RAINFALL-INDUCED LANDSLIDE SUSCEPTIBILITY MODELLING FOR PREDICTION AND EARLY WARNING

Landslide detection and prediction rely on Digital Elevation Models (DEMs) as well as historical documentation of past events. Utilizing multiple elevation datasets and a landslide inventory, logical optimization methods to partition terrain into Slope Units (SUs) used as modelling units for landslide prediction is introduced. Landslide area, counts as well as terrain characteristics like homogeneous slope aspect are used and later unified into segmentation metrics to assess their performance in landslide susceptibility models. These terrain units are continued to be used in dynamic landslide susceptibility modelling through Generalized Additive Models (GAMs) to approach real-time landslide warnings for disaster and risk management. Particularly for rainfall-induced landslides, the integration of rainfall as a dynamic covariate in statistical modelling is approached. Keeping the selection of the mapping unit to SUs, first we utilize cumulative antecedent rainfall information by testing multiple time-windows to select the most-representative temporal bracket for the region. Secondly, while the first technique still remains using rainfall in a scalar form, we implement a functional GAM to utilize the time-series information and present rainfall as a temporally dynamic variable. The model performances undergo a number of validation techniques and the drawbacks of lacking information is explained with varying prediction power. These enhances in methodologies are presented to support Landslide Early Warning Systems (LEWS) for disaster risk and mitigation measures.