Learning Displacement-Robust Representations for Landslide Early Warning under Rainfall Forecast Uncertainty

In three linesLandslide early warning system robust to rainfall forecast uncertainties. Uses Rainfall-Motion-Aware Contrastive Learning (RMCL) to learn representations stable under spatial displacement of rainfall fields. Tested on 2 years of data across Japan (19 regions): 37% higher precision than state-of-the-art baselines.

Summary generated by Claude — human-verified