Uncertainty representation of geotechnical design parameters

Site investigation and the interpretation of site data are necessary aspects of sound geotechnical practice. As such, the characterization of geotechnical variability should play a central role in reliability-based design. This chapter discusses the uncertainties associated with the most basic soil/rock property evaluation task, which is to estimate a design parameter from a field test. The coefficient of variation in the estimate must be a function of the natural variability of the site, measurement error associated with the field test, and transformation uncertainty about the regression line that relates the field data to the design parameter. In addition, soil/rock properties are spatially variable. This autocorrelation (correlation between values measured at different spatial locations for the same property) effect can be quantified, for instance, by the scale of fluctuation. Useful statistical tables and guidelines for the coefficient of variation and the scale of fluctuation derived from a comprehensive survey of soil and rock databases are presented in this chapter. The cross-correlation (correlation between different properties at the same spatial location) effect is discussed in Chapter 4. Stratigraphy is also spatially variable, but this geologic uncertainty is not well studied in the literature at present. The coefficients of variation derived from soil and rock databases may be larger than those encountered in a specific site, because they are applicable in a generic "global'' sense. These generic statistics are useful as prior information in the absence of site-specific data. Measurement error is not site-specific because it is typically related to the equipment, procedure, and operator. Natural variability and transformation uncertainty are potentially site-specific. It is possible to update the statistics for natural variability and transformation uncertainty in the presence of site-specific data using Bayesian methods. However, statistical uncertainties associated with inference from spatially correlated data should be handled carefully.