Conceptual model uncertainty in groundwater modeling: Combining generalized likelihood uncertainty estimation and Bayesian model averaging

Rodrigo Rojas, Luc Feyen, Alain Dassargues
2008 Water Resources Research  
1 Uncertainty assessments in groundwater modeling applications typically attribute all sources 2 of uncertainty to errors in parameters and inputs, neglecting what may be the primary source 3 of uncertainty, namely, errors in the conceptualization of the system. Confining the set of 4 plausible system representations to a single model leads to under-dispersive and prone to bias 5 predictions. In this work we present a general and flexible approach that combines 6 Generalized Likelihood
more » ... ty Estimation (GLUE) and Bayesian Model Averaging 7 (BMA) to assess uncertainty in model predictions that arise from errors in model structure, 8 inputs and parameters. In a prior analysis, a set of plausible models are selected and the joint 9 prior input and parameter space is sampled to form potential simulators of the system. For 10 each model the likelihood measures of acceptable simulators, assigned to them based on their 11 ability to reproduce observed system behavior, are integrated over the joint input and 12 parameter space to obtain the integrated model likelihood. The latter is used to weight the 13 predictions of the respective model in the BMA ensemble predictions. For illustrative 14 purposes we applied the methodology to a three-dimensional hypothetical setup. Results 15 showed that predictions of groundwater budget terms varied considerably among competing 16 models, although that a set of 16 head observations used for conditioning did not allow 17 differentiating between the models. BMA provided consensus predictions that were more 18 conservative. Conceptual model uncertainty contributed up to 30% of the total uncertainty. 19 The results clearly indicate the need to consider alternative conceptualizations to account for 20 model uncertainty. 21 22
doi:10.1029/2008wr006908 fatcat:bfmngyvj25gsdfy3l4ayu67sdq