Integrating Local Uncertainty in Geostatistical Seismic Inversion

Traditional iterative geostatistical seismic inversion methodologies are based on two key main ideas: the perturbation of the model parameter space with stochastic sequential simulation and convolution, and the use of a global optimizer based on a genetic algorithm with a cross-over principle. One of the main drawback of these methodologies is when at earlier stages of the iterative inversion procedure it achieves very high local correlation coefficients, which can translate in similar elastic models during the entire iterative procedure.

In this work is propose the extension of traditional iterative geostatistical seismic inversion methodology in order to incorporate a new stochastic sequential simulation methodology that is able to take into account local probability distribution functions as the perturbation technique of the model parameter space. The proposed methodology is able to avoid the use of co-simulation during the iterative procedure increasing the computational efficiency of the inversion methodology, and avoid being trapped in potential local minima due to early high local correlation coefficients. The proposed methodology was successfully applied to synthetic and real datasets to invert fullstack volumes for acoustic impedance models.