Prediction of changes in landslide rates induced by rainfall

S. Bernardie (1), N. Desramaut (1), J.-P. Mallet (2), M. Azib (2), G. Grandjean (1)

(1) BRGM, Orléans, France (s.bernardie brgm.fr),
(2) Institut de Physique du Globe de Strasbourg, CNRS UMR 7516, EOST, Université de
Strasbourg, France

This work focuses on the development of a combined statistical-mechanical approach to predict changes in landslide displacement rates from observed changes in rainfall amounts. The forecasting tool associates a statistical impulse response (IR) model to simulate the changes in landslide rates by computing a transfer function between the input signal (e.g. rainfall) and the output signal (e.g. displacements) and a simple 1D mechanical (ME) model (e.g. visco-plastic rheology) to take into account changes in pore water pressures.
The models have first been applied to forecast the displacement rates at the Super-Sauze landslide (South East France), one of the most active and instrumented clayey landslide in the European Alps. The performance of different combinations of models (IR model, ME model, and a combination of the IR and ME models) is evaluated against observed changes in pore water pressures and displacement rates at the study site. Results indicate that the three models are able to reproduce the displacement pattern in the general kinematic regime with very good accuracy (succession of acceleration and deceleration phases) ; at the contrary, extreme kinematic regimes such as fluidization of part of the landslide mass are not being reproduced : this statement, quantitatively characterised by the Root Mean Square Error between the model and the observations, constitutes however a robust approach to predict changes in displacement rates from rainfall or groundwater time series, several days before it happens. The models have then been applied to other landslides. The variability of the results, depending in particular on the geological context, is discussed.
An automatism tool has finally been developed to daily compute, analyze and compare the predicted data to the observation, in order to automatically deliver alert messages.