We developed ^GASAKe, Genetic Algorithm-based Self-Adaptive Kernel, a new model to predict the time of occurrence of rainfall induced landslides.

^GASAKe predicts the time of occurrence of single landslides or groups of similar landslides, both shallow and deep-seated, using a threshold than when exceeded determines the initiation of the landslides. The triggering threshold is defined using historical information on rainfall and landslides.

^GASAKe is a “black box” model based on the assumption that the stability of a slope depends on rainfall in a linear and stationary way. Compared to other models, ^GASAKe uses a discreet moving window (a “kernel”) and implements a “self adaptive” calibration procedure based on “genetic algorithms”.

Exploiting the discrete kernel, ^GASAKe is flexible and suitable to simulate very complex interactions between rainfall and slope stability. The self adaptive procedure allows ^GASAKe to modify iteratively the shape of the kernel, based on the modelling conditions.

Main outputs of ^GASAKe include a “mobilizing function” that allows to predict the time of occurrence of one or more landslides, and a critical rainfall initiation threshold.

Results

We have successfully applied ^GASAKe to predict medium-size landslides in Calabria (San Benedetto Ullano, Acri, San Fili) and shallow landslides in the Sorrento Peninsula, Campania.

Calibration and validation of ^GASAKe in the different study areas provided encouraging results, that we attribute to the performance and flexibility of ^GASAKe.

The functions obtained by ^GASAKe can be integrated in early warning systems for the possible occurrence of rainfall induced landslides.