Using big data and AI to boost geospatial insights for hazard prediction
summary
Natural Resources Canada runs public projects that turn satellite and aerial archives into hazard models: GeoDeep Learning on GitHub publishes code for U‑Net variants, DeepLab v3 (ResNet‑101) and fully convolutional nets, plus transfer learning experiments, losses (cross‑entropy, focal) and optimizers (Adam, SGD, AdaBound) for semantic/instance segmentation and fast inference. The national air photo library holds about 6 million air photos (3M digitized) covering ~60% of Canada, enabling pixel‑level time‑series change detection to predict floods, wildfire spread and invasive‑species trajectories.
AI-written summary · about
chapters (9)
- 0:00Intro
- 1:12Science Leadership
- 2:22Science Impact Unique role as performer, user, funder and communicator of S&T
- 4:40Some Applications of Big Data and Machine Learning
- 8:57GeoDeepLearning - Earth Observation
- 11:31GDL Architecture
- 13:52GDL Setup and Hyperparameters
- 15:52National Air Photo Library - Earth Observation
- 19:49Earth Observation Data Management System
related talks
18mDeep learning for Space imagery applications
33mWhat and where? - Machine learning for geospatial image analysis
46mDeep Learning for Satellite Image Processing, Lessons Learned
58mImproving disaster response with machine learning
11mTerraKit: Standardising AI-Ready Geospatial Data...
48mTaking the planet’s pulse: AI to empower adaptation to global change
01h 36mFrom Maps to Models: An End-to-End Journey in Geospatial Machine Learning with Python
18mGeospatial AI: Where Can AI Help to Tackle Today’s Real-World Challenges?