Reading (1996) defined facies in two ways: a facies is a body of rock with specified characteristics; and a facies should ideally be a distinctive rock that forms under certain conditions of sedimentation, reflecting a particular process or environment. These two definitions highlight why, when building static and dynamic models of reservoirs, it is important to include facies modelling as a critical part of the process. The distribution of reservoir properties such as porosity, permeability, and clay content, including connectivity and degree of heterogeneity, that define the static and dynamic characteristics depends on the distribution of facies, and the distribution of facies is controlled to a large extent by geology or geological processes. Therefore, geological insight should be used to construct and constrain models of facies distribution, which in turn controls to a significant degree the distribution of reservoir properties. Reservoir models thus derived are more likely to be geologically realistic, certainly compared to a purely geostatistical distribution or other trend-guided interpolation of the same properties. However, geological knowledge is often not sufficient to ensure that the models are representative of the particular reservoirs under study, particularly if well coverage is limited. Seismic data, which are often present over the entire field, provide a means to constrain the property models to

It has long been recognized that the inversion of seismic data can often add valuable information to quantitative interpretations but that the value of this information is susceptible to bias depending on the selected low frequency model required to generate absolute instead of relative elastic properties. Although the theoretical or conceptual link between the quality of seismic inversion results and the input low frequency model is well documented, the impact on quantitative interpretations is often ignored or dismissed in practice. Various examples are presented herein to highlight the significance of low frequency models constructed using varying degrees of sophistication currently seen in practice.

No abstract

No abstract

The interpretation of deterministic seismic inversion results can be greatly enhanced through the generation of lithology probability cubes (e.g., Avseth et al., 2005). This is particularly true when there are more than two lithology types or when the inversion has produced estimates of P-impedance and S-impedance (or some combination of these). One method to generate the lithology probabilities is to derive probability density functions (PDFs) that relate the inverted elastic properties of the rocks to the probability of each lithology type (e.g., Sams and Saussus, 2010). These are applied to the elastic property volumes, within a Bayesian framework, to produce volumes of lithology probability. Ideally the process captures all of the uncertainties in the data, as well as those resulting from the inversion workflow and due to limited seismic resolution. The resultant lithology probability volumes can be visualized and manipulated for improved qualitative interpretation. For example, the sand probability cube in Figure 1a has been derived from the P-and S-impedance volumes shown in Figure 1a and 1b. The probability volume is intuitively easier to understand and therefore interpret than the elastic property volumes, particularly for those who have little insight into the elastic properties of rocks. Thresholds on probability and 3D connectivity criteria can be imposed to provide potential shapes of geological bodies.