Abstract

This report discusses the idea and fundamental principles of model calibration under uncertainty and  in particular its place in the uncertainty (UQ) ecosystem. Calibration cannot and does not exist on its own. It is only possible in the presence of adequate uncertainty identification, characterisation and propagation protocols, as well as comprehensive verification and validation, and sensitivity analysis procedures.

Even though calibration appears under many different names and forms, its underlying principles remain unchanged, to use empirical information about the modelled system or object in order to improve the model used by engineers. What is here meant by improve refers specifically to the reduction of model discrepancy, defined and discussed in this report. The topic of calibration is presented from three different perspectives, namely, the Bayesian, the frequentist and the transprobabilistic views. Each perspectives comes with its unique advantages, assumption, and limitations.

These are discussed in detail, together with the most influential methods from within each perspective.