ABSTRACT
Geostatistical methods have been applied to underground projects for spatial interpretation of geological-geotechnical conditions. A key question, however, is how reliable are such developed models and how does typical site-investigation approaches adopted for tunneling influence the reliability. This paper investigates the performance of the geological (categorical) geostatistical models in predicting specific geologic features critical to tunneling applications. A 3D conditional random field is modeled with actual site-investigation data from a soft-ground tunneling project. The approach is based on extraction of information from a deterministic conceptualization of the sub-surface domain to develop models for varying sampling density. A stochastic simulation algorithm is used to characterize the sub-surface domain. The goodness of the geostatistical models is assessed using metrics derived from confusion matrix. The results indicate that geostatistical model developed with typical site-investigation approaches in well-stratified and sequenced geologic environment is about 70% efficient in capturing complex transitions between geologic units.
