Keywords
hydrological modelling; calibration; evaluation; discharge uncertainty
Location
Session B5: Managing Uncertainty
Start Date
11-7-2016 3:10 PM
End Date
11-7-2016 3:30 PM
Abstract
Rainfall-runoff models are usually defined and assessed comparing simulation outputs to discharge measurements. The latter is calculated from stream stage measurements using stage-discharge relationships. This unavoidable step introduces uncertainties which can be significant especially during high flows. Unfortunately, most of the methods used to evaluate rainfall-runoff models neglect those data uncertainties and overestimate discharge measurement knowledge. The main objective of the paper is to integrate the real information content of discharge time series with a novel modelling errors formulation. Moreover, special attention is given to aggregating modelling errors into a score, such that the prediction accuracy level of each part of the hydrograph, required by the aforesaid assessment, is controlled by the user. The proposed objective function, named Discharge Envelope Catching efficiency (DEC), meets the objectives of: i) taking into account discharge uncertainties, ii) providing a calibration in accordance with hydrological modelling assumptions and end-user expectations. The DEC efficiency is used as an objective function in a novel calibration methodology (DECC). The latter is finally tested to assess a physically-based distributed model dedicated to flash flood modelling. Its performances are compared with results from GLUE (Beven and Binley, 1992) and other calibration methods proposed by Croke (2007) and Liu et al. (2009). GLUE method is taken as a reference while the other tested calibration methods are representative of existing calibration approaches that integrate discharge uncertainties. The DECC method shows the best representation of the confidence interval of observed discharge. The application on the studied case highlights the usefulness of the DECC method for identifying the weaknesses of model structures at reproducing some hydrological processes. These results emphasize the added value of considering discharge uncertainty in calibration and refining modelling errors formulation according to hydrological model aims.
Included in
Civil Engineering Commons, Data Storage Systems Commons, Environmental Engineering Commons, Hydraulic Engineering Commons, Other Civil and Environmental Engineering Commons
Including discharge uncertainties into an adaptable objective function for rainfall - runoff model calibration and evaluation
Session B5: Managing Uncertainty
Rainfall-runoff models are usually defined and assessed comparing simulation outputs to discharge measurements. The latter is calculated from stream stage measurements using stage-discharge relationships. This unavoidable step introduces uncertainties which can be significant especially during high flows. Unfortunately, most of the methods used to evaluate rainfall-runoff models neglect those data uncertainties and overestimate discharge measurement knowledge. The main objective of the paper is to integrate the real information content of discharge time series with a novel modelling errors formulation. Moreover, special attention is given to aggregating modelling errors into a score, such that the prediction accuracy level of each part of the hydrograph, required by the aforesaid assessment, is controlled by the user. The proposed objective function, named Discharge Envelope Catching efficiency (DEC), meets the objectives of: i) taking into account discharge uncertainties, ii) providing a calibration in accordance with hydrological modelling assumptions and end-user expectations. The DEC efficiency is used as an objective function in a novel calibration methodology (DECC). The latter is finally tested to assess a physically-based distributed model dedicated to flash flood modelling. Its performances are compared with results from GLUE (Beven and Binley, 1992) and other calibration methods proposed by Croke (2007) and Liu et al. (2009). GLUE method is taken as a reference while the other tested calibration methods are representative of existing calibration approaches that integrate discharge uncertainties. The DECC method shows the best representation of the confidence interval of observed discharge. The application on the studied case highlights the usefulness of the DECC method for identifying the weaknesses of model structures at reproducing some hydrological processes. These results emphasize the added value of considering discharge uncertainty in calibration and refining modelling errors formulation according to hydrological model aims.
