Uncertainty in geomechanical and rock engineering problems is directly related to the data-limited nature of our field. In situations where varying the numerous input parameters result in substantial variance in output, models are poorly constrained. The inherent uncertainties, which can plague our engineering designs, must be constrained by model calibration.
Numerical model calibration is the process of correlating observations of actual ground behaviour to numerical model output, thereby refining the site-specific strength criterion. Qualitatively calibrated numerical models provide indications of different classes of behaviour under varied loading conditions, while quantitatively calibrated numerical models provide direct indications of likely rock mass stress/strain behaviour and seismic potential. Both approaches facilitate risk management (pertaining to safety and economics); however, quantitatively calibrated numerical models typically have reduced uncertainty, meaning rock engineering designs may be optimized through use of a lower factor of safety.
Quantitative calibration is the supreme goal of numerical back analysis; however, it is exceptionally more difficult to achieve and requires high-quality instrumentation data. Truly quantitative calibration requires instrumentation to be installed sufficiently early relative to the groundwork process so that the complete stress-deformation path is captured without significant gaps. The physical measurements provided by instrumentation must be taken at numerous locations in order to differentiate global-scale rock mass response from localized influences on ground reaction. For these reasons, qualitative or semi-quantitative calibration is often the best that can be achieved in global-scale numerical modelling applications.