Where the model says the pollution moves when I-5 southbound closes. Mean change over 8 paired seeds, inner Portland, segments changing by 1 g or more where at least 7 of 8 seeds agree on the direction (the control sets 6, 7 or 8). Base model, mixed fleet, one steady-state hour. The map is the campaign's raw output: the graded predictions are the corridor totals and the station directions. Map colors are NO2 change in grams per segment over the hour (NO2 = 0.30 x NOx, the convention used throughout the project); the corridor bars are NOx, as registered.
Stations: the 13 PORTAL detector stations frozen for the October comparison, colored by the registered direction of change. The two stations south of the I-84 merge have no registered direction. The two upstream approach stations are expected down but are not graded.
Routes: twelve travel-time routes logged hourly since August 18, colored by the registered expectation. Numbers give the frozen October rank of modeled slowdowns (1 = largest): the I-5 to I-405 detour trip first, the closed-span trip second, the Vancouver commute third, then N Interstate, then MLK. Six routes are declared inside seed noise and take no rank.
Registered corridor predictions
Registered corridor predictions: mean percent change in route NOx, 8 paired seeds. I-405, the signed detour: up strongly, all 8 seeds agree, supported. I-205, the regional detour: up weakly, 4 of 8 seeds, not at the bar. The other routes sit inside seed noise. Verdict rule frozen before the run: unanimous sign and |t| > 3. Supported means the simulation bore out an expectation written before it ran; the real-world test is October's.
I-5 southbound closes at the Rose Quarter from September 11, 2026, for up to five weeks. This page shows what a traffic simulation predicted for that closure, registered before it happened.
The predictions were written down and pushed to a public GitHub repository on August 14, 2026, before any simulation task ran (commit f76d27c). Results were appended afterward with the registered sections unchanged (commit 8c185c0).
The model: 16,500 vehicles on Portland's street network, run eight times with different random seeds, each seed once with I-5 southbound open and once with it closed on the same demand. Base driving model, mixed vehicle fleet, one steady-state hour.
Registered predictions: I-405 southbound gains the most, up strongly (+84.9% NOx, all 8 seeds agree, supported). I-205 southbound up weakly (+3.1%, 4 of 8 seeds, not at the bar). The closed span falls to the local-access residual.
Known limits, stated before the closure: the real one-lane local access is not modeled; demand is 2021 commute data plus a fixed 15% through share, with no trip evaporation, time shift, or mode shift; vehicles choose a route once at free-flow times and never replan.
Scoring happens in October against ODOT PORTAL loop detectors and an hourly travel-time log kept since August 18, under rules registered in advance: direction and rank of the changes, not absolute volumes. Observed closure data are not shown here until that scoring is complete, so the predictions cannot be adjusted after seeing them.