Probabilistic evaluation of slope performance model error using field and reduced-scale tests

Geotechnical performance prediction is affected by errors associated with a prediction model, such as uncertainties in constitutive behavior, insufficient model representation and boundary conditions. The input parameters are also influenced by spatial variability, and errors from sampling, testing and other sources. Hence, performance predicted from models often deviates from reality. Past reliability analyses primarily focused on the effect of input parameter uncertainties. Model uncertainties were recognized but not sufficiently evaluated due to inadequate information or lack of methodology to utilize the available information. Yet information ranging from historic performance records to specific laboratory or field tests may be available, which could shed light on the bias and error of the prediction model. This research project will develop a methodology for probabilistic

characterization of model uncertainties by effectively incorporating available pertinent information. Knowledge of the model error statistics would provide essential input for a more complete and realistic evaluation of performance reliability. Moreover, relative levels of model uncertainties between prediction models could facilitate the choice of model for analysis. The formulation of the framework will be heavily based on Bayesian methodology. The prediction model for slope stability will include a crude limit equilibrium analysis and a coupled stress-deformation analysis. Performance data will be drawn from the extensive slope performance records of the Geotechnical Engineering Office of the Hong Kong SAR

and a number of large-scale landslides in the Three-Gorges Reservoir Zone. Results from specific field and centrifuge tests will also be utilized. The relative effectiveness of landslide case histories, field tests and reduced-scale model tests in terms of model calibration and error reduction will be investigated. Results will provide valuable information for achieving safer and more cost effective geotechnical design and planning of test programs. Both the engineering community and academia will benefit.