Premium
Bayesian multiresponse calibration of TOPMODEL: Application to the Haute‐Mentue catchment, Switzerland
Author(s) -
Balin Talamba Daniela,
Parent Eric,
Musy André
Publication year - 2010
Publication title -
water resources research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.863
H-Index - 217
eISSN - 1944-7973
pISSN - 0043-1397
DOI - 10.1029/2007wr006449
Subject(s) - calibration , parametric statistics , uncertainty analysis , robustness (evolution) , bayesian probability , sensitivity analysis , bayesian inference , computer science , gibbs sampling , metropolis–hastings algorithm , uncertainty quantification , statistics , mathematics , markov chain monte carlo , biochemistry , chemistry , gene
This paper introduces a general framework that evaluates a numerical Bayesian multiresponse calibration approach based on a Gibbs within Metropolis searching algorithm and a statistical likelihood function. The methodology has been applied with two versions of TOPMODEL on the Haute‐Mentue experimental basin in Switzerland. The approach computes the following: the parameter's uncertainty, the parametric uncertainty of the output responses stemming from parameter uncertainty, and the predictive uncertainty of the output responses stemming from an error term including, indiscriminately in a lumped way, model structure and input and output errors. Two case studies are presented: The first one applies this methodology with the classical TOPMODEL to assess the role of two‐response calibration (observed discharge and soil saturation deficits) on model parameters and output uncertainty. The second one uses a three‐response calibration (observed discharge, silica, and calcium stream water concentrations) with a modified version of TOPMODEL to study the uncertainty of the parameters and of the simulated responses. Despite its limitations, the present multiresponse Bayesian approach proved a valuable tool in uncertainty analyses, and it contributed to a better understanding of the role of the internal variables and the value of additional information for enhancing model structure robustness and for checking the performance of conceptual models.