Improved Modelling of Typhoon-Triggered Landslide Susceptibility in the Philippines Using Single and Multi-Event Landslide Inventories
Abstract
In the Philippines, typhoon-triggered landslides cause hundreds of fatalities and significant economic damage every year. Efforts to manage and mitigate the hazard posed by typhoon-triggered landslides (e.g., via early warning systems) is hampered because the Philippines lacks detailed high-resolution typhoon-triggered landslide susceptibility models. As such, we here use logistic regression techniques implemented alongside a LASSO (Least Absolute Shrinkage and Selection Operator) for variable selection to model typhoon-triggered landslide susceptibility in the Itogon municipality of Luzon, the largest island of the Philippines, using landslide data from two typhoon events in 2018 and 2009. We find that susceptibility models developed using landslide data from a single typhoon event are excellent at predicting independent landslides from the same typhoon event (AUROC values = 0.81 0.82), but have approx. 10% less accuracy when predicting independent landslides from the other typhoon event (AUROC = 0.71 0.72). However, models developed using a combined multi-typhoon-event landslide inventory have a good all-round accuracy at predicting independent landslides from both 2018 and 2009 (AUROC = 0.75 0.81). This suggests that whilst the highest prediction capabilities for a single event are obtained by developing a susceptibility model from a single typhoon season, models developed using multi-event landslide data will be more reliable for predicting future typhoon events. This corroborates previous results from Nepal and Italy, which also show that multi-event or multi-temporal landslide inventories produce more accurate susceptibility models for use in multi-temporal prediction. We thus conclude that multi-typhoon-event landslide susceptibility models are more appropriate for use in landslide early warning systems and for landslide mitigation strategies in the Philippines.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFMNH34A..06J