., (2016)? Moreover, the author pointed out that “ Thus, water scarcity may be underestimated in major cities in this simulation...” this seems to be contrary to Wada’s work. Detailed discussions are needed.

4. On the difference of using global and local water resource model. The author said that data localization and model localization are important for global high resolution hydrological modeling. As I am not expert at water resource modeling, I am confused that what the difference between high resolution global modeling and high resolution regional modeling? If the difference is due to the input data (e.g., forcing, reservoir operation, water use data), then what the difference between the following pathways: 1. use global water resource model, 2. use regional models (e.g., SWAT) and apply them at all global catchment.?

The authors present a study comparing the impact of global datasets vs. localized datasets when parameterizing hyper-resolution models for water resources applications. The paper is well written, and I appreciated the authors' efforts to obtain the best possible detailed localized data for their study site. Results show that, as expected, localized models performed better. While this is an interesting exercise, the paper would benefit from a better contextualization and insights on how we can leverage their work/their findings towards hyper-resolution modeling at the global scales, as the title suggests. Here are some major comments that should be addressed before publication:

1. L 131. "it was designed for applications to any spatial domain and resolution." One of my biggest concerns when applying large-scale designed hydrologic models to hyper-resolutions is the lack of or inappropriate processes representation at the fine scales. These large-scale models were often designed under the large grid cell assumption, in which local-scale and non-linear integrations between water and land and climate were negligible. However, when moving to fine scales, processes such as lateral water flow, the interaction between the river network and the land, and surface water pumping uphill, are no longer negligible. Rather than just increasing computational grid and parameter regionalization, hyper-resolution modeling also comes with the need to further understand and represent hydrologic processes at these scales. The authors do attempt to address the need for improving human activities representation at these scales with the implicit aqueduct estimation. It would be good to have a subsection where the authors describe the H08 efforts towards also improving hydrologic processes representation at the fine scales.
2. This modeling exercise demonstrated how models with localized inputs perform better than with global inputs. It would be great if the authors could provide a sensitivity analysis of the different input datasets to identify which are the most critical for improved performance. I propose a validation analysis using a leave one input out approach. In this way, besides just reporting what we know is already expected (localized models perform better), this paper has the potential to actually inform the scientific community of which of the inputs for hyper-resolution modeling we should be focusing on improving. Of course, all of them are important, but ranking them would greatly value future work in this field. Is that crop data? Precipitation? Water use and withdraws, etc.
3. L 630. "it opens the door to applications of the model to hyper-resolution global hydrology." Not really, the results actually show the opposite: that without localized data global hyper-resolution modeling of water resources is inaccurate, and that perhaps we should be focusing on localized models. If the authors can demonstrate what are the large sources of uncertainties (e.g., which are the localized input variables that drive the uncertainties), then you could say that improving a given X input at global scales can enable more accurate hyper-resolution global hydrologic modeling.
4. Model calibration is an important aspect of the model validation performance, as such, the details should be moved from the supplemental material to the main text.
5. Section 3.1.4, in the implicit aqueduct scheme, does the water only moves downhill? How about water pumping schemes? While they are often negligible at the coarse scale they are very much relevant at the local scale.

Moderate/Minor:

1. Section 4.3, as mentioned earlier, authors should also address hydrologic uncertainties in this section.
2. Figure 3. This figure could be improved for clarity. Inter-basin transfer you mean between grid cells? Does the grid cell comprise a basin? Are the rives the blue arrows or the blue grids?
3. Table 5 would benefit from a more descriptive header