A Global Spatiotemporal Landslide Dataset with Pre-event Deformation from MT-InSAR

Climate change is contributing to an increase in extreme rainfall events, increasing the need for rapid and reliable landslide detection. Recent landslide datasets have increasingly incorporated multimodal and multi-temporal Earth-observation data; however, the temporal evolution of ground deformation remains insufficiently represented. In this proposal, we aim to develop a standardized ML dataset that integrates MT-InSAR displacement time series with SAR, optical imagery, terrain information, and landslide inventories. The dataset will provide both pixel-based landslide masks and slope-unit-based labels, together with metadata describing observation quality and event characteristics. Using this dataset, we will evaluate whether pre-event deformation information improves ML-based landslide detection across multiple events and environmental conditions. As a secondary task, we will also examine whether the same dataset can support forecasting using only pre-event observations. The proposed dataset will provide a common benchmark for evaluating the contribution of ground-deformation dynamics to landslide detection and for extending detection-oriented ML research toward future forecasting applications.