Integrating of initial guess models into geostatistical seismic inversion methodologies(Conference Paper)

Iterative geostatistical seismic inversion methodologies allow the inference of the petro-elastic properties of the subsurface (e.g., density, shear and compressional impedances) that are essential during the geo-modelling workflow. The resulting inverted models are always approximations to the real subsurface geology and contaminated by errors and uncertainties that are an intrinsic part of the inverse solution within a geostatistical setting. The main drawbacks of this framework is the relatively high number of iterations needed to reach high correlation coefficients between real and inverted seismic reflection data. On the other hand, deterministic inverse solutions are fast to achieve high convergence values but lack handling uncertainties about the system we aim to model. In this work we introduce a new iterative geostatistical seismic inversion methodology that integrates a deterministic initial guess model as part of the objective function. In this way we can explicitly model the lowfrequencies missing from the seismic reflection data, as in deterministic methodologies, while decreasing the computational time associated with the iterative geostatistical setting. The proposed methodology was applied to synthetic and real case studies and the results show a clear improvements both in the terms of the reproduction of the real subsurface Earth models and the convergence rate between the inverted and real seismic data. When compared with the traditional seismic inversion methodologies, the proposed approach allows retrieving best-fit inverse models that better reproduce the statistical moments as retrieved from the available experimental. © 2016 SEG.