Behind the scenes of Mapbox Satellite imagery
summary
Mapbox describes how a global satellite tileset is built: cloud removal composites a cloudless atlas, color balancing uses the open-source Rio Color and an ensemble formula, and seam smoothing uses Zblend so zoom transitions don’t "snap." The pipeline ingests Maxar plus EU and NAIP sources, produces 512px WEBP/PNG tiles via the Pixel Monster engine, and reports having purchased 137 million sq km of 50 cm imagery for global coverage.
AI-written summary · about
chapters (21)
- 0:00Introduction by Chris
- 1:26Meet the raster team
- 5:27History of Mapbox Satellite
- 8:23Early data sources
- 12:17Expanding global coverage
- 14:26Prioritizing updates
- 17:20Why satellite imagery matters
- 18:46Processing imagery pipeline
- 21:13Satellite vs aerial imagery
- 23:17Color adjustment tools
- 25:44Zblend for smooth transitions
- 26:26Ensemble color balancing
- 28:38Mosaicing and compositing
- 29:41Pixel Monster engine
- 31:05Serving tiles with speed
- 32:28Use cases for imagery
- 35:40Feature extraction methods
- 36:43How users interact with tiles
- 38:10Ongoing improvements
- 40:19Beyond true color imagery
- 42:08Q&A session begins
