The application of an innovative inverse model for understanding and predicting landslide movements (Salazie cirque landslides,Reunion Island)

The prediction of landslide movement acceleration is a complex problem, among others identified for deep-seated landslides, and represents a crucial step for risk assessment. Within the scope of this problem, the objective of this paper is to explore a modelling method that enables the study of landslide function and facilitates displacement predictions based on a limited data set. An inverse modelling approach is proposed for predicting the temporal evolution of landslide movement based on rainfall and displacement velocities. Initially, the hydrogeology of the studied landslides was conceptualised based on correlative analyses. Subsequently, we applied an inverse model with a Gaussian-exponential transfer function to reproduce the displacements. This method was tested on the Grand Ilet (GI) and Mare-à-Poule-d’Eau (HB) landslides on Reunion Island in the Indian Ocean. We show that the behaviour of landslides can be modelled by inverse models with a bimodal transfer function using a Gaussian-exponential impulse response. The cumulative displacements over 7 years of modelling (2 years of calibration period for GI, and 4 years for HB) were reproduced with an RMSE above 0.9. The characteristics of the bimodal transfer function are directly related to the hydrogeological functioning demonstrated by the correlative analyses: the rapid reaction of a landslide can be associated with the effect of a preferential flow path on groundwater level variations. Thus, this study shows that the inverse model using a Gaussian-exponential transfer function is a powerful tool for predicting deep-seated landslide movements and for studying how they function. Beyond modelling displacements, our approach effectively demonstrates its ability to contribute relevant data for conceptualising the sliding mechanisms and hydrogeology of landslides.