Global Vector Terrain

Presented at NACIS 2014 Practical Cartography Day

Transcript

1. Designing global hillshades & landcover as vectors AJ Ashton, Mapbox
2. Intro What is Mapbox Vector Terrain?
4. 3 layers Elevation Contours Hillshading Generalized landcover
6. Vector tiles • Same location & numbering scheme as web mercator image tiles • Contain points, lines, polygons, & data fields • Designed to be turned into images very quickly with any custom stylesheet • OpenGL / WebGL provide even more speed & flexibility
7. Global coverage • 25+ raw data sets compiled into one • 45,000 raw data files
8. Global coverage • USA: US Geological Survey, NASA • Canada: Geobase • Mexico: INEGI • Norway: Kartverket • France: IGN • UK: Ordnance Survey • Denmark: Geodatastyrelsen • New Zealand: Geographx • +More
9. The process, roughly: • Gather the best elevation & landcover raster data sources we can • Normalize & mosaic the various sources, prioritizing by highest resolution • Generate raster hillshade & landcover visualizations • Programatically trace vector polygons from the processed rasters
10. Built on open-source GDAL, PostGIS, Mapnik, GraphicsMagick ⬇ managed by custom Bash & Python scripts
11. Powered by Amazon Web Services • 9,800 jobs managed by Simple Queue Service • Parallel-processed across 250 Elastic Cloud Compute (EC2) virtual machines • Vector tiles uploaded to Simple Storage Service (S3)
12. Hillshading Process & design
13. Classic approach at the foundation • Standard GDAL hillshading algorithm • Traditional lighting from the northwest • Raster input, raster output
14. Accounting for scale • DEMs resampled for each zoom level 2-14 • Hillshading vertical exaggeration set appropriately for each zoom level
15. Accounting for Web Mercator • Mercator 'meters' are only simlilar to real- world meters near the equator • At 60° north or south, 2 mercator meters fit into one real-world meter • This affects slope & hillshade calculations
16. Accounting for Web Mercator
17. Accounting for Web Mercator Horizontal scaling = 1 / cos(latitude)
18. Default scaling (none) Default scaling (none)
19. Latitude-appropriate scaling Latitude-appropriate scaling
20. Raster to vector
21. Raster to vector gdal_polygonize.py with a little help from GraphicsMagick
31. Landcover Process & design
32. Similar to hillshade process • Start by normalizing raw data sources • Reduce rasters to a small number of colors • Trace with gdal_polygonize.py • Simplify to remove pixelization
35. Elevation Contours Process & design
36. Straight-forward • Start with the same DEMs we processed for hillshading • Generate contours with gdal_contour • ✔
37. But why vector terrain?
44. References & Resources • shadedrelief.com by Tom Patterson • reliefshading.com by Bernhard Jenny & Stefan Räber • Cartographic Relief Presentation by Eduard Imhof • gis.stackexchange.com posts by William Huber & many others
45. mapbox.com/mapbox-studio