Density Estimation Framework for Model Error Quantification


The importance of model error assessment during physical model calibration is widely recognized. We highlight the challenges arising in conventional statistical methods accounting for model error, and develop a density estimation framework to quantify and propagate uncertainties due to model errors. The reformulated calibration problem is then tackled with Bayesian techniques. We demonstrate the key strengths of the method on both synthetic cases and on a few practical applications.

Apr 7, 2016
Lausanne, Switzerland