• pre-stack inversion

This thesis presents an ensemble-based approach to infer the elastic properties from prestack
seismic data. This approach iteratively updates initial ensembles of models based on the misfit
between the predicted and observed data. The initial ensemble is generated using a Multivariate
Normal Random Numbers function and under the assumption of a Gaussian prior. The optimization
procedure is driven by Ensemble Smoother with Multiple Data Assimilation (ES_MDA) an iterative
algorithm that performs a Bayesian updating step for each iteration. Ensemble size has a significant
impact on computational time. In this context, this thesis also assesses the influence of the number
of ensemble members on the quality of the inversion results and the associated uncertainty.
Furthermore, a specific model reparameterization can alleviate the curse of dimensionality issue. In
this work, the 1D Discrete Cosine Transform (DCT), an orthogonal transform, is used to compress
the model parameters. The objective of using DCT is not only to reduce the number of unknown
parameters but also to act as a regularization operator in the model space and allow for the
preservation of the vertical continuity of the elastic properties in the recovered solution. In this
research, the synthetic inversion is assessed over a realistic model mimicking a gas saturated
reservoir hosted in a turbiditic sequence. Later the inversion was run in the compressed model space
through 1D DCT. Finally, the robustness of the inversion in errors in the estimated source wavelet
and different noise levels that contaminated the observed data have been assessed. The results of
the work demonstrate that ensemble-based inversion successfully solves the elastic prestack
inversion and includes the DCT into the proposed algorithm, it significantly speeds up the
computational time of the inversion procedure and ensures that the recovered models retain their
vertical continuity.