GENERAL COMMENTS

The authors present an assessment of the implications of bias correction methods for the assessment of the effect of climate change impacts on hydrological drought in a Mediterranean catchment. They applied a well-known bias correction method and chose to evaluate the performance of each RCM based not only on conventional statistics, but also on drought statistics. The authors discuss methodological issues related to the comparison of the RCM performance through the application of a rainfall-runoff model at the monthly time scale.

The topic is relevant for the audience of Natural Hazards and Earth System Science, the objectives are properly identified, the methodology for the analysis is adequate and the conclusions are relevant and correctly supported by the results and discussion. The overall organization of the manuscript is adequate, and it is clearly written. The analysis clearly shows the agreements and discrepancies between results obtained with different climatic forcings for the hydrologic model of choice. Therefore, I support publication of the work in Natural Hazards and Earth System Science.

SPECIFIC COMMENTS

I have several suggestions and comments, which I believe would improve the paper:

• a) On section 4.2, the authors present their first assessment of discrepancy between historical observations and RCM control simulations. From Fig 4 and Fig 5, I gather that most models do a poor job at reproducing observed climate in the case study basin, particularly in seasonality of rainfall and temperature. I suggest adding a table with a comparison of mean annual values of precipitation and temperature to provide an objective comparison.
• b) The application of the quantile mapping technique is a critical step in the analysis. However, the authors do not provide much information on the procedure or the results while applied to the case study. There is a very brief introduction in the methodology section, with no details on how the original series are transformed. Regarding results, we can only see that bias for the three basic statistics has been eliminated. I think the authors should provide more information on the application of the technique to the case study and illustrate it with at least a figure showing the quantiles.
• c) The authors chose to use SPI as drought index to characterize precipitation, but they should state the aggregation time step chosen in the analysis. The descriptive statistics used later in the paper (frequency, duration, magnitude, and intensity) should be formally introduced.
• d) I was a bit confused by the classification procedure. If I understood correctly, the RCM are assigned penalty values from 1 to 10 according to their ranking in each of 7 statistics. The final classification is obtained by averaging of the penalization for all statistics. However, the index chosen is divided by a normalizing value to allow comparisons across statistics. Why not directly use the index values instead of the penalties based on the ranking, to account for the relative deviations shown by each model?
• e) On section 4.3, line 195, the authors state that there is a “correlation” between the order classification of corrected RCMs for meteorology and hydrology. By looking at Figure 9, I am not sure of this and I am afraid I must disagree. Figure 9 shows a scatter plot of nine values. The fitted regression line for the nine points has an R2 of 0.34, which is very low to conclude that there is a correlation (what is the significance level?). Even the blue line, which corresponds to only to 4 points, has a very low R2, of only 0.46. Finally, the authors should refrain from plotting the regression line for the two points corresponding to classification order <2, which obviously renders a perfect fit because there are only two points. By looking at the figure, I can also see an opposite “correlation” for the 5 points corresponding to classification order >4. The fitted regression line would have a negative slope, contradicting the initial statement. I think this discussion should be reformulated. We all agree that good bias correction would improve the agreement between climate models and observations, but the authors need to provide objective results to draw this conclusion, which, by the way, is a central part of their contribution. I suggest separating the analysis of conventional statistics and drought statistics, since the bias correction procedure is specifically focused on fitting the results of climate models to observations and therefore one can expect (as shown in Figure 4 and Figure 5), that the index values are very low. This does not necessarily have to be the case for drought statistics, which are linked to the tail of the distribution. Perhaps showing the scatter plots of the actual index values obtained with all models would illustrate better the comparison of performance for meteorological and hydrological drought.

TECHNICAL CORRECTIONS

From the formal standpoint, the paper is well written, correctly organized and adequately illustrated with tables and figures. I think the authors should rethink Figure 9 entirely.

The authors should consider changing the term “asymmetry” coefficient for “skew” coefficient.

Page 4, line 119. I believe the normalizing value used in the denominator of equation 1 is useful for comparisons across statistics, not across RCMs, because the normalizing value (historical observations) is the same for all RCMs.

Figures 8 and 13. Please change SPI into SSI, since the plots refer to streamflow droughts.

Although I am not a native English speaker, I believe the following expression should be corrected:

On page 4, line 114, … applying the “following” error index.

On page 4, line 119, … in order to make it comparable .

On page 6, line 174, … when “they” compared different statistical techniques.

GENERAL COMMENTS

This study aims to provide further insights on the selection of Global Climate Models (GCMs) -Regional Climate Models (RCMs) combinations, according not only to their skills to reproduce the local climate during the selected historical  period but also the local hydrology. Concretely, the authors calculate an error index for basic and drought statistics and use it to classify the GCMs-RCMs combinations according to their reliability for the assessment of meteorological and hydrological impacts. The selected methodology involves the bias correction of climate models' outputs through a quantile mapping (QM) approach based on empirical quantiles, the use of a lumped rainfall-runoff model to simulate monthly inflows from climate data and the use of standardized indices for drought characterization (namely the Standardized Precipitation Index (SPI) and the Standardized Streamflow Index (SSI)).

In my opinion, the paper addresses an important issue on the use of climate models' outputs for the assessment of climate change impacts at the basin scale. Besides, it is properly written and well presented. However, I miss a more critical approach to the potential shortcomings of the selected methodology, such as the underlying assumption of stationary bias, the impact of bias correction on the tails of the distribution (e.g. induced changes on the original climate change signal of the climate models), the pros and cons of pre-processing and post-processing the variables derived from climatic ones with regard to bias correction (e.g. performance of the hydrological model simulations over the validation period) or the potential effects of neglecting the inter-variable dependence of climate variables (e.g use of univariate bias correction methods against multivariable ones or ignoring the role of temperature in drought onset) on the assessment of climate change impacts on water resources.

SPECIFIC COMMENTS

Lines 27-29: "For instance,  we  have  Palmer  Drought  Severity  Index  (...)". I will also mention the Standardized Precipitation Evapotranspiration Index (SPEI, Vicente-Serrano et al., 2010).

Vicente-Serrano S.M., Santiago Beguería, Juan I. López-Moreno, (2010) A Multi-scalar drought index sensitive to global warming: The Standardized Precipitation Evapotranspiration Index - SPEI. Journal of Climate 23: 1696-1718.

Lines 91-92: "This is the reason that justifies  the selection of quantile mapping (using empirical quantiles) for this study". Have the effects of inter-variable dependence been considered before selecting an univariate bias correction method? In the case of alpine catchments, Meyer et al. (2019) found that incorporating or ignoring inter-variable relationships between air temperature and precipitation data could impact the conclusions drawn in hydrological climate change impact studies.

Meyer, J., Kohn, I., Stahl, K., Hakala, K., Seibert, J., Cannon, A. J. (2019). Effects of univariate and multivariate bias correction on hydrological impact projections in alpine catchments. Hydrology and Earth System Sciences, 23, 3, 1339-1354, https://hess.copernicus.org/articles/23/1339/2019/

Lines 105-109: "The meteorological drought analysis was developed by applying the Standard Precipitation index (SPI)". What about the role of temperature? As multiple authors have already pointed out, SPEI usually shows more severe increases in future drought events than those from SPI (e.g. García-Valdecasas Ojeda et al., 2021) and therefore I recommend to include it in the analysis. Which aggregation periods, statistical distributions and thresholds are considered for both the SPI and the SSI?

García-Valdecasas Ojeda, M., Gámiz-Fortis, S.R., Romero-Jiménez, E., Rosa-Cánovas, J.J., Yeste, P., Castro-Díez, Y., Esteban-Parra, M.J. (2021).  Projected changes in the Iberian Peninsula drought characteristics, Science of The Total Environment, Volume 757, 143702, ISSN 0048-9697, https://doi.org/10.1016/j.scitotenv.2020.143702.

Lines 136-146: In my opinion, the climate and hydrological regime of the Cenajo basin should be properly characterized in the Case study section.

Line 149: "CORDEX project (2013)". Reference?

Lines 151-152:  "We also  used  official  monthly  natural streamflow data within the Cenajo basin for the historical period 1972 -2001 (adopted as reference)". This reference period is not consistent with the calibration period of the rainfall-runoff model (October 1971 to September 2007, line 157).

Line 153: "(...) Spanish Ministry of Agriculture, food  and environment". The competences of this former ministry have been assumed by the current Ministry for the Ecological Transition and the Demographic Challenge.

Lines 156-161: What is the validation period? Goodness of fit for the validation period?

Lines 178-180: "The fit of the corrected control simulation series of streamflow to the historical series is not as good as for precipitation and temperature, but a remarkable improvement is observed". What could be the reasons for this? What about the performance over the validation period? (e.g. see Chen et al., 2021).

Chen, J., Arsenault, R., Brissette, F. P., & Zhang, S. (2021). Climate change impact studies: Should we bias correct climate model outputs or post-process impact model outputs? Water Resources Research, 57, e2020WR028638. https://doi.org/10.1029/2020WR028638

Line 189: "Note  that  in  this  case  we  refer  to  the  Standard  Streamflow  Index  (SSI)". This index should be properly defined previously (in the Methodology section), along with an appropriate reference.

Lines 198-199: "Therefore, we demonstrated that RCMs that allow better approximations of the meteorology provide better assessments of hydrological impacts". Although it seems quite straightforward (as rainfall-runoff models require climatic variables as inputs), I think that this statement should be carefully discussed before generalizing it: would it hold true if basins with very different hydrological regimes were considered? (e.g. important groundwater or snowmelt components?).

Lines 208-209: "Both  RCMs  predicts  a  decrease  of  the  variability  in  precipitation  and  an  increase  of  the variability  of  temperature  in  the  future". This is an interesting result, as precipitation variability is generally expected to increase in a climate change context (e.g. Pendergrass et al., 2017). Concretely, for the Mediterranean regions, Polade et al. (2017) concluded that a decrease in the frequency of daily precipitation events, combined with an increase in the amount of precipitation delivered in relatively rare heavy events, yielded greater year-to-year variability in total precipitation. In my opinion, this result should be discussed in the context of existing literature on future climate variability in the Mediterranean area. Which could be the potential role of bias correction in this result? For example, Maraun (2013) investigated the role of bias correction in modifying relative trends in annual precipitation maxima from a RCM and found that the RCM underestimated observed variability, which led to substantial amplification by quantile mapping of modeled trends in extremes. Besides, it would be interesting to examine the future trends obtained from the rest of the GCM/RCM combinations.

Maraun, D. (2013). Bias correction, quantile mapping, and downscaling: Revisiting the inflation issue. J. Climate,26,2137–2143, doi:10.1175/JCLI-D-12-00821.1

Pendergrass, A.G., Knutti, R., Lehner, F. et al. Precipitation variability increases in a warmer climate. Sci Rep 7, 17966 (2017). https://doi.org/10.1038/s41598-017-17966-y

Polade, S. D., Gershunov, A., Cayan, D. R., Dettinger, M. D., & Pierce, D. W. (2017). Precipitation in a warming world: Assessing projected hydro-climate changes in California and other Mediterranean climate regions. Scientific reports, 7(1), 10783. https://doi.org/10.1038/s41598-017-11285-y

Line 211: "predict significant decreases of streamflow (-43.5 and 57.2%)" Should it be -57.2%?

Lines 215-217: "In  the  case  of  the  meteorological  droughts  the  first  SPI  threshold  for  which droughts periods are detected in the historical scenario is -3.0. In the future scenarios this value is -5.2 and -4.6 for the RCM2  and  RCM9  respectively". I think that it will be interesting to assess the changes in the parameters of the future distribution with regard to the historical one (even if only the historical distribution is used to obtain the future SPI).

Lines 219-224: Check the signs of the SPI values.

Lines 231-257: in my opinion,  the Discussion section does not address properly the limitations of the selected methodology (see my previous comments).

The manuscript evaluates the ability of 9 CORDEX RCMs to simulate meteorological (i.e., precipitation) and hydrological (i.e., streamflow) variables, as well as drought statistics, in the Cenajo basin (Southern Spain). The best RCMS are then used to generate future scenarios.

General comments

The manuscript is interesting and within the scope of the journal. From a methodological viewpoint, some relevant details need to be specified to understand the validity of the proposed approach, with special reference to drought analysis. Some sections should be re-organized. Also, the language must be improved in some parts.

Major comments

The title of the manuscript is wordy and redundant. Please rephrase.

LL 56-57: The authors state that “In literature few works analyze the reliability of RCMs considering meteorological droughts.” Please add references to previous studies on this topic and highlight the main differences with your study. In particular, the manuscript would benefit from a comparison with a recent study by Peres et al. (https://doi.org/10.5194/nhess-20-3057-2020), dealing with a statistical methodological framework to assess the skill of the EURO-CORDEX RCMs to simulate historic climate (temperature and precipitation) and drought characteristics (duration, accumulated deficit, intensity, and return period), at seasonal and annual timescales, in Southern Italy.

L 148: The authors have performed a lumped analysis in the Cenajo basin. To this end, they have to specify:

• If the reference grids of both the historical data and the CORDEX simulations are equivalent; If not, how do they pair the information from the two grids?
• how many grid cells fall within the Cenajo basin;
• If the gridded historical and simulated precipitation data are spatially aggregated at the basin scale level and how.

The authors apply the SPI for meteorological drought analysis. However, it is not clear which time scale is used to aggregate monthly precipitation (1, 2, 3 months?) and which probability distribution is fitted to such data (gamma distribution?) for SPI computation. The authors should be aware that if they simply calculate the standard normal values corresponding to the differences of monthly precipitation data and the related monthly means, divided by the related monthly standard deviations, they do not obtain SPI, but another index known as the Standardized Rainfall Anomaly (Jones and Hulme, 1996), which is equal to the SPI only if aggregated precipitation data are normally distributed.

Moreover, once that the SPI series is computed by using different threshold values, drought characteristics such as frequency, length, magnitude, and intensity are determined. The authors do not clarify how these characteristics are computed. Nonetheless, I believe that, for instance, drought magnitude for drought events longer than one month has been computed as the sum of SPI values over the length. Is it correct? If so, the approach is misleading since a SPI value already quantifies the magnitude of a dry or a wet period occurs during the considered aggregation period.

The number of drought events identified for each considered threshold should be indicated in a table, together with the mean values of the corresponding characteristics. I am afraid that for the control scenario, very few droughts are identified for threshold values corresponding to severe and extremely dry conditions. Thus, I wonder how fair could be the comparison between observations and simulations? In addition, if the analysis is lumped (i.e., a single series for the whole basin is considered for each variable), it would be interesting to ascertain whether the drought statistics evaluated on RCM simulations correspond to the same drought events identified on the historical series.

Finally, bias correction through quantile mapping applied to SPI (if precipitation) or to SSI (if streamflow) series is a little confusing since these series are standard normal distributed by definition, therefore I do not expect big differences between the historical series and the control simulation series, unless due to sampling variability. Please clarify this point and explain the results illustrated in Figures 7 and 8.

Minor comments

L 10: Hydrological impacts of what? Maybe, change with “hydrological response”.

LL 10-12: “It assumes that … when they provide better approximation to the historical basic and drought statistics.” This sentence is rather unclear and must be rephrased.

LL 18-20: In the last sentence there is no reference to the future scenarios of hydrological droughts.

LL 103 and 114: The term “goodness of fit” is usually applied to describe how well a statistical model (e.g., a probability distribution) fits a set of observations. I am not sure it is appropriate for RCM simulations.

L 162 and L 192: Sections 4.2 and 4.3 have the same title. Merge the two sections.

L 219: “the threshold of “-“ 1.7 of SPI (considered to define extreme droughts …)”. Usually, -2 is used for extreme droughts.