BAYESIAN SEQUENTIAL INDICATOR SIMULATION OF LITHOLOGY FROM SEISMIC DATA

摘要

A discretized lithologic model of the subsurface isdefined by a regular array of pixels. Each pixelcorresponds to one of a finite number of possiblelithoclasses such as sand, shale or dolomite. Thelithoclasses are unknown except at a small number ofsparsely distributed control pixels associated withborehole locations. Associated with each pixel there isa multivariate record of seismic attributes that may bestatistically correlatable with the local lithology. AMonte Carlo method is used to simulate the lithoclassspatial distribution by combining the lithologic data atcontrol pixels with the seismic-attribute data records.Using Indicator Kriging, a prior probabilitydistribution of the lithoclasses is calculated for eachpixel from the lithology values at neighboring pixels.The likelihood of each lithoclass is also calculated ineach pixel from the corresponding conditionalprobability distribution of seismic attributes. Aposterior lithoclass probability distribution isobtained at each pixel by multiplying the priordistribution and the likelihood function. The posteriordistributions are sampled pixel-by-pixel to generateequally probable models of the subsurface lithology.