Landslides are natural hazards that occur in inclined environments, such as alpine or coastal regions. They have historically caused great damages and losses, while in view of climate change, landslide risk is expected to increase. To mitigate the risk to humans and infrastructure in susceptible areas, reliable early warning systems and temporal forecasting techniques are needed.

The inverse velocity method by Fukuzono (1985a) represents an empirical approach that models instable slope behavior for deriving failure time estimates. Its linear variant is widely used today and provides the theoretical foundation of recently developed, more comprehensive procedures, such as the prospective failure time forecast model (PFTF) by Leinauer et al. (2022). Its nonlinear variants, that describe slope behavior more accurately, find less usage. Hence, in this thesis, the feasibility of extending PFTF into the realm of nonlinear acceleration models was evaluated with regard to forecasting performance in terms of accuracy and uncertainty.

In the analysis of 32 data sets of 13 historic landslide events, the approach of computing failure time estimates by means of nonlinear regression procedures was deemed feasible, but found more complex and less robust. Nonlinear acceleration models improved forecasting accuracy in cases where inverse velocity exhibits convex trends from the beginning of tertiary creep. Moreover, small variations of up to ±2 days in the forecasting starting points had, in general, marginal effects on overall forecast accuracy. Furthermore, the estimation of best-fitting values of parameter α confirmed that linear acceleration models are indeed a suitable choice for describing slope behavior in tertiary creep. In conclusion, it is not reasonable to apply nonlinear acceleration models in isolation, as the inverse velocity method’s linear variant is simpler and more reliable. Instead, their potential lies in augmenting linear fore- casting procedures in applicable cases to improve overall accuracy and uncertainty quantification.