Statistical rock physics inversion of elastic and electrical properties for CO2 sequestration studies

SUMMARY We present a statistical rock physics inversion of the elastic and electrical properties to estimate the petrophysical properties and quantify the associated uncertainty. The inversion method combines statistical rock physics modeling with Bayesian inverse theory. The model variables of interest are porosity and fluid saturations. The rock physics model includes the elastic and electrical components and can be applied to the results of seismic and electromagnetic inversion. To describe the non-Gaussian behaviour of the model properties, we adopt non-parametric probability density functions to sample multimodal and skewed distributions of the model variables. Different from machine learning approach, the proposed method is not completely data-driven but is based on a statistical rock physics model to link the model parameters to the data. The proposed method provides pointwise posterior distributions of the porosity and CO2 saturation along with the most-likely models and the associated uncertainty. The method is validated using synthetic and real data acquired for CO2 sequestration studies in different formations: the Rock Springs Uplift in Southwestern Wyoming and the Johansen formation in the North Sea, offshore Norway. The proposed approach is validated under different noise conditions and compared to traditional parametric approaches based on Gaussian assumptions. The results show that the proposed method provides an accurate inversion framework where instead of fitting the relationship between the model and the data, we account for the uncertainty in the rock physics model.

Quantitative seismic interpretation of rock brittleness based on statistical rock physics

Rock brittleness is one of the important properties for fracability evaluation, and it can be represented by different physical properties. The mineralogy-based brittleness index (BIM) builds a simple relationship between mineralogy and brittleness, but it may be ambiguous for rocks with a complex microstructure; whereas the elastic moduli-based brittleness index (BIE) is applicable in the field, but BIE interpretation needs to be constrained by lithofacies information. We have developed a new workflow for quantitative seismic interpretation of rock brittleness: Lithofacies are defined by a criterion combining BIM and BIE for comprehensive brittleness evaluation; statistical rock-physics methods are applied for quantitative interpretation by using inverted elastic parameters; acoustic impedance and elastic impedance are selected as the optimized pair of attributes for lithofacies classification. To improve the continuity and accuracy of the interpreted results, a Markov random field is applied in the Bayesian rule as the spatial constraint. A 2D synthetic test demonstrates the feasibility of the Bayesian classification with a Markov random field. This new interpretation framework is also applied to a shale reservoir formation from China. Comparison analysis indicates that brittle shale sections can be efficiently discriminated from ductile shale sections and tight sand sections by using the inverted elastic parameters.

Integrating statistical rock physics and pressure and thermal history modeling to map reservoir lithofacies in the deepwater Gulf of Mexico

We have developed an approach integrating statistical rock physics with pressure and thermal history modeling for quantitative seismic interpretation (QSI). Extending the training data for lithofacies classification and deriving distributions for scenarios not available in the original training require knowledge about geologic processes affecting the elastic properties in the subsurface. We model pressure and thermal history and corresponding smectite to illite diagenesis with a basin model across Thunder Horse minibasin in the Gulf of Mexico. By comparing the mapped lithofacies with and without basin-modeling extrapolations against the results of a reference workflow, we found the value of integrating basin modeling results and statistical rock physics with QSI workflows. The reference workflow uses all available data from two wells in the QSI. The first workflow performs the same lithofacies classification with data from only a single well and does not account for spatial trends away from the well. In the second workflow, we use data from only a single well, the same well as in the first workflow, and bring in extrapolation from the basin and petroleum system modeling at the location of the second well. Results for the first workflow indicate significant differences with the reference workflow in the training data, the quality of the inverted impedance volumes, and the classified reservoir lithofacies. In the second workflow, the guided extrapolation of the training data accounts for spatial trends away from the well and the quality of the impedance inversion significantly improves. The predicted lithofacies map in this scenario shows only minor differences from the reference workflow, and the posterior probabilities of lithofacies show less uncertainty compared with the first workflow. The superiority of the second workflow demonstrates the added value of the integration workflow to QSI in cases of spatially limited well control.