Digital Earth Pacific DEP cloud native approaches to detect mineral extraction
summary
Cloud-native pipelines on Digital Earth Pacific run logistic regression and multiclass random forest with change detection over Sentinel-1/2 (10 m) to flag mining symptoms: bare lands, stockpiles, and water and map expansion (example: Daa Su catchment showing upstream movement outside permit areas). Workflow runs 32 iterations, uses elevation, SWIR and bare‑soil indices as top features, and deploys training, prediction and change detection via Argo/Kubernetes. Results isolate access roads and stockpiles but struggle to disaggregate cyclone-driven clearing and small upstream gullies; next steps add temporal segmentation and sediment‑transport modelling.
AI-written summary · about
chapters (9)
- 0:00Introduction and background
- 2:34PhD research and Fiji focus
- 6:28Remote sensing approaches and challenges
- 8:14Classifiers and multiclass models
- 10:41Change detection results
- 12:27Workflow and model training
- 13:51Digital Earth Pacific applications
- 14:55Next steps and future work
- 15:18Q&A session
related talks
17mFree satellite imagery and DEM analyses enabling NRM in the developing world
31mFrom Radar Echoes To Real-Time Recognition: Deep Learning on Another Planet -An Open Source Workflow
24mEnvironmental Monitoring with Large-Scale Land Cover Classification
33mUsing EOLearn to build a machine learning pipeline
48mTaking the planet’s pulse: AI to empower adaptation to global change
36mDeep Learning - Segmentation, Identification and Classification
45mFrom Radar Echoes To Real-Time Recognition: Deep Learning on Another Planet an Open Source Workflow
21mUnderstanding World food economy with satellite