|In my previous revision of the manuscript of Piras et al, I requested a major revision since no proper validation of the downscaling methods and the hydrological model are presented, in order to assess the uncertainties. Usually a major revision requires additional results or some changes in the methodology, instead the authors preferred to argue and just modify some text in the manuscript. The authors even refuse to show some simple comparisons, such as requested by reviewer n°1 in his first comment (“one-to-one (coarse vs high-resolution setup) comparison”).|
To my opinion this is not sufficient to address the issues mentioned in the previous revision. I recommend either a straight rejection, if the authors cannot provide new results to justify their methodology, or a major revision, again, if they are willing to revise their paper in a constructive way. At the very least, in the conclusion it should be clearly stated that there are strong uncertainties at the different steps of the modeling chain.
Please find below some comments/suggestions to the answers provided by the authors:
>Regarding the comments related to the uncertainty, we acknowledge that a complete approach >would require the use of several combinations of Global and Regional Climate models, multiple >downscaling methods, and multiple hydrologic models.
This is true, however at least it is possible to validate the different methods chosen for the catchment of interest.
>In our case, to simulate the complex hydrologic process of a Mediterranean basin, we selected a >physically-based hydrologic model to conduct high-resolution simulations. As it is well-known, >this type of models requires a significant computational effort when run on multidecadal periods, >as those involved in climate change studies. For example, the 256 years of simulations took us >880 hours of CPU time over 64 processors.
This probably overkill, by comparison with the available data (3 years in the beginning of the XXth century). A clear justification about the use of such a model is lacking.
>In a future study, we are planning to compare outputs of the hydrologic models that have been >applied on the RMB by other research groups within the CLIMB project. This will allow addressing >somehow the uncertainty of hydrologic simulations. This has been added in the paper conclusions >(page 23, line 8-10).
The identification of a proper hydrological modelling strategy is probably the first step, prior to make future projections with a given model with high uncertainties.
>The hydrometeorological dataset used to calibrate the hydrologic model included:
>- Daily discharge data at the Rio Mannu basin outlet for 11 years from 1925 to 1935. These data
>were acquired from the technical reports of the Italian Hydrologic Survey. After a quality
>control based on the analysis of the stage-discharge relations published every year and other
>notes present in the reports, we were able to identify three years that we judged as those with
>the most accurate data.
>- Daily rainfall data from 12 gages within or near the basin in the same period 1925 -1935.
>- Daily minimum and maximum temperature from one thermometric station located close to the
>basin in the same period.
This is clearly no enough hydro-meteorological data, to perform a climate change impact study with such a complex modelling chain.
>The precipitation downscaling model was calibrated using:
>- Precipitation records at 1-min resolution from 204 automatic rain gages observed during the
>years 1986–1996 in the coarse grid of 104 km x 104 km shown also in Fig. 1b of this
>manuscript. Note that this domain entirely includes the catchment.
>The downscaling algorithm that allows deriving the hourly potential ET was calibrated using:
>- Hourly meteorological data from 1 station over the period 1995–2010. The observed
>meteorological variables include all the variables required to apply the Penman-Monteith
On the opposite, I agree this dataset allows calibrating the downscaling methods over the domain size. What about validation for the catchment of interest? Sometimes downscaling methods perform well on a coarse resolution but have some troubles locally. For example the validation could be performed with a split sample test in the catchment of interest.
Another idea would be to use this data set to force a hydrological model during the period 1986-1996, since observed discharge is not available, it could be a hydrological model that requires either no calibration (simple water balance models) or a model with tables allowing to relate its parameters to land use and soil properties (ex: SCS-CN, SWAT…). Can the parameters of the RMB model be fixed a-priori without calibration?
>We agree with Reviewer 1 that a longer period of calibration and validation would increase the
>robustness of the hydrologic model. The Rio Mannu basin was selected as one of the study sites >of the CLIMB project (funded by the 7th Framework Program of the European Union) because of >the presence of agricultural fields and of an experimental farm where several data of crop >productivity are being collected. One goal of CLIMB is, in fact, to evaluate the impact of climate >change on local economic activities. This basin possesses then ideal characteristics to accomplish >the project goals.
Ideal except for the hydrometric data availability.
>Unfortunately, in our watershed, observed discharge data were only available in the period 1925–>1935. Furthermore, as mentioned in the answer to comment 1, after a quality control, we were >able to identify only three years (1930-1932) that we judged as those with the most reliable data.
If there are some agricultural activities in the catchment, it is likely that some land use change has occurred since 1932 (new agricultural practices, changes in irrigation systems..etc.), therefore the associated hydrological processes may be different in the period 1970-2000.
>As a matter of fact, the problem of lack of discharge data is common for other basins in Sardinia >and for most Mediterranean catchments. Thus, if we want to estimate the impact of climate >change on water resources and local economic activities in these study regions, the limitation of >observed discharge records needs to be accepted most of the times.
I strongly disagree. We should first focus on sites with sufficient data and then try to adapt the methods and modelling schemes to un-gauged or partially gauged catchments. This is the same strategy for regionalization: the methods are first validated in catchments with sufficient data. When not enough data is available, novel modeling strategies should be used to cope with the lack of data instead of hiding the uncertainties behind an overly complex modelling framework.
>Regarding the hypothesis of stationarity of bias correction and more in general calibration >relations and model parameters, which was assumed to hold in current and future climate, we >acknowledge that, in principle, it may not be valid. However, as a matter of fact, it is assumed in >the great majority of the climate change studies.
Indeed, many climate change studies are performed with little scientific grounds. I have nothing against climate change studies, but prior to make future projections the robustness and the skills of the methods should be assessed. This is the main motivation for projects such as COST-VALUE (http://www.value-cost.eu/). Since the hypothesis of stationarity can be tested, at least during historical conditions, there is no valid reason why this test should be avoided. See Maraun 2012, Tramblay et al. 2013 (references listed in the previous revision) or Teutschbein and Seibert 2013 (http://www.hydrol-earth-syst-sci.net/17/5061/2013/hess-17-5061-2013.html).