Automated landslide detection outperforms manual mapping for several recent large earthquakes.

Earthquakes in mountainous areas can trigger thousands of co-seismic landslides, causing significant damage, hampering relief efforts, and rapidly redistributing sediment across the landscape. Efforts to understand the controls on these landslides rely heavily on manually mapped landslide inventories, but these are costly and time-consuming to collect, and their reproducibility is not typically well constrained. Here we develop a new automated landslide detection algorithm (ALDI) based on pixel-wise NDVI differencing of Landsat time series within Google Earth Engine accounting for seasonality. We compare classified inventories to manually mapped inventories from five recent earthquakes: 2005 Kashmir, 2007 Aisen, 2008 Wenchuan, 2010 Haiti, and 2015 Gorkha. We test the ability of ALDI to recover landslide locations (using ROC curves) and landslide sizes (in terms of landslide area-frequency statistics). We find that ALDI more skilfully identifies landslides than published inventories in 10 of 14 cases when ALDI is locally optimised, and in 8 of 14 cases both when ALDI is globally optimised and in holdback testing. These results reflect both good performance of the automated approach but also surprisingly poor performance of manual mapping, which has implications not only for how future classifiers are tested but also for the interpretations that are based on these inventories. We conclude that ALDI already represents a viable alternative to manual mapping in terms of its ability to identify landslide-affected image pixels. Its fast run-time, cost-free image requirements and near-global coverage make it an attractive alternative with the potential to significantly improve the coverage and quantity of landslide inventories. Its simplicity (pixel-wise analysis only) and parsimony of inputs (optical imagery only) suggests that considerable further improvement should be possible. [ABSTRACT FROM AUTHOR]