Printable Congressional District Maps: Behind The Scenes

Today I’m releasing print-quality maps of congressional districts, with street-level detail and county border lines. This has been one of the most sought-after resources based on emails I’ve received over the last some four years and I don’t think you can find this anywhere else. (At least not comprehensively for the whole nation. Local state clerk’s offices may have them. NationalAtlas.gov has maps but not with very much detail.)

This was a solid 2-day project with less than 300 lines of code and it’s something that only recently became this easy to do. I used Amazon Web Services (AWS), Census TIGER/Line cartographic data in an AWS pubic data snapshot, OpenStreetMap for the street detail in an AWS snapshot prepared by MapBox.com, Mapnik to render the maps (pre-installed on an AWS machine image prepared by MapBox), and the Python modules osgeo (for OGR) and PIL.

Here’s what  I did. This took a lot of trial and error, but in the end the steps were relatively simple.

Setting up the EC2 instance and the OpenStreetMap (OSM) planet data:

• Start up a new Amazon EC2 Linux instance using the AWS machine image (AMI) prepared by MapBox linked above.
• Create Amazon Elastic Block Storage (EBS) volumes for the two data sets (OpenStreetMap and Census TIGER/Line) in the same availability zone as the EC2 instance. If you do it in the AWS console, you’ll just need to search for the snapshots by ID or name (see the links above).
• Attach the two volumes to the running EC2 instance as /dev/sdf (OSM) and /dev/sdg (TIGER).
• Log into the EC2 instance with SSH.
• Mount the two volumes: mkdir /mnt/osm; mount -t ext3 /dev/sdf /mnt/osm; mkdir /mnt/tiger; mount -t ext3 /dev/sdg /mnt/tiger
• Following the MapBox instructions, attach the OSM data to Postgres, change the Postgres configuration to remove password protection, and restart Postgres.

To set up Mapnik, I followed the OpenStreetMap wiki which shows how to reuse their map styling. Most of the steps can be skipped because the data has already been set up in Postgres by MapBox. That involved:

• Getting the OSM Mapnik files from their SVN repository.
• Downloading some extraneous boundary information.
• Create a new style definition that controls how map features are rendered based on the OSM defaults.
• Editing the defaults a) so it actually works, and b) so it looks good at high DPI for printing (increasing font sizes, removing some icons). This took a lot of trial and error since I didn’t understand what was going on and regenerating a map takes some time.

The last step was to write a Python script that invokes Mapnik for each congressional district and generates a high-resolution map image.

• The Census’s TIGER/Line cartographic data has a Shapefile-format file for each state containing the congressional districts in the state. The osgeo/OGR Python module can load the file and tell you the latitude/longitude bounds of the congressional district (among other things).
• Then the Mapnik Python bindings are used to create a new map with the given size, loading in the OSM street data.
• Additional layers are added from the TIGER/Line data for place names (CDPs and county subdivisions if you’re familiar with Census data), county names and borders, state borders (and shading of other states), and the boundaries of the congressional district itself and shading of other congressional districts.
• After rendering the map, which takes ~30 seconds, I used the Python Imaging Library module to add header and footer text with a nice translucent effect.

Generating the maps at three resolutions for all of the congressional districts (except districts at-large) took several hours. I let it run overnight. They’re stored on Amazon S3 (the s3cmd tool is really useful for that).

There’s still room for a lot of improvement. After playing with the style instructions I got too much local road detail that in some places just ruins the whole map at low resolution. And in many places the county names aren’t showing up. Maybe because there’s too much detail. It’ll take some more trial and error to fix.

The source code (which includes all of the preparation steps in detail) is posted here.