Comparison of different approximation techniques for uncertain time series arising in ocean simulations

Comparaison de différentes techniques d'approximation de séries temporelles incertaines issues d'écoulements océaniques

Pierre Sochala (1) , Mohamed Iskandarani (2)

1 BRGM - Bureau de Recherches Géologiques et Minières (BRGM)
2 University of Miami [Coral Gables]

The analysis of time series is a fundamental task in many flow simulations such as oceanic and atmospheric flows. A major challenge is the construction of a faithful and accurate time-dependent surrogate with a manageable number of samples. Several techniques have been tested to handle the time-dependent aspects of the surrogate including a direct approach, low-rank decomposition, auto-regressive model and global Bayesian emulators. These techniques rely on two popular methods for uncertainty quantification, namely Polynomial chaos expansion and Gaussian processes regression. The different techniques were tested and compared on the uncertain evolution of the sea surface height forecast at two location exhibiting different levels of variance. Two ensembles sizes were considered as well as two versions of polynomial chaos (ordinary least squares or ridge regression) and Gaussian processes (exponential or Matern covariance function) to assess their impact on the results. Our conclusions focus on the advantages and the drawbacks, in terms of accuracy, flexibility and computational costs of the different techniques.

Domains: Modeling and Simulation

Pierre Sochala, Mohamed Iskandarani. Comparison of different approximation techniques for uncertain time series arising in ocean simulations. SIAM Conference on Uncertainty Quantification, Apr 2018, Garden Grove, United States. ⟨hal-01715823⟩