A Non-Stationary Framework for Landslide Hazard Assessment Under the Extreme Rainfall Condition

The rate of occurrence and size of shallow landslides may be significantly impacted by alterations in rainfall pattern under changing climate. Thus, it becomes apparent to develop hazard maps instead of susceptibility maps for the effective landslide hazard mitigation. The probability of landslides for the different time horizons is accounted by considering exceedance probability of rainfall as a proxy measure. The stationarity assumption of rainfall data to estimate the temporal probability of landslides is no longer appropriate in a changing environment. In this study, the rainfall non-stationarity for the multiple sites of the Wayanad district of Kerala is taken into account when developing the landslide hazard map (LHM) considering extreme rainfall event of 2018 In the first phase of the preparation of LHM, the landslide susceptibility map (LSM) of the district is developed using Random Forest (RF) model. The probability of exceedance of rainfall threshold was then evaluated using generalized extreme value (GEV) analysis using a novel multi-site rainfall information approach that accounts for antecedent rainfall. The temporal probability of rainfall is then integrated with LSM to develop the LH maps of Wayanad district for the different discrete time horizons up to 50 years. The proposed framework for developing LH map is found to be more robust than the conventional practice of hazard zonation based on LSM under a non-stationary rainfall condition in changing climate scenario.