Summary of the paper:

This article calibrates a sophisticated hydrological model with diverse datasets in a large basin in the Tibetan plateau and then uses the model to understand sensitivity of hydrological fluxes to possible changes in temperature and precipitation. The entire calibration exercise is very interesting, as it is done with varied dataset (streamflow, snow coverage, glacier mass balance, and stable water isotopes). By calibrating across these datasets, it is assumed that the model parameters closely mimic the underlying hydrologic processes. The article then does climate perturbation studies and predicts that temperature increase changes internal water flux partitioning within the catchment with limited changes in absolute amount of streamflow, whereas precipitation changes has a significant impact on streamflow amounts (and not on flux partitioning). The article also highlights certain interesting threshold processes occurring in this region, where a small increase in temperature may lead to decrease in streamflow whereas a larger increase in temperature increases streamflow due to enhanced glacial melt. Additionally, there are lots of other small interesting results within the article which might be very relevant to researchers working in this region.

The flow of the article is well drafted and it explains modeling and results reasonably well. I have a few suggestions to further improve the article and a few questions regarding the model application.

1. A key advantage of using a hydrologic model is that it allows estimating uncertainty within the modeling result. However, the article does a very poor job in describing uncertainties in model outputs. Except in Figure 5(a, d), uncertainty in model outputs have not been described in text or in figures. For e.g. in Table 4, change in CP days (<2 days in the most extreme scenario) have been inferred which might be deceiving unless the uncertainty bounds introduced by natural climate variability is taken into account. It is quite possible that the day of peak runoff varying by 2 days can be explained by natural variability. To ensure the robustness of these results, I suggest adding uncertainty to most plots and do statistical significance tests. I have pointed this out findings in my more detailed comments below. My general suggestion will be to introduce uncertainty bounds in every figure (wherever feasible) and include statistical significance of any future results (e.g. runoff decrease of 20%under a given scenario at what significance level?). This will make the discussions more robust.
2. One key limitation in the model calibration is that the calibration period uses the entire period of collected hydrometeorological data, and there is no proper validation period. In a way, the article replaces time with space, where calibration is done over time and validation is done scantly over space. Please provide a justification for this calibration scheme, and why hasn’t the model been calibrated and validated in a conventional way?
3. The hydrological model consistently underestimates peak streamflow across sites (only exception is 2008 period of Figure 3f). This suggests that the model has a consistent bias which hasn’t been adequately discussed in the article. Please provide more context to this, why this bias exists and what should be done to remove this bias.

L201-203: How different is glacier meltwater isotopes from snowmelt isotopes, and how have these two been differentiated within the tracer module of the model?

In the mathematical equations, a lot of variables are depicted by small words (like PET). I will suggest using a single alphabet for a variable, in-line with proper mathematical convention. Please address this in all the equations.

Figure 4. Looking at difference between Y-axis of Fig 4a and 4d, its clear that annual runoff is not significantly impacted by increasing temperature. Is the conclusion about reduction in annual runoff in different temperature scenarios statistically significant?

Figure 5. One interesting takeaway from the figure is that precipitation increase has a disproportionate impact on summer month runoff whereas temperature increase has higher impact on winter runoff. This is a very interesting result as it suggests winter baseflow is more influenced by temperature changes. This should be highlighted in the text.

L366-367: concentration ratio and concentration period have not been defined in the text. Please define them

L369-372: Was CP decrease of 2 days statistically significant?

L380-383: if subsurface runoff is 70%, rainfall runoff 20%, glacial melt runoff 10% and snowmelt runoff 5%, then where is the remaining 5% water flux? I suggest using uncertainty bounds (or standard deviation values) with these fluxes to solve these issues

L404-408: I don’t understand the discussion around proportion of three other components. Also these values look very very small to make meaningful conclusions. Please include error bands around these values

L453-457: This is a very interesting result, it highlights the most dynamic regions within the basin, which can keep shifting between energy vs water limited stages. Have past studies identified such regions? If yes, please mention them in discussions. Also, please highlight this part in the abstract and conclusions part.

L478-479: Please show the other figures with insignificant correlations in the supplementary material

L488-489: How can glacier coverage be higher in warmer regions? Shouldn’t the presence of glaciers be more in colder places?

Figure 10: This is a useful figure but I think change in runoff is very low in T+5 scenario (maybe bulk of points lie between -10% to 10%). What is the uncertainty range of simulation of runoff?

Figure 10: I suggest reproducing this sort of a figure for increasing precipitation scenario. In the text, it was mentioned that in certain areas streamflow increased by >10% if precipitation increased by 10% and in certain areas streamflow increased by <10% if precipitation increased by 10%. In a way, different regions are showing different streamflow elasticity. I suggest producing a figure of streamflow elasticity in increasing precipitation scenarios and highlight which areas are closest to 1, as they would likely show non-monotonic behavior i.e. shift from <1 to >1 in different precipitation. Does glacier area ratio play a pivotal role there as well?

L542-543: Can you point to the figure which highlights this?

L553-554: Decrease of 0.1%-3% is likely very small and not statistically significant. Can you clarify?

Figures

Figure 1b: The yellow color of tributary station YBJ is blending with the yellow color of the underlying DEM. Figure 2b should be redrawn and the contrast between legends and background should be increased. Its currently very hard to read

Figure 4: Please describe subplots (d) and (h) in the figure description. Add error bands in all the subplots

Figure 5a. X-label 350 matches with “0” of subplot (b) making 350 look like 3500. Please resolve this. Also add error bands to Figure 5e and 5f

Figure 6c,d: Add uncertainty bands

Figure 7e-h, m-p: Add uncertainty bands

Minor comments:

L16: replace exist with “existed”

L23: can use “multiple datasets” instead of “Datasets of multiple objectives”. It makes reading easier

L38-40: I cannot understand what is being said in this line. Please rephrase

L81: replace “processes” with “models”

L97: replace “likened” with “compared”

L121: remove “and”. It can be better framed as “Snow, glacier, isotope data and observation ..”

L140: Missing unit “km2” next to 2x10^5

L145-146: Why is the acronym of Yangjia “TJ”? It is not at all intuitive

Table 2: Instead of using column name as “Sample number”, use “Number of samples”. Sample number gives the impression that it’s the laboratory sample number of a collected water sample

Eq.6: Please move the equation to L255 as it is a continuation of that sentence

Table 4: Add error bands (or standard deviation values) in this table

This work aims to disentangle the long-standing question of hydrological sensitivities to climate change. This is done by feeding climate forcing into hydrological simulations. Specifically, the climate forcing is perturbed in order to assess the sensitivity of the hydrological response in a large mountainous basin. Of particular interest is the use of a detailed hydrological model capable of modelling not only streamflow but also snow cover and glacier mass balance, both key elements for medium-term water management and water resources optimisation. The authors point out that the novelty of the work reside in both the assessment of "patterns and drivers of local hydrological sensitivities" and to the case study. This is also due to the fact that there are contrasting results from similar analyses in the same region of interest. However, in my opinion, one of the most interesting features of this work is the adoption of a sort of multi-objective calibration based on streamflow, snow cover area, glacier mass balance and stream water isotopes, alongside with a spatial validation of the calibration. Therefore, I believe that this paper is worth publishing in HESS after answering the following questions.

The model calibration is one of the weakest parts of the papers. There is no reference to the algorithm used as well as all the details regarding its applications which, in my humble opinion, should not only be mandatory to ensure the replicability of the paper, but are also important to understand the calibration results. For example, if the PSO scheme is adopted, the sensitivity analysis of the hydrological model parameters is not possible. Conversely, a Monte Carlo procedure could help to define the confidence band of the hydrological results. This should be explained according to the desired research objectives.

I have some doubts about the objective functions. If I understand correctly, the model seems to have been calibrated using a single objective function composed of four equally weighted functions (NSE for streamflow at one site; NSE for isotope; RMSE for SCA; and RMSE for GMB). In this way, however, the authors mix different types of metrics. I try to be clear: the NSE varies between minus infinity and one, while the RMSE varies from 0 to plus infinity (theoretically, of course). As a consequence, different metrics have different impacts on the aggregated objective function. I therefore believe that the NSE (or RMSE) must be used for all the objects under consideration. Otherwise, one factor could be weighted more than the others. I would suggest to have a look at "Madsen, Henrik. (2003). Parameter estimation in distributed hydrological catchment modelling using multiobjective automatic calibration. Progress in Water Resources. 26. 205-216. 10.1016/S0309-1708(02)00092-1.,Section 2.2, Equation 3.

I also disagree with the definition of multi-object calibration. Basically, the authors use a single-objective calibration. In other words, they choose a specific solution in the multi-objective space. I believe that this strategy is fine if the aim of the paper is to focus solely on the uncertainty of climate forcing, although a more precise definition of the objective function and calibration is required, in my opinion.

Model calibration and evaluation section. I propose not only to carry out a spatial validation of the hydrological model, but also a temporal validation with a calibration period and then a test period. The hydrological response could then be carried out taking into account all 15 years from 2001 to 2015. At this stage it is indeed important to ensure the reliability of the hydrological model.

Line 303: I propose to support the sentence "but were at acceptable levels with "D.N. Moriasi, J.G. Arnold, M.W. Van Liew, R.L. Bingner, R.D. Harmel, T.L. Veith Model Evaluation Guidelines for Systematic Quantification of Accuracy in Watershed Simulations".

The limitation of hydrological performance in terms of either maximum flow or low flow is consistent with several studies showing how the use of one metric in calibration can lead to less than ideal results for other metrics, as each is sensitive to particular characteristics of the time series and has its own limitations and trade-offs. See for example "Schaefli, B. and Gupta, H. V.: Do Nash values have value?, Hydrol. Process., 21, 2075-2080, 2007; "Gupta, H. et al.: Decomposition of mean squared error and NSE performance criteria: Implications for improving hydrological modelling, J. Hydrol., 377, 80-91, 2009. "Mcmillan, et al. Five guidelines for the selection of hydrological signatures. Hydrol. Process., 31, 4757-4761, 2017. "; "Fenicia, F., et al.: Signature-domain calibration of hydrological models using approximate Bayesian computation: Empirical analysis of basic properties. Water Resour. Res., 54, 3958-3987." and "Majone, B. et al. Analysis of high streamflow extremes in climate change studies: How to calibrate hydrological models? Hydrol. Earth Syst. Sci. 2022, 26, 3863-3883". . I suggest that this consideration be taken into account when rewriting the section about the evaluation of model performance".

A few words should be devoted to the description of the concentration ratio and the concentration period, as is the case for NSE, LnNSE, RMSE ecc ecc.

For the sake of clarity, I suggest deleting Figures 5a and 5b. They do not add any additional information respect to Figures 5c and 5d.

Figure 7: I do not understand figures from 7e to 7h and figures from 7m to 7p. It is not clear that the sum of the components is 100%. I suggest using a different type of figure style.

I suggest using the correlation values (r) in the text to identify and comment on positive\negative correlations and strong weak correlations (see Ratner, Bruce The correlation coefficient: Its values range between +1/-1, or do they not?). The p-value only indicates that there is a relationship between two groups.

Limitations section: I suggest adding that future work should address the problem of sensitivity analysis of hydrological models, multi-objective calibration and goal-oriented calibration.

The work is a good, well performed, exercise of modeling. After the model calibration, the authors did some scenarios simulations to assess the effects of temperature increase and precipitation change. This should be the purpose when modeling climate changes effects.

Correctly the authors listed also the limitations of the work (chapter 4.3). At L572-577 they state that the simulation of climatic change was at least too abrupt. By sure when arriving at +5° the glaciers will not be the same than now. They already are reduced since the simulation period (2015). But furthermore, I would remark that the temperature increase due to the climate change is far from being linear. We currently measure mainly longer warm periods (sometimes with higher temperature maximums) and shorter cold periods, and this leads to an increased mean temperature. Simulating a linear increase is of course easier than introduce a climate (stochastic?) model, but the effects on hydrology are surely different. Remarkably on ice or snow melt, but also, for instance, on evapotranspiration estimates. A second deep source of uncertainties is due to lack of data over such a large basin. “The CMFD was made through fusion of ground-based observations with several gridded datasets from remote sensing and reanalysis” (He, J., Yang, K., Tang, W. et al. The first high-resolution meteorological forcing dataset for land process studies over China. Sci Data 7, 25 (2020). https://doi.org/10.1038/s41597-020-0369-y). The CMFD, like all large-scale datasets, is the product itself of a model that incorporates from sparse ground data (about 735 CMA stations over 9.5x10⁶km²) and remote sensing data combined by some algorithms and interpolated. As well known, eventually, the uncertainties of a cascade of models become rapidly great.

In my opinion these limitations severely reduce the usefulness of the work.

L55 – the melting processes of frozen water are determined by the energy budget.

L75 - I presume that you mean that the contribution of melt ranges from 0.86% to 40.59%, or it was estimated as 0.86% ± 40.59%. The use of a minus sign is not very clear due to the very wide range of values. Also at L45 and L77.

L140 – the drainage area is 2 x10⁵  …    km²? The measure unity misses.

L165 – 1.875° at 30°N is a rectangle of about 180x208 km, 3.7x10⁴ km². This means just a little more than five full pixels on a 2x10⁵ km² basin.

L180 – eight subzones. If I understand correctly  the subzones should be 12 (6 surface x 2 subsurface)

L188 – does the model consider even the melt caused by (liquid) rainfall? It could be rather important mostly (but not only) in the changing rainfall scenarios.

Table 3 – I presume that DDF_G is the degree-day factor for ICE (glacier) melt.

Table 3 – -4.28C is a rather cold temperature for melting threshold. Of course, this happens because of the daily average, but it is better to explain it.

L245 and following. This could be the main problem of the work. The temperature increase due to the climate change is far from being linear. See my general comments on modeling limitations.

L258 - Did you combine also temperature and precipitation perturbations in a single simulation?

L269 – Figure 2. Streamflow discharge appears well simulated, NSE = 0.82 is more than satisfactory in such a long period. SCA is more confused, but still well acceptable, as the isotopes simulation. The average glacier mass balance comparison with observed values (L280) has no meaning. Even a linear mass loss could simulate those 11 points.

L316 – decreasing rate

L333 – it is hard to believe that evaporation is not limited by water condition. Did you ever try to compare it with potential evaporation?

L362 – possibly in April the fraction of runoff due to snow melt increases. But SCA is going to decrease …

L385-386 – please control the measure unit (km³/yr and km³/s).

L388 – the glacier surface is (almost) impermeable but runoff from bédières rather quickly falls to the bottom through crevasses, moulins, wells. So, melt runoff flows mostly on the bottom of the glacier.

L498 – in my experience the melt component due to (liquid) rainfall can amount up to 15% of the total melt. In the REWs with larger GAR the increase of rainfall can lead to a greater runoff, if this component is considered in the model. Fig. 9d means that over a glacierized area a precipitation increase corresponds to a runoff decrease. Projecting the interpolation line, we can suppose that for a 35% of GAR there are no runoff changes when precipitation increases to 120%. The 35% of glacierized area literally eats the rainfall increase over the whole basin. Rather surprising!

L510 – the equation is useless, the correlation is high. But on fig. 10 also the “decreasing” REWs seems to have a correlation.

L513-517 – for “increasing” REWs the correlation between GAR, MAP and the increasing runoff is poorly understandable by the blue color intensity. Could you find a way to represent it with a number?

L535 – glacier meltwater increases with temperature, but glaciers are shrinking and, soon or late, this leads to a runoff maximum followed by a decrease. It has already happened to many glaciers.

L542-543 – and for the other regions the future will be the same, moved a little further.