Uncertainty quantification of lithofacies trends from seismic and borehole data

Abstract:

The spatial distribution of lithological facies represents a major source of geological heterogeneity. Facies heterogeneity often determines the distribution of rock petrophysical and fluid properties in subsurface reservoirs such as oil & gas, groundwater or geothermal. Quantifying uncertainty in facies modeling therefore plays a critical role to earth resources appraisal and development. However, in most geostatistical modeling, the global trend proportion of facies is usually fixed, thereby potentially underestimating uncertainty. In this work, we developed a data-driven Bayesian approach that quantifies both global and spatial facies uncertainty using seismic and borehole data. The global trend uncertainty is due to the uncertain relationship between seismic data and facies proportions, simply because of the lower resolution of seismic. To quantify this global trend uncertainty for each facies, statistical relationships between seismic responses and facies proportions are first built by learning from Monte Carlo simulations on forward seismic modeling. Such statistical relationships, once conditioned to real acquired 3D seismic, allows generating multiple realizations of facies proportions trends. Then with the sampled posterior facies proportions as trends, a Sequential Indicator Simulation is performed to generate facies models. This approach is applied to a channelized turbidite system. It shows that the generated facies models can preserve the global uncertainties captured by proportion trends while locally matching to borehole facies observations. When compared to conventional approaches that use deterministic trend, it also shows that our statistical learning approach can avoid the problem of underestimating reservoir storage uncertainty by incorporating the trend uncertainty. More importantly, the facies models from the proposed method are less likely to be falsified than those built with deterministic trend.