Every number on this site is computed from one of the sources below. Nothing is hand-entered, and nothing is copied from a report — each one is fetched by a named module in the repository, so a figure can be traced from the page to the code to the file it came out of.
Every original below needs either a 35 MB national download, a PDF read off its bar charts, or a bounding-box query before anybody can see the study area in it. These are the same data after that step — small enough to open in a spreadsheet, and the fastest way to check a number on this site without running the pipeline.
The same files the maps read, unmodified.
Three separate crash files, because no one of them answers the question on its own. FARS has a coordinate on every death and nothing below a death. The state's file has everything and, in published form, no coordinates. The vulnerable-road-user assessment has both, for pedestrians and cyclists only.
A census — not a sample — of every crash on a US public road that killed somebody within thirty days. Used for where people are killed and the per-corridor death rate. 147 crashes, 161 deaths inside the study area, 2018–2023.
Every Uniform Crash Report the state holds — any incident on a public road with a death, an injury, or $500 of damage — published back as a report per county and per town under NMDOT contract. Used for the hourly curves that need no exposure denominator: drink as a share of each hour's crashes, the injury share, and the all-severity hourly shape. 59,849 crashes, 2007–2021 across the three counties. The numbers are read off the bar charts in the PDFs and checked against each report's own Table 1 — 72 checks, all exact.
Federally mandated under the infrastructure act, and New Mexico published the underlying records as an open map service rather than only a PDF. Every pedestrian and pedalcycle crash from 2013 with a latitude, an hour, a KABCO severity and the lighting condition — 758 inside the study area, against FARS's 40 deaths, and unlike those it includes the people who were hit and lived. The companion layer is the state's own High Injury Network: 89 corridor segments here, each with a crash severity index.
Sidewalks, bike lanes and marked crossings on state routes, with a shape, a type and a condition on each. Used to measure how far a crash was from the nearest marked crossing, and to draw the gaps. 2,827 sidewalk runs, 796 bike lane runs, 7,964 crossings inside the study area. State highways only — the arterials that carry the killed-and-seriously-hurt are covered and a residential street is not in it at all, which is why every figure computed from it reports how much of its own evidence sits on a surveyed road.
The road network the simulation runs on: 32,877 edges, 22,760 junctions, 164 traffic lights. Also the street names the walking router reports, and — through Nominatim — the addresses it accepts.
Home-to-work flows at block level, which is the whole of this model's demand. 85,717 commute pairs, 99,580 workers. It is also this model's biggest limitation: LODES knows the drive to work and nothing about freight, shopping, school runs or tourism, which is why the simulation carries about 14% of measured traffic and why its hourly shape has a hole in the middle of the day.
When people leave for work, which shapes the morning peak. The return trip has no equivalent published anywhere, so it is an assumption and is labelled as one wherever it appears.
The measurement the model is calibrated against, and the exposure denominator wherever a road has one. 7,149 count segments intersect the study area; 2,567 carry modelled traffic to compare. State highways only — which is why 12% of the deaths sit on roads with a rate and the rest are drawn as points and not ranked.
The traffic simulator itself — microscopic, open source, maintained by the German Aerospace Center. Everything specific to northern New Mexico is in this repository; SUMO supplies the model and the network format.
Nominatim turns a typed address into a point, bounded to the study area, one request per submitted address and never per keystroke. The basemap is Esri's dark canvas; the road network drawn on top of it is this project's own, and the pages are legible with no basemap at all.
Every dataset above is fetched by the module named beside it, cached on
disk, and re-fetched with --force. Nothing is committed to the
repository except the processed output this site serves. The code is at
github.com/deknapp/otowi
— MIT licensed, and the data keeps the terms of whoever published it.