America’s passenger rail network used to be enormous. For most of railroad history, our passenger network was by far the largest in the world. We were an example to other nations, and many would say that the railroad was critical in America’s growth and prosperity. Today, our network has shrunk to a fraction of its former size, and our passenger rail quality lags far behind many nations in Asia and Europe.
I’ve always been curious as to just how extensive our network was, so I took it upon myself to build a website to document it. ironroadatlas.com shows what our network looked like every year from 1830 to the present. It’s still very much a work in progress, but it’s already quite entertaining to play with.
At its core, this site is built on massive amounts of railroad data. Over the past 200 years, hundreds, if not thousands of companies operated passenger rail service throughout the United States. Covering all of those lines is a massive undertaking, and some information may even be lost to history. Using AI, I’m working through the lines that are verifiable from data on the internet. Each passenger rail company is its own json file in the repo. The data I’m collecting doesn’t just stop at lat/long coordinates - I’m also trying to figure out how frequent trains were, and how fast they travelled. For now, that data isn’t being rendered on the website, but it would be cool if one day I had enough data to show exactly how long the journey from any two US cities would be in any given year.
This site was possible to create because I have Claude Code. I want to be transparent about the fact that I wasn’t tracing old routes or poring through old timetables to accumulate this data. Claude did all of that for me. A lot of this data was put on the internet by passionate volunteers, and I certainly don’t want to take credit for their work. As such, I’ve put all of the data I’ve collected so far into a public repo that anyone can read and contribute to. Many people have told me that this work is the type of thing that OpenHistoricalMaps would benefit from. I will look into how I can bring this work into that tool, but in the meantime I think making the data open to the public is a good first step.
I’ve laid out some rules for Claude, which also generally regulate what data I want to include in this project:
- Include passenger rail only. Freight railroad is to be excluded. I care about how we moved people around historically because I want to show exactly what we’ve lost.
- Interurban rail and trolleys may be included, but they’re not the first priority. Eventually I’d like to have everything in this dataset, but even just recording all railroads is a herculean task.
- Try to source exact track data from OpenStreetMap where track data is available. It’s ok to approximate using straight lines connected by known points, but this must be labelled and explained in the json.
- Don’t assert what isn’t verifiable. The nature of using AI for a project like this is that there are going to be mistakes. I do want to minimize them, so I’ve added a rule to be conservative and to document any uncertainties. All sources are to be cited as rail lines are added. As AI works through a list of over 1,000 railroads, it can note exactly what sources it used to compile the data, and what parts of the line it wasn’t able to verify. In this way, humans may be able to pick up where AI left off to create a truly comprehensive dataset.
- Every line is to be labelled with the operating company. The map allows users to filter by rail operator, so it’s important that when data is added it’s added to the correct company’s json file.
- No horse-drawn/animal powered rail. Especially for urban trolley systems, many electric/steam powered systems used to be driven by literal horsepower. We have to draw the line somewhere, and this is one line I wanted to draw. Fundamentally, horse-drawn rail is a different technology.
- Every line is to be labelled <Operating Company : Line name as advertised to passengers>.
- We’re only covering rail in the contiguous United States. Again, the line has to be drawn somewhere.
Please check out ironroadatlas.com, and feel free to reach out with any questions or concerns. I’ll again reiterate that this project is far from finished, and it may have inaccuracies (though I’m hoping they’re at a minimum). If for whatever reason you feel compelled to contribute feel free! If you’d like to use the data yourself, go for it!