Feature-Based Geostatistics: An Application to a Submarine Channel Reservoir

Abstract

Characterization of developing submarine channel reservoirs requires geostatistical methods that are capable of reproducing channel facies morphologies. Such deep marine settings are generally characterized utilizing seismic data while in the development phase. Typically, the seismic is analyzed by means of clustering methods to partially interpret facies. Then, facies models are generated using geostatistical simulation methods. Yet, due to the incompatibility between the pixelbased nature of many geostatistical methods and the patterncatching ability of seismic, making the resulting reservoir models satisfy seismic, geological interpretation and actual well-log / core data is an outstanding challenge. Object-based techniques are not viable for this problem because they are difficult to constrain to well-log data or partially interpreted geobodies. This paper attempts to solve the above problem by proposing a novel geostatistical approach making use of socalled "features". Instead of using single pixels, feature-based geostatistics attempts to work directly on templates of pixels, which represent geologically realistic object pieces, thereby decreasing the dimensionality of the pattern matching problem and promoting geology reproduction. In this sense, featurebased geostatistics is a middle-ground approach between pixel-based and object-based methods. The paper includes a submarine channel case study where the seismic is available and analyzed using a clustering method. A feature-based simulation technique is applied to this data. The method manages to reproduce the facies morphologies desired and conditions to data easily at considerable less CPU demand than either pixel-based or object-based methods.

7. Seismic, rock physics, spatial models, and their integration in reservoir geophysics

“…This makes the calcula tion of conditional probabilities numerically more tractable than for continuous properties. In some implementations, the conditioning properties are further compacted in major pat terns to additionally reduce the set of combina tions of conditioning configurations (Arpat et al, 2002). Modern multipoint methods can handle both categorical and continuous proper ties, and the algorithms are based on methods from computer graphics.…”

Evaluation of Gaussian approximations for data assimilation in reservoir models

“…Geostatistical techniques can then be used to generate realizations of reservoir parameters conditioned to static data [8]. In some cases, information concerning geologic facies may also be incorporated [2]. On the other hand, with the aid of downhole permanent sensors, measurements of reservoir flow can be continuously acquired.…”