Benchmarking Bayesian methods for spectroscopy

Francois Andrieu (1) , Frederic Schmidt (1, 2) , Guillaume Cruz-Mermy (1) , Ines Belgacem (3) , Thomas Cornet (4)

1 GEOPS - Géosciences Paris Saclay
2 IUF - Institut universitaire de France
3 ESA - Agence Spatiale Européenne = European Space Agency
4 Aurora Technology BV

Four inversion strategies have been compared to retrieve surface properties of Europa using a radiative transfer model inversion. This work showed the significant improvements of the multi-chain MCMC method using the SNOOKER sampler, compared to others. This method will be used in future work. In addition, this work stresses the fact that even MCMC sampler are by construction asymptotically sampling the posterior distribution, they may present an apparent converged chain that is actually biased. This study may be extrapolated to other datasets, and other models, with large parameters spaces (at high dimension), and correlated parameters, such as photometric inversions.

Domaines: Planétologie et astrophysique de la terre [astro-ph.EP]

Francois Andrieu, Frederic Schmidt, Guillaume Cruz-Mermy, Ines Belgacem, Thomas Cornet. Benchmarking Bayesian methods for spectroscopy. EuroPlanet Science Congress (EPSC) 2022, Sep 2022, Granada, Spain. ⟨10.5194/epsc2022-502⟩. ⟨insu-03902095⟩