Try here:

• Globe: ⠀⠀⠀⠀⠀⠀ https://vossr.github.io/Earth-elevation-Web-Mercator/example_minimal-globe
• MapLibre GL JS: https://vossr.github.io/Earth-elevation-Web-Mercator/example_maplibre
• Mapbox GL JS:⠀ https://vossr.github.io/Earth-elevation-Web-Mercator/example_mapbox

JAXA AW3D30 to WGS84 Web Mercator zxy quadtree conversion

• jaxa description
• opentopography description

Data download (220 GB)

• Install aws cli: https://docs.aws.amazon.com/cli/latest/userguide/getting-started-install.html
• aws s3 cp s3://raster/AW3D30/ . --recursive --endpoint-url https://opentopography.s3.sdsc.edu --no-sign-request
• Took 50min@586MB/s
• python3 -m pip install numpy pygeodesy rasterio opencv-python
• download online egm96-5.pgm, EGM96 geoid (18 MB)

Elevation encoding

• The pseudo base-256 rgb24 encoding
• -10000 + ((R * 256 * 256 + G * 256 + B) * 0.1)
• Can runtime decode to float texture, so vertex shader don't need to spam the conversion

Usage

• edit generate_webmercator.py select res and z
• python3 generate_webmercator.py
• python3 generate_lower_levels.py fast combine quadtree 2x2 tiles to upper tiles
• can gdal_translate to .mbtiles (SQLite database caches to RAM)

Conversion speed optimizations

• Precompute heightmap for egm96 (gravitational undulation calculations are slow)
• Operate with lists instead of single elevation samples
• Don't save tiles with no elevation data (image encoder slow)
• Early exit by testing output tile corners for elevation data existence
• multiprocessing
• After optimization image encoder remains as a bottleneck
• Level 9 output size 13GB

License

• JAXA AW3D30 https://earth.jaxa.jp/en/data/policy/
• My code CC0
• Maplibre, Mapbox something other

Image data under jaxa licence: