Evaluating the Performance of Objective Functions and Regional Climate Models for Hydrologic Climate Change Impact Studies: A Case Study in the Eastern Mediterranean
Abstract. The robustness of hydrological models used in projections of future fresh water resources is compromised due to non-stationary climate conditions. This study aims to (i) develop a method for selecting a skillful hydrological model parameterization under changing climate conditions and (ii) apply a calibrated hydrological model to assess streamflow projections for 38 mountain watersheds in the eastern Mediterranean island of Cyprus over the next decades (2030–2060). A matrix-based approach was developed to evaluate six objective functions by eight performance measures. Using the GR4J hydrological model, evaluation matrices were computed for multiple 5-year simulation runs covering 1980–2015. The matrices covered 14 model calibrations and 182 validations in total, as well as 4 sets of validations under different climate change conditions, for each watershed. Based on the matrix method, the Nash-Sutcliffe Efficiency with square-root transformed streamflow resulted in the best performance for streamflow simulations in Mediterranean watersheds experiencing drying trends. This method is transferable and can be applied in different climate regions to identify the most suitable objective function and model parameterization for hydrologic climate impact assessments. Eighteen Regional Climate Models (RCMs) were bias-corrected, downscaled to 1 km and used to simulate streamflow with GR4J for 1980–2010. Nine RCMs underestimated the fraction of wet period precipitation (60–73 % instead of 82 % of annual precipitation), causing streamflow biases up to 40 %. The remaining nine RCMs selected for the study simulated the seasonal precipitation cycle accurately. The median of future projections showed a 6 % reduction in precipitation and a 17 % reduction in streamflow. In the worst case, reductions could reach 16 % and 39 %, respectively. Notably, during the driest years, streamflow reductions could reach 70 % relative to the driest years in the past. Our findings suggest that terrestrial water resources in the eastern Mediterranean may significantly deteriorate in the coming decades.
In this manuscript the authors demonstrate a robust hydrologically modelling method for simulating streamflow under projected future climate conditions.
While there is not anything particularly new in this manuscript the overarching method is well considered and supported by comprehensive modelling experiments. The manuscript is well structured, and the scientific literature well referenced throughout. Figures and tables are appropriate. Manuscript is generally well written but, in some cases, mixes tenses – benefit from further proofread to improve clarity.
Subject to revision this manuscript would make a useful addition to the scientific literature.
Specific comments
Abstract
Ln 12 : Insert word …..“conceptual” hydrological models….Need to make it clear early that you are referring to conceptual hydrological models here. i.e. physically based models may not be compromised by non-stationary climate conditions.
Ln 14 Is “assess” the correct word here?
Ln 18 here and later it is not clear to me how multiple 5-year windows between 1980-2015 resulted in 14 calibration and 182 validations. A little more explanation is required in the main body of manuscript, as it is not intuitive.
Ln 19 Matrix method. Reword, ambiguous.
Ln 24 “…used to simulation streamflow with GR4J…” reword to make it clear streamflow was simulated using GR4J using inputs from the RCMs.
Ln 30 Here and elsewhere I don’t really like the term deteriorate in this context. It is ambiguous. Could you be more specific e.g. mean annual streamflow will decrease.
Ln 69 Hageman et al. (2013) is more than 12 years old. Is there not a more recent study using CMIP models?
Ln 78 Which phase of the CMIP?
Ln 80 RCP? So this manuscript is using CMIP5 models. It does beg the question how different CMIP5 is from CMIP6 over the Mediterranean region? Hopefully this is covered in the discussion as it would be necessary to place the results of this study into context with the latest climate modelling.
Ln 82 insert word “mean”? e.g. “…highlighted a MEAN annual precipitation reduction of…”
Ln 80-100 The introduction doesn’t make clear to me what the new scientific contribution this manuscript makes.
Data and methods
My main comment with respect to methods is there is no justification for the adoption of the 5-year calibration (and validation) window length. I understand that one wants windows short enough to have distinct wet/dry phases and I understand models were selected based on calibration and validation performance but why 5-years? However, considering the principle of ’equifinality’ is 5-years sufficient for calibration? Would the results/conclusions be different for a longer window length (minimum of 10 years is typically used)?
Ln 106 There are only two transformations not three. i.e. “ 1) no transformation; 2…”
Ln 127 Not clear how the validation were undertaken. Did they also have a 1 year warm up period?
Ln 149 method not methodology. Methodology is a study a methods (e.g. a study of different farming systems is a methodology).
Ln 156 I think this needs rewording as I don’t know what is a “..typical annual and interannual variability in precipitation of Mediterranean climates”? South-eastern Australia and South Africa have mediterranean climates and they are among the most variable in the world.
Ln 159 I assume these are all unregulated with minimal landuse change over the experimental period? This isn’t stated anywhere.
Results
Figure 3 it is difficult to see change in the heat map. Can the gradient be modified to better show changes (ie introducing a third colour into the colour ramp?)
Ln 366 dam storage? I think dam yield would be more appropriate? Besides changes in runoff and changed in dam yield under future climate projections are not always the same, so knowledge of the former doesn’t necessarily translate to the latter.
Figure 4 Reference evaporation is designated as ET in this manuscript, however, in this figure PET is used?
Discussion
My understanding is that this was based on CMIP5 data, how does CMIP5 data compare to CMIP6 data for this region? A brief discussion would be useful to put results of this manuscript into context of more recent CMIP6 data.
Ln 520 Yes this is true for mid-to-high flows there is more uncertainty in future climate inputs but for low-flows it has been found that there is more uncertainty in the hydrological models than climate inputs e.g. See Petheram et al. (2012), Teng et al. (2012),
References
Petheram C, Rustomji P, McVicar TR, Cai WJ, Chiew FHS, Vleeshouwer J, Van Niel TG, Li LT, Creswell RG, Donohue RJ, Teng J, and Perraud J-M (2012) Estimating the impact of projected climate change on runoff across the tropical savannas and semi-arid rangelands of northern Australia. Journal of Hydrometeorology. 13(2), 483-503, doi:10.1175/jhm-d-11-062.1; (IF 3.573; GSC: 5).
Teng J, Vaze J, Chiew F, Wang B, Perraud J-M (2012) Estimating the relative uncertainties sourced from GCMs and hydrological models in modelling climate change impact on runoff. Journal of Hydrometeorology 13(1), 122-139, doi: https://doi.org/10.1175/JHM-D-11-058.1