A Non-Stationary Approach for Temporal Probability of Landslide using Hydrometeorological Thresholds

Landslides are one of the natural hazards that endanger life and property. Landslide research emphasises prediction based on the probability of triggering factors such as rainfall for use in early warning systems, and has implications for effective risk mitigation. Recent studies have focused on the probability of a landslide occurrence depending on hydrological factors such as soil moisture. The objective of the current study is to determine the temporal probability of landslide occurrence in a non-stationary framework using hydrometeorological parameters such as soil moisture and rainfall. The study was conducted in the Wayanad district of Kerala, India and area was divided into different zones inorder to account the spatial variation of rainfall and the topographical influence on the soil moisture. The non-stationary temporal probability estimation was performed using the generalised extreme value analysis. The hydrometeorological parameters, gridded rainfall and soil moisture data collected over a 42-year period (1981–2021), were analysed for the non-stationarity characteristics using the statistical tests for trend detection and Pettit test for the change point analysis. A monotonical trend in non-stationarity of the parameters were observed in the different regions of Wayanad. The temporal probability estimation for the future time periods was performed using the bias corrected GCM data and the landslide inventory data. The results showed that the exceedance probability of soil moisture based on the covariates improves the temporal probability of landslides when compared to the rainfall-based approach. The study is a novel and effective method for improving landslide prediction based on hydrological and meteorological factors under changing climate conditions, and for incorporating the same in early warning systems.