18 Journal of Canadian Petroleum Technology Introduction Rock properties of reservoir engineering interest, such as absolute and relative permeability, formation factor, resistivity index and capillary pressure curves are commonly determined by laboratory core analysis. It is common knowledge, however, that all of these properties are, to a significant extent, controlled by the complex geometry and topology of the pore space (1) . In this respect, proper interpretation of experimental results requires some knowledge of the microstructure. Most importantly, suffi- ciently accurate information on the connectivity and size distribu- tion of pore space channels in reservoir rocks is required for the prediction of their properties using network models (2) . To date, such tools have found limited application as quantitative predic- tors, mainly because of the extreme difficulty in obtaining reliable geometric and topological descriptions of the porous microstruc- ture (3) . Such information is, unfortunately, not accessible by 2D measurements on sections through a porous medium (4) . This fact poses a severe limitation on existing methods for the prediction of petrophysical properties from 2D images of the pore space (5, 6) . Because such methods cannot capture the 3D connectivity of complex solid-void bicontinua, they rely for their predictions on empirical correlations or extensive calibration with experimental data. Recently, computer reconstruction of detailed pore structure data, obtained by serial sectioning of pore casts (7, 8) or magnetic resonance imaging (9) (MRI) of rock samples, has been employed to determine the pore and throat size distributions and pore con- nectivity measures required by network simulators. Unfortunately, the best resolution achievable with MRI to date (ca. 10 μm) is not sufficiently high and serial sectioning remains an extremely labo- rious technique. Thus, neither method seems suitable for routine acquisition of detailed pore structure data. Stochastic simulation of porous media in 3D promises an attractive alternative to serial sectioning or tomographic imag- ing (10) . This technique utilizes statistical information, obtained by analyzing binary images of sections through a sample, to create a stochastic reconstruction of the porous medium in three dimen- sions. The main principle of stochastic simulation is that the model and real microstructures must have identical statistical properties. The statistical properties used as input for the creation of simulated microstructures correspond to the first two moments of the binary phase function Z(r → ), a function taking the value of unity if a point r → in space belongs to the void phase or the value of zero otherwise. In this context, binary images of the pore space are nothing but discrete maps of the phase function Z( r → ) . Such images can be readily obtained by segmenting back-scatter scan- ning electron micrographs of petrographic thin sections. Under the assumption of statistical homogeneity, only the first two moments of the phase function are used as input by stochastic simulation. These are the porosity, φ, and the autocorrelation function, R z (u → ), defined as the following statistical averages: ............................................................................................(1) ...............................................(2) where u → is a lag vector measuring the separation between two points in space. Different variants of stochastic simulation have been recently employed to simulate statistically homogeneous and isotropic microstructures (11-15) . When applicable, this method replaces the arduous task of obtaining serial sections through a porous medium with the task of measuring a small number of sta- tistical properties on binary images of the pore structure. Recent work (16) has demonstrated that statistical information conveyed by the autocorrelation function R z (u → ) can significantly improve the predictive ability of empirical porosity-permeability correlations. Notwithstanding, the question of whether stochasti- cally reconstructed porous media are faithful representations of Z R u Zr Zr u () () ( ) r r r r = - [ ] ⋅ + - [ ] - φ φ φ φ 2 φ= Zr () r Computer Enhanced Core Analysis for Petrophysical Properties M.A. IOANNIDIS, I. CHATZIS, M.J. KWIECIEN University of Waterloo Abstract This paper presents the main results of an effort to integrate 2D image analysis, geostatistics, 3D computer reconstruction, and network modelling for the purpose of creating a new tool for the evaluation of various transport and capillary properties of reservoir rock samples. The new approach utilizes statistical information, obtained from high-resolution binary images of thin-sections, to create a 3D representation of the porous microstructure using stochastic methods. The result is a 3D model porous medium honouring the statistical properties mea- sured in thin sections of the real sample. Availability of such a model permits the determination of geometric and topological attributes of the 3D microstructure which critically control the petrophysical properties of reservoir rocks. Prediction of various properties is then possible by incorporating information about the geometry (pore and throat size distributions) and connectivi- ty (coordination number) of the pore space into network simula- tors. This paper outlines the main features of this new methodol- ogy and presents comparisons of model predictions of perme- ability, formation factor, and resistivity index to experimental data for seven sandstone and carbonate samples from three dif- ferent formations in Western Canada.