抄録
Most geological systems are three-dimensional in their spatial aspect, anisotropic, and poorly sampled, and the general reconstruction problem for such systems is to determine from limited data the forms of the particular objects that make them up, and their interrelationships. This must involve first determining the topology of each object, and then its geometry. Real geological data, however, are never good enough to eliminate all but the actual topology and geometry of any object, and so only its most likely topology and geometry may be predicted, on the basis of the available data. Subsurface horizons can be reconstructed from seismic and well data by first subdividing the subsurface into a set of triangular prisms controlled by these data, and then using a random walk technique within each triangle to determine the most likely topology. The geometry of the horizons is then obtained by straightforward surface fitting. This approach has a local, probabilistic basis, and can incorporate anisotropy. It is also computationally very reasonable to implement.
