General comments
This is the second time I review the paper and it is nice to see that it has been improved from the last version. However, with regards to the presentation I would still urge the authors to revise the text thoroughly to improve the clarity in the presentation of their work. I have given only a few technical comments below in this respect.
With regards to the content, the main thing that needs to be addressed is a better explanation of the fitting of the variogram and Figure 4, see comments below.
Specific comments
P2, line 25-26. But the low flow performance was not so good for the un-nested case, wherefore this conclusion should be modified. Also exclude emphasis words like “very”. I would also suggest focusing the abstract more on what was done in the study. I would suggest presenting the three study aims and the corresponding results, including more information about the TND indices and the sensitivity analyses that were done and reducing the general sentences about FDCs in the beginning.
P14. This section should refer to the new information in the supplementary material.
P15. L8-9. It should be specified how many bins were used. From Figure 4 it looks like there are fewer bins than the 5 parameters in the variogram fitted to them, which seems like overfitting?? You should also comment on how well the variograms were fitted to the binned data, the fit does not seem to be that good from Figure 4? Additionally for Figure 4:
• It should be “semivariance” on the y-axis and the x and y axes should go down to zero so the whole variogram can be seen. What is the difference between the red diamonds (and the other markers) of varying size? The dark red diamond is very difficult to discern from the bright red.
• The semivariance cloud could be shown in the background so the appropriateness of the binning used can be assessed and also the quality of the fit.
• Also it is unclear what 300km2 vs 75 km2 means, is it referring to catchment pairs with areas in the intervals 1-300km2 vs 1-75 km2?
P15 L22, could specify again that d is duration
P19, line 19-25. “A comprehensive error index…” but this seems to be just the summed absolute errors at the selected points? How were these points chosen? In the same way as on P15? The authors could consider moving the whole paragraph about the evaluation methods on page 15 and 16 to this section about the performance indices so that the prediction method is discussed first and the evaluation of it later to keep a clearer structure of the methods section.
P21. Line 11-14. Please rephrase this in a more unbiased way.
P22. Line 15. Please comment on the negative kriging weights and why these occur (screening effect, etc, see Isaaks and Srivastava, 1989 and other standard kriging references). Were there any negative weights included in the prediction results for n=6? This could lead to more extreme values of predicted TND values than the distribution of the empirical values and could then potentially lead to prediction of negative streamflow quantiles in catchments with low low-flow values, which would be good to be aware of for users of the method.
P22 line 14-15 it says that “….(2) any nested pair of catchments is associated with a high or very high Beta value.” Whereas in the figure all nested catchments are on the left-hand side with small Beta values.
P22. Section 5.5.2 and Discussion. It would be interesting with more discussion about the reasons for the difference between the MAF and MAP* results.
P23. Line 3. “NSE are insensitive to n” From the figure it does not seem insensitive, NSE is getting better with a higher n (after the initial drop) with about the same difference that it is getting worse with higher n for MAF. Please rephrase this in a more unbiased way.
P23. Lines 18-20 is contradictory to Lines 15-17 where it says that the information was neglected after point 2 (and thus the estimation of the variogram).
P23. Line 23. Even if the results are analogous the NSE and LNSE values could easily be reported here. It would also be interesting with some discussion as to why the performance was mainly worse for low flows when removing the nested catchments.
P24. Line 25-26. The fact that the prediction method is using a weighted linear average would also mean that more extreme values than in the data pool cannot be predicted (without negative kriging weights, see above).
P26. Future analyses. I would suggest the following analyses as well: 1) It would be interesting to compare to other methods that use weighted combinations from dynamic pooling groups (e.g. Burn et al, 1990; Holmes et al., 2002, these could also be discussed in the introduction) and that incorporate other information than spatial proximity. 2) It would also be good to test the un-nested prediction capability against other regional models that are also applied to the same un-nested dataset.
Technical comments
P3. Line 3. “has been” instead of “can be” when it refers to the historical period.
P3. Line 13. “Peculiar” means “odd”, “strange”, “weird”, etc in English and this is not a good choice of word in this type of text. This is used in several places in the text and should be replaced. In this place “…their advantage of being readily understandable…” could be used instead. On P7, line 12 “A special feature…” can be used, etc.
P11. Line 25. Delete “The discriminant between the two ways resides in the fact that”, it is unnecessary and start the sentence at “The uncertainty…”
P12. Line 25-26. This sentence can be deleted since the method is presented generally for just “TND” in the following text.
P15 L18-19. Replace “each and every” with “each”, here and at many other places in the manuscript to improve the language.
P22, Line 3. 365 is a number and not “a … resampling scheme”
P22, line 11, end of line, add “i,j=1…18 and” after “with”
References
Burn, D. H. (1990), Evaluation of regional flood frequency analysis with a region of influence approach, Water Resour. Res., 26(10),2257–2265.
Holmes, M., A. Young, A. Gustard, and R. Grew (2002), A region of influence approach to predicting flow duration curves within ungauged catchments, Hydrol. Earth Syst. Sci., 6(4), 721–731.
Isaaks, E. H. and R. M. Srivastava (1989). Applied Geostatistics. New York, Oxford University Press. |