Dear authors, please find the comments in the attached PDF. 

This manuscript addresses an important issue in hydrological modeling: the challenge posed by equifinality in model parameters, especially in response to significant environmental disturbances such as wildfires. The authors effectively demonstrate that equifinal parameter sets can yield widely varying streamflow predictions following a large disturbance. To mitigate this uncertainty, the authors propose leveraging nonstationary information, such as dynamic vegetation maps by leveraging a hybrid approach, combining a physically based model (streamflow simulation) with a statistical metamodel (to account for non-stationarity). Overall, the manuscript is logically structured and addresses a critical gap in ecohydrological modeling. The methodology is well-designed to address the stated research questions; however, additional clarification of the complex approaches implemented would facilitate easier understanding. The conclusions drawn by the authors are clearly justified by the analyses presented.

However, there remain several areas that could be improved or clarified before publication. Specifically, the following points should be addressed:

• Parameters to account for uncertainty in meteorological data:

I understand the idea that meteorological data are uncertain and can significantly influence model outputs. However, the parameter range used for correcting meteorological biases (±25% for precipitation, ±4°C for temperature) appears very large. Are there existing studies or evidence supporting this magnitude of potential meteorological bias specifically within this basin? My concern is that using such broad correction ranges might allow the hydrological model to artificially add or remove water to better match streamflow observations, which themselves are uncertain due to reconstruction processes, thereby compensating for potentially missing or poorly represented processes.

• Reconstructed streamflow data:

The presence of a reservoir near the gauge station considerably disrupts natural streamflow patterns. The approach of reconstructing streamflow to remove these effects (naturalized flows) partially restores the basin's natural characteristics but leaves the "observed" data highly uncertain. Such uncertainty can significantly impact calibration experiments. Could you clarify the method used to reconstruct the streamflow? Specifically, does the reconstruction explicitly account for changes introduced by the 2020 Creek Fire disturbance?

• Calibration thresholds on the objective functions:

In defining the “behavioral” parameter sets, thresholds were applied selectively to NSE, log NSE, yearly MAPE, and April-July MAPE, but notably not to the snow criteria. Could you elaborate on why snow metrics were excluded from this subjective filtering? Additionally, could you clarify whether these subjective thresholds are truly necessary or if the results would have differed significantly by simply selected the 30 "best" members from the initial ensemble across all calibration metrics? A more detailed justification for choosing these thresholds would enhance transparency and reproducibility.

• Calibration and evaluation consistency:

If my understanding is correct, the 30 DHSVM parameter sets were derived from calibration experiments using dynamic vegetation only. Thus, comparing model performance under conditions for which it was explicitly calibrated (dynamic vegetation) against for which it was not calibrated (static vegetation conditions) might be unfair. To address this concern, would it be possible to conduct an additional calibration experiment using static vegetation maps? This would allow a more balanced and fair comparison, using these static-calibrated parameter sets as an appropriate benchmark against the dynamic-vegetation approach presented here.

Additional minor comments and technical corrections can be found in the attached pdf.

Overall, the manuscript presents significant advancements in the field and should be published following minor revisions to address the points mentioned above.

Sincerely,

Cyril Thébault