The calibration objective in the large scale applications is not to provide an optimal model for a specific region, but rather to identify a model that achieves best possible performance averaged across many gauged basins at all scales.
E-HYPE was calibrated and evaluated, using a stepwise a multi-basin approach. First, parameters related to snow melt and evapotranspiration were calibrated using satellite remote sensing and in-situ snow data (ESA GlobSnow and Former Soviet Union snow course data, respectively). Runoff generating parameters were then tuned to groups of representative gauged basins for the relevant soil and landuse types for which the parameters were being tuned (116 discharge stations). General routing parameters for river flow and lake dampening were then tuned to the same set of 116 calibration stations. Specific parameters for lake discharge and dam regulation and spillway curves were then calibrated using the nearest discharge stations below lakes, when available, for the largest rivers. Parameters for irrigation were tuned to a group of discharge stations in irrigated areas. Finally, a subbasin specific regionalisation of the parameters across the entire model was made based on links between catchment characteristics and flow signatures identified using cluster analyses and calibration.The years 1980 to 1999 were used for calibration. A further 750 gauged basins were used for independent validation.
Different performance criteria are used here to define the performance of the model towards the observed discharge. The criteria presented here investigate the relative adequacy of the model with regard to timing, variability and volume error. In here, we present model performance at different stations in terms of the Nash-Sutcliffe Efficiency (NSE; Nash and Sutcliffe, 1970), the correlation coefficient, the relative error in mean, the relative error in standard deviation, and Kling-Gupta Efficiency (KGE; Gupta et al., 2009). The optimum value for each criterion (describing a perfect model) is not the same; NSE, KGE and correlation coefficient have optimum value at 1, while the remaining relative error based criteria have optimum value at 0.
The colour of the circle corresponds to a range of model performance, which segments are presented at the histogram plot; depicting the number of stations that have performance within each segment.
The overall model performance in terms of mean annual discharge is presented in the “Simulated versus observed river discharge” plot. A perfect agreement between simulated and observed mean annual discharge would correspond to dots lying on the red 1:1 line. The user can click on a specific dot in the graph, and the selected station will appear in the map. The user can further select one of the “evaluation criteria” and the histogram is automatically updated for the selected criterion. Additional information for individual stations can be presented by Selection of individual stations (simply by left-clicking a circle) would provide additional information about the station (name and upstream area) and basic flow characteristics, i.e. simulated and observed mean discharge. Selecting the “view time-series chart”, the observed (black dots) and modelled (red line) discharge series are presented allowing visual model evaluation.
Nash, J. E., & Sutcliffe, J. V. (1970). River flow forecasting through conceptual models. Journal of Hydrology, 10, 282–290.
Gupta, H. V., Kling, H., Yilmaz, K. K., & Martinez, G. F. (2009). Decomposition of the mean squared error and NSE performance criteria: Implications for improving hydrological modelling. Journal of Hydrology, 377(1-2), 80–91. doi:10.1016/j.jhydrol.2009.08.003