I have read with interest the manuscript entitled ‘Guidance on evaluating parametric model uncertainty at decision-relevant scales’. The study examines the sensitivity of the simulations of a spatially distributed ecohydrological model to the model parameters for calibration purposes. Sensitivity is considered with respect to different model output metrics that correspond not only to performance metrics assessed at the basin outlet, but also statistics of the model output calculated for the different hillslopes of the basin where no output observations are available.

The study is a welcome contribution to the field of sensitivity analysis of spatially distributed models, which is challenging due to the high dimensionality of the parameter space of these models and which requires further investigations. The study discusses the issue of calibration and uncertainty estimation in absence of output observations, in particular at internal locations of a river basin where model-based information is critically needed to support decision making.

Overall, the manuscript is well written, the analyses were performed with care and the experiments are well documented in the supplements. However, I have a number of suggestions and I think that a number of points need clarification, in particular regarding the choice of (output) sensitivity metrics and the analysis of the multipliers, as detailed below.

p1 L8 ‘parameter multipliers‘: I suggest adding ‘for spatially distributed parameters’ for clarity.

p1 L15-16 ‘for some parameter multipliers […] reducing dimensionality.’: This needs clarification.

p2 L31 ’sensitivity metrics’: I suggest specifying what this term refer to for clarity (e.g. performance measure or statistics of the simulated model output).

p2 L34-35: Could you provide some references/examples for this?

p2 L37-39 ‘Matching […] with controlling extremes.’: A link between this sentence and the rest of the paragraph is missing.

p3 L92-93 ‘will be evaluated […] error model’: The authors should clarify whether they refer here to future studies that may use the guidance presented in the manuscript.

p4 L107 ‘performance measure’: Consider revising the terminology. The metric of Eq. (2) is not really a performance measure, as it does not use observed values.

p5 L130-131 ‘Therefore, […] the TN estimation method’: This sentence is not clear to me. Observed data are also used for the streamflow objectives (Eq. (1)). In addition, does the water quality objective only consider the basin outlet or also hillslopes?

p5 Section 2.3 A justification for the choice of the four calibration performance measures is missing (e.g. a justification could be that these metrics are typically used in previous studies, and in this case some references to some of these studies should be provided). In addition, I think that the metric of Eq. (1) could also be used for calibration purposes. Therefore the difference between the decision-relevant and calibration-relevant sensitivity metrics is fuzzy.

p6 L170 ‘EEs for each parameter […] in the parameter domain.’: Please add a reference for this (e.g. Pianosi et al. (2016)).

p6 L178 ‘Step changes’: Does this refer to the quantity Delta_{s+1,s,p} of Eq. (7)? Please clarify.

p6 L178-179 ‘to allow for a uniform […] within the specified bounds.’: This needs to be better explained.

p8 L217 ‘mean EE value’: Clarify whether this refers to the metric defined in Eq. (8).

p8 Sect. 2.6: I do not understand this point. The use of a multiplier for a certain parameter type is based on the assumptions that the value of the parameters of this type vary in the same proportions in different locations of the basin, and I do not understand why these parameters should have similar sensitivity (here EEs).

p11 L300-303: This is based on a rather strong assumptions that the C-Q relationship obtained using the WRTDS regression method can be extrapolated to other flow conditions. This may not be appropriate for in-depth nitrogen study, in particular in agricultural areas with highly varying nitrogen inputs. I understand that it is not the point of this study to have a sophisticated nitrogen routine. However, I think that some comments should be added to further highlight the simplifications/limitations of the data-driven nitrogen routine used. My comment also refers to the sentence p21 L503-505.

p13 L359 ‘healthy ecosystem’: I suggesting replacing this expression by something like ‘more humid ecosystem’ for clarity.

p14 L373-374: ‘This result demonstrates […] to calibrate.’: This sentence needs clarification.

p21 L511-514: The authors could refer to the study by Cuntz et al. (2016), which also demonstrates the importance of including in the calibration some parameters that are typically set to fixed values and in particular hard-coded parameters, using the NOAH-MP land surface model.

p22 L553-555: I think that it should also be emphasized that, because of the issue equifinality, calibration strategies that identify an ensemble of possible parameter sets (as compared to a unique ‘best’ solution) and that therefore consider parameter uncertainty are more appropriate.

Minor edits

p3 L61: I suggest replacing ‘will’ that ‘are typically’.

p7 L192: replace ‘smaller’ by ‘smallest’.

References:

Cuntz, M., Mai, J., Samaniego, L., Clark, M., Wulfmeyer, V., Branch, O., et al. (2016). The impact of standard and hard-coded parameters on the hydrologic fluxes in the Noah-MP land surface model. Journal of Geophysical Research, 121(18), 10,676-10,700. https://doi.org/10.1002/2016JD025097

Pianosi, F., Beven, K., Freer, J., Hall, J. W., Rougier, J., Stephenson, D. B., & Wagener, T. (2016). Sensitivity analysis of environmental models: A systematic review with practical workflow. Environmental Modelling & Software, 79, 214–232. https://doi.org/10.1016/j.envsoft.2016.02.008

1. L8-10: “We evaluate six sensitivity metrics that align with four decision objectives; two metrics consider model residual error that would be considered in spatial optimizations of engineering designs.” This sentence is confusing -- do the six sensitivity metrics add up to the four decision objectives + two model residual error metrics?
2. L39: extremes or high flows?
3. It would be helpful to somewhere define “engineering controls”.
4. L 116: here are you write flooding, low flow, reservoir water supply objectives, but earlier you wrote flooding, low flow, and all other flows. If these are the same, that should be explicitly stated.
5. L320: “The goal of this sensitivity analysis is to inform the selection of parameters to calibrate a RHESSys model that could be used in such a reforestation optimization.” is this the overall goal of this paper? If so, this should be stated in the introduction section much earlier.
6. L 328: are the elementary effects for all the parameters normalized on a percentage basis? Why compare the 95th percentile for the elementary effects to the overall mean of all parameters’ elementary effects, if that is what is being explained in this sentence? What does the 95th percentile estimate for the elementary effects mean?
7. Figure 3 seems to be referred to before figure 2 (L335).
8. L349: does an elementary effect value of exactly 0 mean that this parameter has no effect on the stream flow or hillslope metric? It would be helpful to state this explicitly.
9. The text discussing figure 2 is useful, (line 348 in the rest of this paragraph), but without knowing what the specific parameter numbers are in figure 2, I'm not sure what to take from this graphic.
10. Line 480: I see now that engineering designs are not explicitly evaluated in this paper. My earlier comment (comment 3), asked about what engineering controls meant. The focus on engineering controls in the introduction section led me to believe that this paper would be about engineering controls. Rather it seems that this paper has implications for where to locate engineering controls but does not directly investigate this placement. If this is accurate, then I would suggest deemphasizing engineering controls from the introduction section.
11. From what I understood of this article, the first main finding was that parameters describing watershed characteristics are sometimes important for modeling hillslope hydrologic response even though they do not affect the streamflow at the model outlet much. The authors state that this might be important in the spatial location of engineering controls. There are many other reasons why getting the hydrology right within the watershed is important (modeling of spatially distributed soil moisture, etc.), but a major limitation is that we don't normally have data to compare to within the watershed, so in practice it would be hard to calibrate these parameters that don't affect streamflow much. The second main finding was that commonly used metrics (e.g., NSE) are not as sensitive to the decision relevant streamflows that we would want them to be. These are both important findings and points to make, but I found the article overall hard to read and understand. The authors may be served by focusing the text on the main findings and reducing discussion of peripheral topics.