Datamaps

This is a tool for indexing large lists of geographic points or lines and dynamically generating map tiles from the index for display.

Dependencies

Modern C compiler like gcc or clang
make
libpng
Ideally a 64 bit machine with >8 GB free memory

Installation

First install make and libpng then type:

make

After the build finishes you will have 4 new command line programs available in the local directory:

encode render enumerate merge

Usage

The basic idea is that if you have a file of points like this:

40.711017,-74.011017
40.710933,-74.011250
40.710867,-74.011400
40.710783,-74.011483
40.710650,-74.011500
40.710517,-74.011483

or segments like this:

40.694033,-73.987300 40.693883,-73.987083
40.693883,-73.987083 40.693633,-73.987000
40.693633,-73.987000 40.718117,-73.988217
40.718117,-73.988217 40.717967,-73.988250
40.717967,-73.988250 40.717883,-73.988433
40.717883,-73.988433 40.717767,-73.988550

you can index them by doing

cat file | ./encode -o directoryname -z 16

to encode them into a sorted quadtree in web Mercator in a new directory named directoryname, with enough bits to address individual pixels at zoom level 16.

You can then do

./render -d directoryname 10 301 385

to dump back out the points that are part of that tile, or

./render directoryname 10 301 385 > foo.png

to make a PNG-format map tile of the data. (You need more data if you want your tile to have more than just one pixel on it though.)

Alternately, if you want an image for a particular area of the earth instead of just one tile, you can do

./render -A -- directoryname zoom minlat minlon maxlat maxlon > foo.png

The "--" is because otherwise getopt will complain about negative numbers in the latitudes or longitudes. For example you could use

./render -A -- dots.dm 12 37.192596 -122.811526 38.070528 -121.702961 > sf.png

to generate an image of the San Francisco Bay Area at zoom level 12 from the encoded data in dots.dm.

The point indexing is inspired by Brandon Martin-Anderson's Census Dotmap. The vector indexing is along similar lines but uses a hierarchy of files for vectors that fit in different zoom levels, and I don't know if anybody else does it that way.

Rendering assumes it can mmap an entire copy of the file into the process address space, which isn't going to work for large files on 32-bit machines. Performance, especially at low zoom levels, will be much better if the file actually fits in memory instead of having to be swapped in.

Merging files

encode will only write to a brand new file. If you want to add data to an existing file, the way to do it is to create a new file with encode and then use merge to combine the old and the new.

$ cat newdata | encode -o new.dm
$ merge -o combined.dm old.dm new.dm

merge also has an option, -u, to eliminate duplicates between the source files while merging them.

Generating a tileset

The enumerate and render programs work together to generate a tileset for whatever area there is data for. If you do, for example,

$ enumerate -z14 dirname | xargs -L1 -P8 ./render -o tiles/dirname

enumerate will output a list of all the zoom/x/y combinations that appear in dirname through zoom 14, and xargs will invoke render on each of these to generate the tiles into tiles/dirname.

You can enumerate a single zoom by specifying both -z and -Z for maximum and minimum. So if you want just z12, enumerate -z12 -Z12.

The -P8 makes xargs invoke 8 instances of render at a time. If you have a different number of CPU cores, a different number may work out better.

If you want to filter the output of render, for example through pngquant to reduce the number of colors, you can do it by having xargs invoke a subshell.

$ enumerate -z8 dirname | xargs -L1 -P8 sh -c 'mkdir -p tiles/dirname/$2/$3; render $1 $2 $3 $4 | pngquant 32 > tiles/dirname/$2/$3/$4.png' dummy

The dummy argument is important because sh -c eats the first argument after the command.

Adding color to data

The syntax for color is kind of silly, but it works, so I had better document it.

Colors are denoted by distance around the color wheel. The brightness and saturation are part of the density rendering; the color only controls the hue.

If you want to have 256 possible hues, that takes 8 bits to encode, so you need to say

encode -m8

to give space in each record for 8 bits of metadata. Each input record, in addition to the location, also then needs to specify what color it should be, and the format for that looks like

40.711017,-74.011017 :0
40.710933,-74.011250 :85
40.710867,-74.011400 :170

to make the first one red, the second one green, and the third one blue. And then when rendering, you do

render -C256

to say that it should use the metadata as 256ths of the color wheel.

Options to render

Input file, zoom level, and bounds

The basic form is

render dir zoom x y

to render the specified tile into a PNG file on the standard output.

-A ... dir zoom minlat minlon maxlat maxlon: Instead of rendering a single tile (zoom/x/y), the invocation format changes to render the specified bounding box as a single image.
-f dir: Also read input from dir in addition to the file in the main arguments. You can use this several times to specify several input files.

Output file format

-d: Output plain text (same format as encode uses) giving the coordinates and metadata for each point or line within the tile.
-D: Output GeoJSON giving the coordinates and metadata for each point or line within the tile.
-T pixels: Image tiles are pixels pixels on a side. The default is 256. 512 is useful for high-res "retina" displays.
-r: Leaflet-style retina, where a request for a tile at zoom level N is actually a request for a quarter of a tile at zoom level N-1. In this case, the quarter-tiles remain 256x256.
-o dir: Instead of outputting the PNG image to the standard output, write it in a file in the directory dir in the zoom/x/y hierarchy. It will also write a basic dir/metadata.json that will be used if you package the tiles with mbutil.

Background

-t opacity: Changes the background opacity. The default is 255, fully opaque.
-w: The default background color becomes white, not black.

-b hex

Specifies hex to be the background color. The default is black (or white, if -w is set.)

-m

Makes the output image a mask: The data areas are transparent and the background is opaque. The default is the opposite.

Color

-c hex: Specifies hex to be the fully saturated color at the middle of the output range. The default is gray.
-S hex: Specifies hex to be the oversaturated color at the end of the output range. The default is white.
-s: Use only the color range leading up to full saturation. The default treats saturated color as the middle of the range and allows the output to be oversaturated all the way to white (or the -S color).

e-n-f / datamaps

readme