Rainfall Threshold Analysis Using Antecedent Rainfall for Shallow Translational Landslides
Abstract
Rainfall-triggered landslides are a major geohazard in many parts of the world, especially in the developing Asian countries including the Indian subcontinent. The minimum amount of rainfall required to trigger a landslide had been the subject of interest for many years. In a common approach to gauge the rainfall threshold, rainfall intensity is plotted against the event duration. Apart from the intensity-duration threshold, the event-duration threshold, which considers the cumulative rainfall over the event, and event-intensity thresholds are typically used. However, such thresholds consider only the immediately preceding rainfall ignoring the effect of moisture content variation through continuous precipitation. Moreover, the identification of contributing rain gauges is a major source of uncertainty. This study seeks to overcome such disadvantages by utilizing antecedent rainfall, obtained from satellite precipitation measures which are gauge corrected using a process called conditional merging, to arrive at rainfall thresholds. In this study, historical landslide activity in Idukki, a highland district in India that borders the Western Ghats, is considered. The predominantly rain-triggered shallow translational landslides of the area are evaluated for daily vs 1, 2, 3, and 5-day antecedent rainfall activity respectively. The bias in the distribution was assessed in each scenario and it was found to move from 73% towards daily rainfall to 65% towards antecedent rainfall as we progress from 1-day antecedent to 5-day antecedent rainfalls. Assessment of the daily vs 5-day antecedent rainfall showed that an antecedent threshold of 324 mm and a daily threshold of 100 mm were able to correctly predict 100% of the landslide occurrences. However, these thresholds came with a high false positive rate which could be reduced by considering a longer duration of antecedent rainfall. Furthermore, separating the data into different clusters was found to improve the prediction statistics.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFMNH15B..02C