TY - JOUR
T1 - Parameterisation, calibration and validation of distributed hydrological models
AU - Refsgaard, Jens Christian
N1 - Funding Information:
The author wishes to thank Børge Storm, Anders Refsgaard and other colleagues at the Danish Hydraulic Institute for assistance in the model work and for many inspiring discussions over several years which have greatly contributed to the crystallization of the ideas presented in the present paper. The financial support from the Danish Technical Research Council is acknowledged. The constructive criticism and detailed suggestions of the two referees, Marios Sophocleous and Rodger Grayson, are greatly appreciated.
PY - 1997/11/1
Y1 - 1997/11/1
N2 - This paper emphasizes the different requirements for calibration and validation of lumped and distributed models. On the basis of a theoretically founded modelling protocol, the different steps in distributed hydrological modelling are illustrated through a case study based on the MIKE SHE code and the 440 km2 Karup catchment in Denmark. The importance of a rigorous and purposeful parameterisation is emphasized in order to get as few 'free' parameters as possible for which assessments through calibration are required. Calibration and validation using a split-sample procedure were carried out for catchment discharge and piezometric heads at seven selected observation wells. The validated model was then used for two further validation tests. Firstly, model simulations were compared with observations from three additional discharge sites and four additional wells located within the catchment. This internal validation showed significantly poorer results compared to the calibration/validation sites. Secondly, the validated model based on a 500 m model grid was used to generate three additional models with 1000 m, 2000 m and 4000 m grids through interpolation of model parameters. The results from the multi-scale validation suggested that a maximum grid size of 1000 m should be used for simulations of discharge and ground-water heads, while the results deteriorated with coarser model grids.
AB - This paper emphasizes the different requirements for calibration and validation of lumped and distributed models. On the basis of a theoretically founded modelling protocol, the different steps in distributed hydrological modelling are illustrated through a case study based on the MIKE SHE code and the 440 km2 Karup catchment in Denmark. The importance of a rigorous and purposeful parameterisation is emphasized in order to get as few 'free' parameters as possible for which assessments through calibration are required. Calibration and validation using a split-sample procedure were carried out for catchment discharge and piezometric heads at seven selected observation wells. The validated model was then used for two further validation tests. Firstly, model simulations were compared with observations from three additional discharge sites and four additional wells located within the catchment. This internal validation showed significantly poorer results compared to the calibration/validation sites. Secondly, the validated model based on a 500 m model grid was used to generate three additional models with 1000 m, 2000 m and 4000 m grids through interpolation of model parameters. The results from the multi-scale validation suggested that a maximum grid size of 1000 m should be used for simulations of discharge and ground-water heads, while the results deteriorated with coarser model grids.
UR - http://www.scopus.com/inward/record.url?scp=0030616293&partnerID=8YFLogxK
U2 - 10.1016/S0022-1694(96)03329-X
DO - 10.1016/S0022-1694(96)03329-X
M3 - Article
AN - SCOPUS:0030616293
SN - 0022-1694
VL - 198
SP - 69
EP - 97
JO - Journal of Hydrology
JF - Journal of Hydrology
IS - 1-4
ER -