<![CDATA[Landslides in tropical regions are often triggered by intense rainfall, causing significant impacts. This study assesses the feasibility of rainfall thresholds for landslide early warning in Kapanewon Samigaluh, Kulon Progo. The objectives include characterizing the rainfall regime, testing the relationship between rainfall and landslide events, evaluating spatial consistency with the landslide hazard map, and establishing and verifying operational thresholds. Landslide data from the BPBD and daily rainfall data from BMKG were processed by aligning the dates and performing quality checks, followed by quantitative analysis. The number of landslide events analysed was 197, with rainfall data collected from a single measurement station. Thresholds were set using the median approach for daily rainfall (CH0) and three-day accumulation (CH−2) from the 2014–2023 series. Verification was conducted on 213 days of the 2024 rainy season using Proportion Correct. Characterization shows a consistent monsoonal pattern with notable interannual variability. A positive tendency is observed between annual rainfall accumulation and landslide frequency. Spatially, around 93% of events occur in high-hazard zones. The median-based thresholds obtained are 31 mm for CH0 and 81 mm for CH−2. Operational verification results in PC values of 84.0% for CH0 and 83.6% for CH−2, indicating acceptable performance. Physically, the intensity of rainfall on event days effectively distinguishes landslide from nonlandslide days, while three-day rainfall accumulation increases risk through soil saturation. These findings support the implementation of locally calibrated thresholds for strengthening early warning, with a focus on monitoring during the rainy season. However, the potential for false alarms related to geological conditions and land-use variability may affect model accuracy. Furthermore, periodic recalibration of thresholds is necessary to address uncertainties resulting from changing hydrometeorological conditions and land-use dynamics.]]>
