| crossing | now | in 30m typical error 5m | in 60m typical error 6m | in 90m typical error 6m |
|---|---|---|---|---|
| peace arch | 0–5m | 0m | 0m | 0m |
| lynden | 0m | 0m | 0m | 0m |
| pacific highway | 0–5m | 2m | 3m | 3m |
| sumas | 5–7m | 6m | 4m | 3m |
live sources
| crossing | WSDOT | CGDW |
|---|---|---|
| peace arch | 5m 10m ago | 0m 10m ago |
| lynden | — 10m ago | 0m 10m ago |
| pacific highway | 5m 10m ago | 0m 10m ago |
| sumas | 5m 10m ago | 7m 10m ago |
how this works
Borderwatch answers "which border crossing should I take": it predicts wait at Peace Arch, Pacific Highway, Lynden/Aldergrove, and Sumas 30/60/90 minutes out and ranks all four against each other, instead of leaving you to compare live camera images or a raw number per crossing yourself.
Predictions come from a LightGBM model trained on CGDW's 5-minute-resolution archive (2023–present) for all four Cascade Gateway crossings, using time of day, day of week, US and Canadian holidays, school breaks, each crossing's recent trend, and cross-crossing diversion patterns (a busy Peace Arch predicts a busier Pacific Highway soon after, since drivers divert) as inputs. One model per horizon (30/60/90 min), beating both a "current wait" and a "historical median for this hour" baseline at every horizon tested.
"Now" values come directly from CBP, WSDOT, and CGDW's live feeds, not the model — they're shown separately, and as a range when sources disagree, because they measure different things (CBP/CGDW: southbound; WSDOT: northbound).
Typical error is 6–8 minutes depending on horizon, worse at hours flagged above (empirically Saturday afternoon/evening and Sunday evening are hardest to predict). This is a predictive estimate from a hobby project, not an official source — verify with CBP or WSDOT directly.