The goal of this project is to create depth polygons and elevation-encoded PNG tiles for use in Mapbox GL basemaps.
GEBCO 2020 data were downloaded from: https://www.gebco.net/data_and_products/gridded_bathymetry_data/
GEBCO 2020 provides global coverage of elevation data on a 15 arc-second grid.
Blue Earth Bathymetry data were downloaded from: http://www.shadedrelief.com/blue-earth/ on 6/23/2020.
GEBCO Compilation Group (2020) GEBCO 2020 Grid (doi:10.5285/a29c5465-b138-234d-e053-6c86abc040b9)
Blue Earth Bathymetry data were processed by Tom Patterson.
Raster data were first clipped to depths <= 0, so that elevation polygons and grids are not created for terrestrial areas.
GDAL VRTs were used to keep the existing individual datasets from GEBCO throughout; merging the rasters into a single file makes for a very big file (>7 GB).
Blue Earth Bathymetry data were preprocessed using bathymetry/preprocess_blue_earth.py.
This uses the Mapzen terrarium encoding in create_tiles.py.
Elevation tiles for zooms 0 - 10 take about 9.5 hours to create. Currently in Mapbox GL JS, these only render correctly to about zoom 7. There appears to be linear chatter in the derived elevation output; we don't know yet if this comes from the underlying GEBCO data or the WarpedVRT resampling used to create the tiles.
TODO: this might come from using cubic; bilinear may be a better option according to https://tilemill-project.github.io/tilemill/docs/guides/terrain-data/
EXPERIMENTAL: Convert blue earth depth data to hillshade (recommendation of z value comes from: http://www.shadedrelief.com/blue-earth/).
TODO: project to 3857 first, then adjust scale (-s) to balance out level of detail.
Using scale of 1000 highlights the steepest gradients, and thresholded at 175 does a reasonable job of accenting major features without too much noise. (fine detail can then be dropped from this). Also scale at 100 and thresholded at 50,100 also does a nice job of level of detail.
gdaldem hillshade -az 335 blue_earth.tif -z 1000 -s 100 blue_earth_hillshade.tifThresholding this at approx 50, 100, 150 seems to separate out sides of a slope reasonably well.
See also: https://gis.stackexchange.com/questions/144535/creating-transparent-hillshade