Geostatistical prediction is applied with a stationary random function (SRF). Stationarity is defined in this thesis as a decision involving five key phases of intervention from the practitioner: (1) choosing domain types, (2) boundary modeling, (3) determining the nature of transitions across boundaries, (4) trend modeling, and (5) predicting with a trend. The framework is a support system for making reasonable decisions of stationarity.;Making a reasonable decision of stationarity is essential for building numerical models with realistic geological heterogeneity. These improved models then lead to improved geological and production uncertainty characterization.;Four new modeling techniques are prototyped within each of the last four phases: (1) boundary modeling with volume functions, (2) near boundary model mixing with a linear mixing model, (3) probability combination schemes for building 3D trend models from lower dimensional trends, and (4) sequential Gaussian simulation with a locally varying transformation to account for the trend. |