A probabilistic approach for assessing landslide-triggering event rainfall in Papua New Guinea, using TRMM satellite precipitation estimates

摘要

Large and numerous landslides can result in widespread impacts which are felt particularly strongly in the largely subsistence-orientated communities residing in the most landslide-prone areas of Papua New Guinea (PNG). Understanding the characteristics of rainfall preceding these landslide events is essential for the development of appropriate early warning systems and forecasting models. Relationships between rainfall and landslides are frequently complex and uncertainties tend to be amplified by inconsistent and incomplete landslide catalogues and sparse rainfall data availability. To address some of these uncertainties a modified Bayesian technique has been used, in conjunction with the multiple time frames method, to produce thresholds of landslide probability associated with rainfall events of specific magnitude and duration. Satellite-derived precipitation estimates have been used to derive representative rainfall accumulations and intensities over a range of different rainfall durations (5, 10, 15, 30, 45, 60, 75 and 90 days) for rainfall events which resulted in landslides and those which did not result in landslides. Of the two parameter combinations (accumulation-duration and intensity-duration) analysed, rainfall accumulation and duration provide the best scope for identifying probabilistic thresholds for use in landslide warning and forecasting in PNG. Analysis of historical events and rainfall characteristics indicates that high accumulation (>250 mm), shorter duration (<15 days) rainfall events and long duration (>75 days), high accumulation (>1200 mm) rainfall events are more likely to lead to moderate- to high-impact landslides. This analysis has produced the first proxy probability thresholds for landslides in PNG and their application within an early warning framework has been discussed. Crown Copyright (C) 2016 Published by Elsevier B.V. All rights reserved.