Dynamic rainfall thresholds for landslide early warning in Progo Catchment, Java, Indonesia

Abstract

Abstract This study set out to derive empirical rainfall thresholds for landslides in the Progo Catchment, Indonesia, using high-resolution satellite-based precipitation products (SPPs) and rain gauge data. The SPPs are the gauge-adjusted version of the Global Satellite Mapping of Precipitation (GSMaP-GNRT) and the bias-corrected version of the Climate Prediction Center morphing method (CMORPH-CRT). First, we evaluate the detection capacity and accuracy of each SPP. Then we determine rainfall events responsible for landslides by using a dynamic window that allows us to adapt rainfall events by extending or shortening their duration depending on the rainfall signal persistence. Based on 213 landslides that occurred in the Progo Catchment from 2012 to 2021, we derive multiple rainfall thresholds corresponding to various exceedance probability levels. Results indicate that both GSMaP-GNRT and CMORPH-CRT products fail to capture high-intensity rainfall in the Progo Catchment and overestimate light rainfall compared to rain gauge observations. Nevertheless, when accumulated to define the rainfall threshold, the overall performance of GSMaP-GNRT and gauge-based data in Progo Catchment is comparable. Gauge-based data performed slightly better than GSMaP-GNRT, while CMORPH-CRT performed the worst for all exceedance probabilities. By maximising true skill scores, the suitable exceedance probability for early warning purposes in Progo Catchment can be decided, e.g., 10% (15%) if using gauge-based data (GSMaP-GNRT). These findings can be viewed as an attempt to improve the landslide early warning system in Indonesia. Further study is required, using a numerical weather model that reliably forecasts weather systems producing the rainfall events triggering landslides.