Binders Full Of TIGER Deserts


The U.S. has binders full of TIGER deserts.

Let me explain. Back in 2007, we imported TIGER/Line data from the U.S. Census into OpenStreetMap. TIGER/Line was and is pretty crappy geodata, never meant to make pretty maps with, let alone do frivolous things like routing. But we did it anyway, because it gave us more or less complete base data for the U.S. to work with. And so we worked. And worked and worked. Fixing spaghetti highways. Reclassifying and arguing about reclassifying. Connecting and disconnecting. And now, four years and change later, we have pretty maps!

Eat that, TIGER! (check it out)

But we don’t all live in Epcot Center (actually, I don’t think any of us do really) and there’s lots of places where we haven’t been taking as good care of the data. Vast expanses of U.S. territory where the majority of the data in OSM is still TIGER as it was imported all those years ago.

The TIGER deserts.

I want those all to go away, but there’s only so many folks mapping here in the U.S., so we may want to prioritize a little bit. So I wanted to look at the TIGER desert phenomenon a little closer. In particular, I wanted to look a the parts of TIGER deserts that cover towns, and even cities. That’s where people actually do live, and that’s where OSM data is most valuable.

So I set out to identify those the only way I know how: using ArcGIS. The thing is: this kind of job has ‘raster analysis’ plastered all over it, and I just don’t know how to do that using FOSS4G tools. So maybe I’ll just explain what I did in ArcGIS, and then you all can chime in with smart comments about how to do this in PostGIS, R, GRASS, QGIS or whatever free software can do this magic. If you don’t care for all that, just scroll to the bottom for the results.

I created a shapefile first with all OSM highways with TIGER tags in Florida using C++ and osmium. (There’s some good example code to get you started if you’re interested.)

Then, with that loaded into ArcMap, I first created a 5km grid with the predominant (in terms of way length) version number as the cell value.

A second grid for the neighborhood way density:

I reclassified the version grid generously – all cells with 1 or 2 a the predominant TIGER way version got true / 1, the rest false / 0. For distinguishing between built-up and boondock cells, a threshold of 1.8 looked good after some tweaking.

And finally some simple map algebra to combine the two variables into the final result grid:

So there we are folks – TIGER deserts and TIGER ghost towns in Florida:

TIGER deserts in Florida

TIGER ghost towns in Florida

Hmm. I hope we can figure out a way to improve this analysis so the situation does not look quite so bleak. GIS does not lie though – these are the 5km cells that have a reasonably high way density and TIGER way versions that are predominantly 1 or 2.

So let me know folks – 1) Is this a good approach for identifying TIGER ghost towns and if not, what is? and 2) how do you do this using FOSS4G tools?