NASA Logo in the header
Data Science
Operational

Global Landslide Detection for Rapid response (GLiDEr)

Foundation models and High-resolution commercial imagery enable near-real-time detection of new landslides following extreme events.

Launch Date
--
Class
Data/Image Computer Models
Website
--
Key Staffs
  • Program Lead
  • Software Developer
  • Project Scientist

  • Goal: Rapidly map new landslides after disasters to support emergency response, replacing slow manual mapping and overcoming the limits of coarse public imagery.
  • Why a Foundation Model: Landslide labels are expensive and human-error prone (lots of interpretability challenges by SME, see example on the right where SME miss several landslides).
  • What’s New: A change-detection deep learning framework using Planet high-resolution imagery identifies event-induced landslides only, separating new damage from older terrain scars.
  • Technical Innovation: Hybrid architecture combining DinoV3 + diffusion-based terrain reconstruction to improve change detection and reduce false positives. Enables near-real-time disaster mapping, supports rapid response decisions, and delivers an open, globally transferable system for landslide hazards and other change-detection applications.