Exploration of Multi-Decadal Landslide Frequency and Vegetation Recovery Conditions Using Remote-Sensing Big Data

Major landslides caused by earthquakes, typhoons, and heavy rainfall are frequently observed in subtropical areas like Taiwan, presenting considerable risks to human life and the economy. Comprehending landslide frequency and their recovery by analyzing vegetation regrowth is crucial for understanding substantial damage and ecosystem resilience. This study automates landslide detection using temporal NDVI gradient analysis with moving-window smoothing in Taiwan from 1990 to 2022, leveraging Landsat time series data, to comprehensively estimate the frequency of landslide occurrences and vegetation recovery by monitoring the changes in vegetation. Validation is conducted through an analysis of five distinct case studies when compared to the government's landslide inventory. Results show that the landslide frequency and vegetation recovery conditions have been aptly estimated in recent years. About 20% of landslides exhibit multiple remobilization, emphasizing the recurrent nature of these events. Northern Taiwan stood out as a region predominantly characterized by landslide activity in the 1990s, while Central, Southern, and Eastern Taiwan saw a surge in landslides after the 1999 Chi-Chi earthquake and the 2009 Typhoon Morakot. The relations among landslide frequency, topography, and vegetation recovery duration are explored. Landslide frequency displays a positive correlation with elevation and slope. Vegetation recovery varies based on landslide frequency, reflecting the impact of multiple landslide events on growth impediment, whereas lower elevations and gentle slopes exhibit shorter durations for recovery. More than 50% of the vegetation has been recovered to reach the pre-event level in landslides after Typhoon Morakot. This study contributes to the understanding of landslide dynamics, leveraging automated detection and analysis methods for comprehensive regional scales.