Accounting for Modelling Errors in Parameter Estimation Problems: The Bayesian Approximation Error Approach
Many parameter estimation problems are highly sensitive to errors. The Bayesian framework provides a methodology for incorporating these errors into our inversion. However, how to characterise the errors in a way that can be efficiently utilised remains a problem in many inversions. Recently the Bayesian approximation error method has been utilised as a systematic way of characterising errors that arise from inaccuracies in the model. We describe the Bayesian approximation error method and demonstrate its use in a homogenisation example. In this example, it is shown that the coarse scale homogenised parameter can be estimated by accounting for the significant modelling error using the Bayesian approximation error method. This modelling error arises from inverting using a model that does not account for the fine scale and has a coarse finite element discretisation.
KeywordsBayesian inversion Modelling errors Homogenisation
- 1.J.P. Kaipio, E. Somersalo, Statistical and Computational Inverse Problems (Springer, New York, 2005)Google Scholar
- 2.J. Koponen, T. Huttunen, T. Tarvainen, J.P. Kaipio, Approximation error method for full-wave tomography. J. Acoust. Soc. Am. 133, 3230 (2013)Google Scholar
- 3.D. Calvetti, P.J. Hadwin. J.M.J. Huttunen, J.P. Kaipio, E. Somersalo, Artificial boundary conditions and domain truncation in electrical impedance tomography. Part II: Stochastic extension of the boundary map. Inverse Probl. Imaging 9(3), 767–789 (2015)Google Scholar
- 4.T.Lähivaara, N.F. Dudley Ward, T. Huttunen, Z. Rawlinson, J.P. Kaipio, Estimation of aquifer dimensions from passive seismic signals in the presence of material and source uncertainties. Geophys. J. Int. 200(3), 1662–1675 (2015)Google Scholar
- 5.R.V. Nicholson, Approaches to multiscale inverse problems. Doctoral thesis, The University of Auckland, Auckland, New Zealand, 2016Google Scholar