peer review report 1 on assessing the role of uncertain precipitation estimates on the robustness of hydrological model parameters under highly variable climate conditions

Journal of Hydrology: Regional Studies (2017) 96–98 Contents lists available at ScienceDirect Journal of Hydrology: Regional Studies journal homepage: www.elsevier.com/locate/ejrh Peer Review Report Peer review report on “Assessing the role of uncertain precipitation estimates on the robustness of hydrological model parameters under highly variable climate conditions” Original Submission 1.1 Recommendation Minor Revision Comments to Author: Comments to Author: This study assesses the model performance of the LISFLOOD rainfall-runoff model in medium sized catchments in South Africa More specifically, streamflow simulation results with seven different satellite-based precipitation products are evaluated using a split-sample test, enabling also to analyze robustness of calibrated parameter values The model runs on daily time-steps, but evaluation is performed with monthly values The main findings of the authors are in line with previous studies including (a) ranking of satellite-based rainfall products and (b) problematic simulation bias if climate in calibration period differs strongly from independent evaluation period Overall this is a classical modelling study applying well known concepts The paper is written well and has a clear structure Some of the figures I found hard to understand (see general comment below) The main contribution of the paper is that the analysis is carried out in a region where there is still lack of similar studies It is also of value that the findings of previous studies (in other regions) are confirmed A drawback is that the study lacks (as is many other publications too) real new, innovative contributions However, I think the study fits well into the scope of the journal and should be published after minor revision I hope my comments are helpful regards, Harald Kling General comments: Non-stationary climate: Throughout the paper the authors use the term non-stationary climate However, given that the simulation period 20022010 (i.e × years) is quite short, I ask the authors to re-think if it is non-stationary climate or simply natural fluctuation between wet and dry years Especially given that for climate analysis the focus is usually on at least 15–20 year periods Selection of basins: The study starts with 10 basins but then six of them are excluded after finding that there is a low correlation between rainfall and streamflow The authors argue that this is an indication that there is significant impact of water resources management (reservoirs) Significant impact of reservoirs might be true, but low correlation between rainfall and streamflow is not a convincing argument because also in undisturbed catchments there might be a low correlation E.g for the upper DOI of the original article:http://dx.doi.org/10.1016/j.ejrh.2016.09.003 2214-5818/$ – see front matter http://dx.doi.org/10.1016/j.ejrh.2016.12.052 Peer Review Report / Journal of Hydrology: Regional Studies (2017) 96–98 97 Zambezi the correlation between monthly rainfall and streamflow is about r = 0.25, but simulation results can be quite accurate with KGE’ = 0.90 and above (see Kling et al., 2014, Journal of Hydrology: Regional Studies: “Impact modelling of water resources development scenarios on Zambezi River discharge”) The reason is that in many African basins the rainfall at the start of the rainy season does not generate runoff as the soil first has to reach a specific wetness Consequently there is low correlation between rainfall and streamflow time-series I suggest the authors simply never mention the six excluded basins (you not need to dwell on the low correlation) A priori parameter values: On page you describe the data sources for a priori parameter estimation for the LISFLOOD model However, later on in the paper you never discuss the simulation results with the a priori parameters It would be interesting to report about the model performance in the two evaluation periods (is it always poorer than with calibrated parameter values when switching calibration/evaluation periods?), what performance sub-components are affected (r, beta, gamma) and what is the relationship between a priori parameter values and calibrated values? The simulation with a priori parameters would serve as a good baseline against which all other simulation results could be compared Kling’s personal communication (2012): On page 15 (line 348–350) you cite Kling’s personal communication for classification of KGE values into good/medium/etc Here is an excerpt from the original e-mail to Vera Thiemig (18.10.2012) serving as the reference: [— e-mail excerpt start —] Given the above three points, I suggest you use much stricter values for classification, such as: 1.00 > KGE > 0.75: good 0.75 > KGE > 0.50: intermediate 0.50 > KGE: weak I expect that for the “weak” simulations, the most likely cause are biased precipitation inputs, and not so much the calibration of the hydrological model This would show up in poor beta values, but good variability ratios (if you base the variability ratio on coefficient of varation) Therefore, I strongly suggest that you base the evaluation not only on KGE, but also on the three sub-components, as this is a great way of learning the causes for poor model performance: poor timing and shape −> poor correlation, poor mass balance −> poor bias ratio, poor distribution of flows −> poor variability ratio [— e-mail excerpt end —] A value of KGE = 0.75 is not yielding a hydrograph simulation which looks perfectly accurate Therefore it might be worthwhile to include an upper class KGE > 0.9 named “excellent” I would also like to point out that KGE is just a summary measure and it depends on the model application what sub-components are important For example, for assessing long-term reservoir inflows the bias is important, but correlation is not so important For flood forecasting applications the bias is not important but correlation is very important For assessing long-term energy generation at run-of-river hydropower plants the bias and variability terms are very important (flow duration curve), but not correlation In summary, you have to know the application of the model, before a reasonable classification into good/medium/etc performance is feasible What is the intended application of the LISFLOOD model? Fig and Fig 9: I understand the dots shown in the figures, but after several minutes of thinking I have to admit I am still not sure how to precisely interpret the bars And I am sure some of the readers will have similar difficulties Please give a better description in the caption and text Specific comments: “a non-stationary hydrological model” The model is not non-stationary, probably you mean “non-stationary” climate? However, see also my general comments above about the use of the term “non-stationary climate” Abstract, first paragraph: “LISFLOOD hydrological, and” −> “LISFLOOD hydrological model, and” page 1, line 12: I suggest to remove “However” at the start of the sentence page 1, line 14,15: I suggest to remove “On the other hand” at the start of the sentence page 3, line 68: “reliable time-series of precipitation” I would remove the word “reliable” when referring to satellite-based precipitation data page 3, line 71: In the references you should also cite: Cohen-Liechti et al., 2012 HESS: Comparison and evaluation of satellite derived precipitation products for hydrological modeling of the Zambezi River Basin page 4, line 92: “variability in will” −> “variability will” page 7, line 162: For TRMM 3B42 V6 & V7 you use the real-time product or the “high quality” product (which lags by about months)? page 9, line 209: 98 Peer Review Report / Journal of Hydrology: Regional Studies (2017) 96–98 Why you use the “standardized monthly mean of the precipitation”? Standardization should not affect your computed correlation values page 9, line 215: “justifies the choice of using only regions A1-4” I not understand the reasoning See also general comment above page 10, line 230: “to he” −>“to the” page 17, line 420: “optimum value” −>“ideal value” See the original KGE publication by Gupta et al (2009) how they use the terms “optimal” and “ideal” page 20, line 488, 489: “KGE’ efficiencies” −>“KGE’ values” KGE = Kling-Gupta Efficiency KGE efficiency = Kling-Gupta Efficiency Efficiency (2x) page 22, line 536: “dominated limited factors” Do you mean “dominating limiting factor”? This has a different meaning Ph.D Harald Kling Hydrologist Pöyry Hydro Consulting, Hydropower, Kranichbergg Vienna, 1120 Austria ... hydrological model, and” page 1, line 12 : I suggest to remove “However” at the start of the sentence page 1, line 14 ,15 : I suggest to remove ? ?On the other hand” at the start of the sentence page... “standardized monthly mean of the precipitation? ??? Standardization should not affect your computed correlation values page 9, line 215 : “justifies the choice of using only regions A1-4” I not understand the. .. some of the readers will have similar difficulties Please give a better description in the caption and text Specific comments: “a non-stationary hydrological model? ?? The model is not non-stationary,