Towards Probabilistic Rainfall Thresholds for Landslide Early Warning System in Yogyakarta and Central Java, Indonesia

Abstract:

A set of rainfall thresholds based on hourly rainfall for Yogyakarta and Central Java, Indonesia, is proposed in an attempt to improve the existing Landslide Early Warning System (LEWS), which applies national thresholds based on 1-day and 3-day accumulated rainfall. For this purpose, first, we compile a landslide inventory for the study area from the existing databases, which mostly covers unspecific location and date of the events. We update the inventory and add additional information, particularly in terms of the specific location and the estimated time of landslide occurrence. Regarding the rainfall events that have triggered landslides in the study area, we explore various rainfall datasets (measurements from rain gauges and estimates derived from remote sensing). The multiple datasets allow us to compare the characteristics of thresholds derived from each dataset and assess the capability of the remote sensing measurements in capturing the occurrence of rainfall that induce landslides.

The proposed thresholds are estimated from the relation between accumulated rainfall and rainfall duration that triggered landslide occurrences in the study area. These rainfall thresholds are represented by a power-law model that is widely used in defining thresholds for landslide early warning. We apply the frequentist method to derive thresholds with several levels of exceedance probability. This scheme allows us to have probabilistic thresholds, representing different probability of possible landslide occurrence. For future operational purpose, we also explore the usage of the high-resolution numerical weather prediction output in simulating the rainfall inducing the landslides.