Prediction in Geoenvironmental Engineering: Recommendations for Reliable Predictive Modeling

Rapid increases in processing power, convenient graphical user interfaces, and the commercial availability of sophisticated two and three-dimensional numerical codes have made detailed simulation of engineered geoenvironmental systems practical for many projects. Predictions made with these codes exhibit many of the nuances and perturbations evident in real systems that are stressed by temporally and spatially variable factors (e.g., rainfall or loads). However, these predictions can be very sensitive to the material properties input by the user, the boundary conditions that are applied, and the characteristics of the constitutive relationships that are selected. Examples are used to illustrate the sensitivity of model predictions to each of these factors when they are varied within a reasonable range. The power of inversion for parameterization and assessment of the conceptual model is also discussed. Comparison with field data is shown to be a key step in checking the reasonableness of predictive models.