GNGTS 2019 - Atti del 38° Convegno Nazionale

GNGTS 2019 S essione 3.3 729 ms, where the analytical approach erroneously interprets a finely layered shale-brine sand sequence as a gas saturated layer enclosed in a thick brine sand sequence. Conclusions. We presented a Markov Chain Monte Carlo (MCMC) inversion algorithm for elastic amplitude versus angle (AVA) inversion. The main advantage of the implemented MCMC recipe is that it is suitable for mixed discrete-continuous inverse problems, non-linear forward modellings and multimodal, non-parametric prior distributions. In other terms, our approach does not require any assumptions (i.e. Gaussianity) about the distribution of the continuous properties in a given facies. The method includes geostatistical constraints for the elastic parameters, a 1D Markov prior models for the facies distribution, and use the exact non-linear Zoeppritz equations as the forward modelling. Our implemented MCMC recipe is especially aimed at decreasing the computational effort and it includes multiple chains, a parallel tempering strategy, a delayed rejection updating scheme and hybridize the standard Metropolis-Hastings algorithm with the Differential Evolution Markov Chain method. Our inversion results and the convergence analysis (this analysis is not shown here for the lack of space) demonstrated that the implemented algorithm efficiently samples from a multimodal non-parametric mixture distribution with a reasonable computational effort. Our synthetic inversion experiments proved the importance of correctly modelling the multimodal behavior of Fig. 3 - Facies classification results provided by the analytical (a) and the MCMC (b) inversion. From left to right we represent: True facies profile; MAP facies solution; Estimated posterior pdf of facies.