Sensitivity analysis in model calibration: GSA-GLUE approach (Q5938255)

This paper addresses the problem of model calibration to observed data for distributed models. Generally speaking, mathematical models are characterized by a certain degree of uncertainty, resulting both from uncertainty in modeled processes and observations errors, and the structural and numerical errors of the mathematical model. The generalized likelihood uncertainty estimation (GLUE) technique used is here an extension of the generalized sensitivity analysis (GSA) methodology. The new approach works with multiple sets of factors, typically via Monte Carlo sampling, and applies likelihood measures to estimate the predictive uncertainty of the model. Model realizations are weighted and ranked on a likelihood scale via conditioning on observations and the weights are used to formulate a cumulative distribution of predictions. This new methodological is applied to real examples.