Create basemap tiles for a custom shape (for reducing the tiles count)

Question

in order to create a world wide tiles (for a basemap) in acceaptable zoom level one must have a huge hd (see in this post). but, most of the world is sea...

so i want to create x levels of the whole world (x is small say 5-10) and then y levels of only the mainland (y is big say 20) and by that save allot of cpu time for the tile creatiion and hd space for saving unneeded tiles.

i use tilemill to create tiles and you can only define a bounding box rectangle.

can anyone recommend a solution (or maybe a different approach?)

thanks!

edit: the tiles are created from osm data

Answer 1

For lower zoom levels, I decided to take the NASA Blue Marble as tile source.

The original is downloadable here: http://visibleearth.nasa.gov/view.php?id=57752

For zoom levels 0 to 4, I took the small one and made the tiles with gdal2tiles:

gdal2tiles --s_srs EPSG:4326 --zoom 0-4 D:\Karten\Geotiff\BlueMarble\bluemarble.tif D:\Tiles\test

Fro zoom level 5 to 7, I took the high-resolution of half the world:

gdal2tiles --s_srs EPSG:4326 --zoom 5-7 D:\Karten\Geotiff\BlueMarble\bluemarble-ost.tif D:\Tiles\test

Gdal2tiles needs a little patch to use the OSM-like tile naming. I described the workflow here: GDAL2Tiles: MapTiles from BSB/KAP are Switched

You only have to do that once, as the world does not change much looking at low zoom levels.

My OSM tiles built with Mapnik start at zoom level 8. This can be defined in generate_tiles.py

To save disk space, I wrote a little recursive batch file to delete empty tiles:

for /R .\ %%N IN (*.PNG) DO if %%~zN LSS 120 del %%N

This works for Windows, you have to test which file size in bytes is best for you.

---

EDIT

Another choice is to set up a bunch of bounding boxes covering the coastlines in a better way than one single bounding box does. This will save creation time as well.

Answer 2

First, you need to have mapnik and its python bindings installed. I guess you do already, since you run TileMill. Make a regular mapnik style out of you cartocss with carto (npm install carto; carto -l project.mml > style.xml) Then, download this script (see the page for required dependencies).

To generate tiles for the whole world, use ./polytiles.py -b -180 -90 180 90 -z 5 10 -s style.xml: your tiles for zooms 5-10 will be put into tiles directory.

To generate tiles for a bbox, specify it with -b option. Alternatively you can create a .poly file with a polygon, or download one from http://download.geofabrik.de (select a region, then find ".poly" link): specify it in -p option.

So for tiles for land.poly the command will be ./polytiles.py -p land.poly -z 11 20 -s style.xml. Run ./polytiles.py -h for a list of all options.

Answer 3

I have done this before in ArcGIS: I made a fishnet of coarse and fine grids, starting at the same origin. Selected and extracted the fine grid and used the extracted fine grid to erase the coarse grid. Then merge and union with the data to be gridded.

I did some cleanup of the erased grids prior to the union, merging two half grids into one etc.. but that's not really necessary.

Answer 4

i just found out that MBTiles format does exactly that:

Maps that cover large areas of solid color like ocean or empty land can contain thousands of duplicate, redundant tiles. As example, a tile in the middle of the pacific ocean might look like this:

At higher zoom levels this could lead to millions of solid blue tiles, all exactly the same. MBTiles solves this problem by referencing tile coordinates to raw imagery. Thousands of tile coordinates could be paired to the same raw image which drastically reduces the filesize required to serve a map of multiple zoom levels.

reference

however, this does not shorten the rendering part itself