A generalized method for calibration of parameters in numerical models for landslide runout prediction

Abstract:

The evaluation of runout is a key aspect in hazard and risk assessments of highly mobile landslides, which frequently cause significant loss of life and property. Both, empirical methods and numerical models can be used for predicting runout behavior, with preference for the latter when estimates of the spatial distribution and time evolution of landslide depths and velocities are required, as in the calculation of expected losses and the design of mitigation works. The input material parameters for numerical models can be directly measured in very few situations due to scale problems or to rheologies not being reproducible in experimental conditions. In most cases, these parameters need to be calibrated by back-analyses where runout simulations are fitted to field observations (footprint, maximum velocities, final depths, etc.). This fitting has normally been performed by comparing the observed and simulated runout area visually and only in few cases quantitative comparisons have been made, but still based only in the planimetric area. The present contribution proposes a new method for quantitatively calibrating material parameters in numerical models for landslide runout prediction. The basis for the procedure is the application of classification statistics to observed runout variables (e.g., planimetric area, depths, velocities) and sets of simulations obtained from a range of input material parameters and rheological models. A first version of the method was put forward recently. This consisted in calculating discrete classifiers for a single landslide and performing simulations for a deterministic set of input variables. The method herein presented is a generalized formulation that allows addressing the following situations: (a) calibration for a set of landslide cases; (b) stochastic input variables; (c) multiple rheologies; and (d) uncertainty in the observed runout variables. The generalized procedure is illustrated with case studies of highly mobile landslides from Central America and Norway.