Diagnostic analysis of distributed input and parameter datasets in Mediterranean basin streamflow modeling

The paper suggests a methodology, based on performance metrics, to select the optimal set of input and parameters to be used for the simulation of river flow discharges with a semi-distributed hydrologic model. The model is applied at daily scale in a semi-arid basin of Southern Italy (Carapelle river, basin area: 506km²) for which rainfall and discharge series for the period 2006–2009 are available. A classification of inputs and parameters was made in two subsets: the former – spatially distributed – to be selected among different options, the latter – lumped – to be calibrated. Different data sources of (or methodologies to obtain) spatially distributed data have been explored for the first subset. In particular, the FAO Penman–Monteith, Hargreaves and Thornthwaite equations were tested for the evaluation of reference evapotranspiration that, in semi-arid areas, represents a key role in hydrological modeling. The availability of LAI maps from different remote sensing sources was exploited in order to enhance the characterization of the vegetation state and consequently of the spatio-temporal variation in actual evapotranspiration. Different type of pedotransfer functions were used to derive the soil hydraulic parameters of the area. For each configuration of the first subset of data, a manual calibration of the second subset of parameters was carried out. Both the manual calibration of the lumped parameters and the selection of the optimal distributed dataset were based on the calculation and the comparison of different performance metrics measuring the distance between observed and simulated discharge data series. Results not only show the best options for estimating reference evapotranspiration, crop coefficients, LAI values and hydraulic properties of soil, but also provide significant insights regarding the use of different performance metrics including traditional indexes such as RMSE, NSE, index of agreement, with the more recent Benchmark Efficiency (Schaefli and Gupta, 2007) and Kling–Gupta Efficiency (Gupta et al., 2009).