General comments

Hofmeister et al. applied WRF 2 km data to a physically based hydrological model with a resolution of 25 m resolution in a small (62 km²) highly glaciated catchment in Austria. They provided two different types of air temperature to the hydrological model, one is the temperature 2 m above the surface, and one the so-called “corrected” temperature, which was derived from WRF temperature on the first and 20^th level (~3 km above ground). They also corrected a drizzle effect in WRF by setting all values to 0 below a monthly determined threshold based on a station nearby the study area.

Comparison with four stations within the catchment showed an improvement for the corrected temperature, especially for the highest station which was presumably wrongly classified as glaciated in WRF. With the corrected temperature also the hydrological model performance improved, e.g. comparison with MODIS determined snow covered fraction improved, the sign of glacier mass balance turned from obviously wrong positive to realistic negative and comparison with observations of streamflow improved.

However, the authors promised in the title, in the abstract and the introduction with the stated research questions much more than this synthesis, which was later not presented in the manuscript. I will provide examples in the next section. Oppositely, the manuscript contains significant parts which do not relate to the stated research questions and rather show interesting applications of such a model chain or a discussion/review on other studies. It is a long manuscript and there is a clear need to focus.

While reading the manuscript it was often not clear to me what motivation was or how the methods were applied. Reading the conclusions at the very end was a (belated) eye-opener for me, and yet questions remain (see next section for details).

Another important issue is the limited availability of data to ensure that the hydrological model used to evaluate the altered temperature inputs does not simply improve the results due to error compensation (see details in also in next section).

I acknowledge that preparing WRF model results and applying a high-resolution, physically based hydrological model as WaSIM is an extensive task (WRF results were probably taken from Altmann et al. (2024) and it is not clear if they were independently generated, but this is not the relevant point here). I suggest that the authors completely revise their manuscript and clearly focus on what they can provide with the presented analysis. However, I have the feeling that a distilled manuscript is not sufficient for an own publication and that the authors should rather proceed with their obvious plans of analyzing model output from long-term runs since 1850, where this here presented part of adjusting air temperatures is a small paragraph.

Detailed comments

Stated but not done (as it first appears)

Small-scale surface representations in WRF

The title and the first research question (line 96) suggest that small-scale surface representations in WRF and their impact on subsequent hydrological models have been studied. The authors want to present the effects of an alternative derivation of near surface air temperature, which should consider elevation and land cover classes (line 100). However, this correction is not based on any small-scale surface properties; the opposite is true: It is removing the influence of the probably too coarse WRF land cover information by simply averaging temperatures over the whole domain at elevation levels where they are relatively insensitive to land surface properties (which is at the end also stated in the conclusion, lines 684-687).

Bridging the scale from WRF to WaSIM

In the abstract the authors stated that the main challenge is to transfer forcing data to the much finer resolution of the hydrological model. The authors wanted to present a workflow for bridging this scale. This relates to the second research question in the introduction (lines 97-98). However, I can only see the application of rather simple WaSIM integrated algorithms (lines 300 to 309) which cannot be meant by a newly presented workflow.

Consistency of bias-corrected RCM data over longer periods

This relates to the last and third research question (lines 98-100). However, the authors have not applied bias-correcting method for temperature using observations. The temperature adjustment presented is not using observations (but still removed biases). The consistency was tested in comparison to a nearby valley. However, this was not done in relation to non-corrected data (and probably only for one reference grid point), so one cannot state an improvement.

For precipitation they removed a drizzling bias using long-term observations which is a classic bias-correction case. However, no consistency results were presented here.

Modelling period

In the abstract it was stated that the modelling period is from 1850 to 2015, however, only data from the 1973 to 2015 are analysed in this manuscript.

Topics outside of the focus

The manuscript can substantially be reduced to topics which are part of the research questions. Example of topics, which are not part of an evaluation of presented temperature and precipitation correction may be the determination of significant change points (unless it is not evaluating the corrected input).

Another example is the long discussion as it does not relate own results with other studies. It can be shortened and put in the introduction or placed in a review paper. The implications of own results are hardly discussed in the discussion.

Clarity

In equation 5 and 6 it is stated how the corrected and uncorrected temperature at a certain station location is calculated. However, it is unclear to me how this is done for WaSIM grid points: In lines 302-304 I can read some information, but how the variable lapse rate is determined is unclear, and whether this is done for both variants. This is particularly relevant as Eq. 6 contains a fixed lapse rate to account for elevation differences.

Equation 5 and 6 differ in two ways: First the mentioned lapse rate is fixed and second the 2 m temperature is used in Eq. 6 for the uncorrected version, while Eq. 5 used an hourly changing lapse rate and the first level temperature. It is unclear to me, which part is the relevant one for improvements. As this is the core part of the paper, I would suggest some additional analysis.

One main argument for using the corrected temperature approach was that the land cover classification too coarse in WRF (conclusions lines 684 – 686). The authors missed to show a map of this classification.

Improvements or error compensation?

In the conclusion the authors state that due to the lack of meteorological stations above 2700 m they evaluate the impacts of the corrected temperatures with a hydrological model. However, it is unclear to me if the authors have included not more issues with this attempt than less. The hydrological model certainly includes uncertainties in the modelling cascade, which are also discussed in the manuscript. One issue is that the study area is not really suited for this evaluation. While I see the point that there is a switch from an unrealistic to a realistic sign in the glacier mass balance trend, there are no measurements of glacier mass balances available. There is also no spatial snow distribution data used in this study. So, the reader does not know whether the shown improvements in fractional snow-covered area, glacier mass balance and runoff comes not from error compensations elsewhere in the hydrological model.

The authors argue with Figure 7 that the improvements stem from air temperature corrections leading to an improved distribution of snow, to an improved snow and glacier melt and finally to an improved runoff.  One indication for an error compensation is the largely biased snow water equivalent (SWE) as well as its insensitiveness to the temperature correction (Fig 7a) at one single station. I think the authors need to provide additional insights in the correctness of their argumentation. For this it may be relevant to evaluate the model chain in regard to a realistic SWE distribution. To my opinion a 500 m resolution MODIS product is not sufficient for this complex terrain, which was also stated by the first author in Hofmeister et al. (2022). Such a study can be done using Sentinel2 data as done by Hofmeister et al. (2022) but needs to be presented for WRF input and for this specific catchment but may be limited to a few winter seasons.

Review of The impact of small-scale surface representation in WRF on

hydrological modeling in a glaciated catchment by Hofmeister et al., submitted to HESS.

The manuscript (now being considered as a technical note - communication with editor on 29-09-2015) principally addresses the implications of considering the WRF land-surface classification upon the distributed air temperatures for input to a hydrological model using a glacierised catchment in the Austrian Alps. The impact of re-deriving air temperature lapse rates from pressure levels of a 2km WRF model is assessed at a very high model resolution of 25m using an intermediate complexity hydrological model with consideration of glaciers and snow. The authors highlight that the representation of glacier mass balance, snow cover dynamics and catchment hydrology are all improved with the representation of air temperature which acts more independently of the prescribed land cover class from the 2km WRF grids. 

The quality of the writing and figures is generally very good and the arguments are reasonably supported for the most part. The impact of the temperature downscaling methodology for long-term glacier mass and related impacts on the catchment hydrology are interesting and potentially useful for the community. The actual long-term modelling results are very nice (despite not being calibrated or reasonably evaluated) and could be very interesting for the community.

My initial feeling upon reading was, however, that the manuscript does not provide a substantial advance with respect to catchment modelling and the work rather describes a methodological approach to improve the WRF forcing data (that is not always so clear and well justified), albeit in a very long article. With this manuscript now being considered as a technical note, the authors must substantially shorten the work and focus on a clear, yet concise argument for the practical applications of air temperature lapse rates from WRF for catchment modelling, ideally testing different combinations of lapse rates/methods and providing a much clearer description of the method and the context for its relevance. The change of article type already constitutes a major revision for the authors, which is consistent with my recommendation, before being acceptable for publication in the journal. My major and minor technical comments are written below.

Major Comments

1. The manuscript is long and overly descriptive in places in order to present a methodological step toward improving forcing for a hydrological model. However, a complete analysis of why and where the hydrological model is improved using the lapsed free-air WRF temperatures is still lacking in places. There is no testing of alternate lapse rate approaches or using different pressure level combinations from the WRF, nor whether ‘ignoring’ the WRF near-surface temperature attribution due to landcover type is in any way dependent upon certain conditions or times of year. The transition to a technical note for the journal should see shortening of other sections describing the WRF and WaSIM modelling, where appropriate, especially where details can be provided in summary tables (main text or SI). 

2. Regarding the methodological details for the derivation of the lapse rates from pressure levels of WRF, the details are actually lacking still, and it's not clear to me exactly how or why the lapse rates have been derived the way they have. For example, from my reading, the authors average temperatures (and geopotential heights) for each level for each time step (equation 1+2), but then use only a linear fit for the 20th and 1st levels (equation 3), and then make an average of all hours (equation 4). The authors make the claim that the spatial averaging on each level makes the lapse rates more robust, though with no real evidence of this. This may be negligible due to the size of the catchment, but the reader has no way to understand if the land surface variations (in this case, most of the glacier cells are to the south of the catchment) could still have some impact on the lapse rates and comparison to the observations and impact on the hydrological model etc. The information in figure 3 also highlights that lapse rates are calculated for each elevation band, but it is unclear how that was derived (stepwise lapse rates?), whether taking different eta levels or removing the lowest level (which retains the impact of surface exchanges of WRF) impacts the results of the study here, and potentially for other applications. The authors need to make a clearer case to justify what lapse rates they apply, how robust their approach actually is, and how their approach might succeed or struggle in other, potentially larger, domains. The true value for the community is therefore minimal at the current time.

3. The main focus of the study is exploring the meteorological and model differences when using pressure-level lapse rate vs. near-surface WRF air temperatures. The title and main aims of the work are somewhat misleading therefore with respect to “impact of small-scale surface processes”. I think that a more clear and specific title, as well as rephrased terminology throughout the manuscript is warranted. The manuscript really deals with the impact of correcting air temperatures due to coarse land surface representation of WRF.

4. The relevance of the corrected vs uncorrected air temperatures needs to be put into perspective when also adjusting and downscaling other meteorological variables (e.g. humidity) in a similar way. Describing their impact on long term snow and glacier evolution and objective ways to choose pressure levels/lapse rates are needed to be seen as a technical advance for the community. This would not require any re-running over heavy WRF simulations (which would be excessive), but rather just the post-processing of corrections for inputs to WaSim.

Technical Comments

Introduction

The authors should reorder and restructure parts of the introduction for a more logical flow related to the aims of the study (the introduction starts talking about geomorphology and fluvial transport, and then moves onto droughts and flooding, which is too specific given the focus of the study). Given the change of article type, the authors should now go directly into the importance of accurate meteorology in hydrological models. 

L42: This should be reworded. Observations since the LIA have only improved. Observations back in time to the LIA, however, are limited in space and time. 

L44: The reference should be “van Tiel”. Likely a citation manager issue.

L55: Given the specific application of WRF and a study about its limitations, the introduction and explicit benefits of HiCAR don’t make sense here, especially as HiCAR is no longer used nor mentioned.

L93-94: Correcting the diel cycle of what exactly? The reasoning of this paragraph needs to be made clearer.

L96: Small scale processes need to be clearly defined here for the reader. It seems that this only refers to the land cover representation from WRF, which is anyway largely removed and averaged in the manuscript’s workflow to model at 25m. In essence, the small scale processes are not really considered, and the model is improved as a result. In line with my major comment above, this needs to be made clear from the title and re-wording of the manuscript. 

L96: RQ2 is not grammatically correct. It is written as a question, but the syntax is not correct. 

Study Site

I guess that the data from Weissseespitze lies too far beyond the modelling period for this work? https://doi.pangaea.de/10.1594/PANGAEA.939830 This should be at least mentioned, explaining that site specific meteo and mass balance data are not available for the catchment of interest. 

L130: Why is the Vergoetschen station not listed?

L136-140: Please describe (in brief) the SWE model. What are the key, perhaps site-specific, parameters?

L186-187: Provide an indication of the seasonality for the strength of the surface cooling effect over ice and snow. Is this related to the albedo and radiative cooling during winter primarily? Or also the summer, density-driven cooling over ice due to glacier wind development? Are there notable differences here due to ice vs. snow?  What is responsible for the anomalous 2m temperatures over shrubland (Fig. 2a)? Does it meaningfully influence the lapse rate derivation if not averaging out the spatial effects (equation 1)? 

L188: Define what is meant by “lower gradients” Which gradient exactly? Does lower mean a shallower lapse rate (a smaller change of temperature with elevation)?

L204: Here the authors are defining T, not TlevelX, I believe. Please check, in case that this is a typo.

L209: As per my major comment, is this actually more robust? And more robust in what way? Evaluated how? The authors provide no evidence of this. This needs to be more rigorously tested and demonstrated for the reader. 

L214: The exact justification for the use of a linear gradient between levels 1 and 20 is not so solid from my reading. Levels 20-25 could also represent the mountain boundary layer. Can the authors provide some generalisable means of excluding the free-air based upon an objective measure? For example, how strong is the relationship of air temperature and elevation (e.g. the R2 of the levels in Fig. 2). More importantly, does it make any difference? How would another study utilise this information (which forms the core part of the manuscript) to improve their own hydrological model simulations? What impact does ignoring the lowest eta have?

Fig3: The exact derivation of the pressure level lapse rate for this figure is unclear. Is this the regression of temperature and elevation for each level, as shown by Fig 2? This approach is however different to how the authors derive lapse rates applied to WaSim, however (by averaging temperatures for each level). What about taking the lapse rate in a similar way to what we see in Fig 3, from the nearest WRF pixel? This would implicitly account for the often shallower lapse rates observed over glaciers. Regarding this figure, the distinction to the methodological approach of the text should be given and explained. Ideally, the authors can also add the approximate elevation from the eta levels for interpretability, marking the elevation range of the catchment on top.

L237: So the complex data results in a single lapse rate value to correct the (2m?) WRF cell to the elevation of the meteo station? This seems a huge simplification of a rich data input from WRF, but likely also depends on the absolute differences of elevation between the two. These elevation differences (HWRF and HStation) should be given somewhere. 

Section 2.4.2: The precipitation frequencies can be highly influential to high resolution, physically-oriented models, so I am happy to see that this has been considered. For the re-structuring of the manuscript, I think these are details which can be considered away from the main text somehow, however. Are there other bias-corrections necessary to improve the representation of glacier mass balance in this catchment?  Related to this, What comparison to geodetic glacier changes has been made, and how sensitive would the model results (i.e. Fig. 7d) be to variations of the lapse rate derivation as I have suggested should be considered? 

Results

Fig. 4: Having some statistics for the improvement of fit would be beneficial here (mean bias, RMSE etc).

L455-456: Citation for this statement is necessary.

Fig. 8 a,c,e,g: The standard deviation of what? The sub-daily simulated values? This should be clarified. 

Fig 9: These are nice results and quite valuable for the community, but they seem to address a broader question about the long-term evolution of the ice and snow in this catchment, but not really related to the value of the authors approach to improve the temperature downscaling. This could be left for successive work on this topic, or revised to demonstrate the long-term implication of different approaches for changing/improving/simplifying WRF’s surface representation.  Also for this figure, the authors should specify whether these are hydrological years or not. I suspect they are, as 2002 has the lowest snow cover and not 2003 (the European heatwave). 

Discussion

The authors have pre-empted the critique of their modelling approach that has overlooked humidity and other variables which are also dependent on WRF’s coarse (and perhaps incorrect) surface characteristics. Nevertheless for a physical modelling approach, the accurate estimation of most or all meteorological variables can be quite important. The relevance of the improvements to the air temperature downscaling cannot be put into perspective, however. I think that upon revising and restructuring the manuscript as a technical note, some testing of the different downscaling of the humidity and precipitation should be made to highlight whether it is as influential to the model calculations as the corrections made to air temperature. 

L606: Why did the authors decide upon the usage of 20CRv3 when ERA5 is a widely adopted reanalysis, especially in Europe. This should be mentioned here. 

L614: Can the authors justify (ideally early in the manuscript) why a very high resolution of 25m was chosen or necessary.