Beschreibung:
Rainfall thresholds are considered one of the quintessential components in the monitoring and prediction of rain-triggered landslides. However, rainfall threshold models are affected by uncertainties in the availability and quality of rainfall and landslide data, the selection of the rainfall parameter, the assessment of event duration, the selection of contributing rain gauge, and the objectivity of the model. Moreover, seasonal variations in rainfall patterns are often overlooked while defining rainfall thresholds. In this study, we propose an approach to address these uncertainties, especially the intra-seasonal variations in rainfall patterns. Idukki, a data-sparse region in the Western Ghats, India, is chosen for rainfall threshold development. A database of 250 landslides is compiled from various sources for the period from 2018-2021. With the aid of gauge-corrected satellite precipitation data, the rainfall threshold for this area is modeled separately for two waves of the monsoon season. Critical values for daily and antecedent rainfall were defined to separate extreme and non-extreme rainfall events for both waves. Ordinary least squares and quantile regression models were compared to identify the best model to represent the non-extreme rainfall threshold. ROC analysis was performed to identify the sensitivity, specificity, efficiency, and false positive rates of the models. Moreover, a rainfall threshold was defined for the entire monsoon without subdivision into separate waves, and corresponding ROC statistics were compared with the former approach. The results suggest that considering intra-seasonal variation is essential in rainfall threshold modeling