ETA accuracy benchmark · 🇭🇰 Hong Kong

How Port Flow's predicted ETA compares to each vessel's own broadcast ETA at Hong Kong, measured on closed voyages over the last 7 days. We publish the number — good or bad. Lower error is better.

Port Flow predicted ETA (RMSE)

22.22 h

MAE 8.56 h · 970 voyages

Broadcast ETA (RMSE)

60.86 h

crew-declared, 301 voyages (same set)

Our advantage

63.5 %

lower error than the broadcast ETA

Broadcast ETAs more than 7 days from the actual arrival are excluded as AIS sentinel / placeholder values (193 excluded). Both figures are computed on the same remaining voyage set — apples to apples.

Closed voyages

error = predicted/broadcast minus actual arrival (hours)


477997254	container	10/12, 05:42 AM UTC	0.57 h	—
413900781	container	10/12, 05:34 AM UTC	-0.25 h	—
477997080	product	10/12, 05:29 AM UTC	-0.01 h	5480.51 h
477997157	product	10/12, 05:25 AM UTC	1.44 h	—
413465120	container	10/12, 05:14 AM UTC	3.56 h	-78.91 h
477995066	product	10/12, 05:11 AM UTC	1.22 h	—
413247850	container	10/12, 04:55 AM UTC	2.88 h	4.08 h
413856426	container	10/12, 04:51 AM UTC	0.44 h	7.64 h
413907171	container	10/12, 04:37 AM UTC	0.62 h	-30.04 h
413865572	container	10/12, 04:32 AM UTC	0.79 h	-2.54 h
477995066	product	10/12, 04:27 AM UTC	1.03 h	—
413465120	container	10/12, 04:24 AM UTC	3.97 h	—
413465410	container	10/12, 04:24 AM UTC	0.43 h	-17.91 h
413901227	container	10/12, 04:17 AM UTC	0.52 h	2992.21 h
477995249	product	10/12, 04:02 AM UTC	2.02 h	—
563161700	general-cargo	10/12, 04:01 AM UTC	0.93 h	-715.52 h
413247850	container	10/12, 03:58 AM UTC	0.81 h	—
477997254	container	10/12, 03:52 AM UTC	1.07 h	—
477997312	container	10/12, 03:49 AM UTC	0.45 h	—
477997254	container	10/12, 03:41 AM UTC	1.09 h	—
352980812	container	10/12, 03:39 AM UTC	2.14 h	-142.66 h
477995065	product	10/12, 03:34 AM UTC	2.22 h	—
477995063	product	10/12, 03:09 AM UTC	1.14 h	—
477997157	product	10/12, 02:59 AM UTC	1.54 h	—
413469660	container	10/12, 02:52 AM UTC	1.56 h	—
477996907	product	10/12, 02:52 AM UTC	1.81 h	6537.13 h
477995065	product	10/12, 02:46 AM UTC	1.32 h	—
412400040	container	10/12, 02:43 AM UTC	0.38 h	-10.72 h
477995137	chemical	10/12, 02:40 AM UTC	0.47 h	—
538008845	container	10/12, 02:38 AM UTC	-11.97 h	-2675.64 h
413865572	container	10/12, 02:33 AM UTC	0.92 h	-0.56 h
413469660	container	10/12, 02:28 AM UTC	0.46 h	-3.97 h
477996397	product	10/12, 02:27 AM UTC	0.76 h	—
477995063	product	10/12, 02:13 AM UTC	1.00 h	—
413762844	container	10/12, 02:10 AM UTC	-0.43 h	-15.17 h
477997157	product	10/12, 02:07 AM UTC	1.33 h	—
413492470	container	10/12, 02:05 AM UTC	0.25 h	-8.10 h
413247850	container	10/12, 02:03 AM UTC	0.64 h	6.93 h
352005274	container	10/12, 02:01 AM UTC	2.64 h	-1.45 h
413865572	container	10/12, 01:59 AM UTC	0.02 h	0.01 h
477995065	product	10/12, 01:56 AM UTC	1.57 h	—
413902744	container	10/12, 01:42 AM UTC	0.30 h	-15.38 h
477997254	container	10/12, 01:39 AM UTC	1.07 h	—
413900781	container	10/12, 01:32 AM UTC	0.27 h	—
477997157	product	10/12, 01:28 AM UTC	0.07 h	—
477996546	container	10/12, 01:22 AM UTC	0.25 h	—
413247850	container	10/12, 01:14 AM UTC	0.06 h	-73.25 h
477996397	product	10/12, 01:14 AM UTC	1.54 h	—
477995338	product	10/12, 01:09 AM UTC	0.58 h	—
477995063	product	10/12, 12:51 AM UTC	0.93 h	—

See it on a live desk

Predicted ETAs, congestion and sanctions screening across 51 ports — try it in one click, no signup.

Open dashboard See pricing →

How it's measured

Predicted ETA: distance-to-berth over speed-made-good, refreshed on every AIS position, with a seasonal correction from rolling medians.
Broadcast ETA: the crew-entered ETA in the vessel's AIS message — the de-facto industry default.
Both are compared to the actual arrival timestamp on the same closed voyages. We report RMSE (penalises large misses) and MAE (typical error).
Broadcast ETAs more than 7 days off are dropped as AIS sentinel / placeholder values, and our model is scored on that same set — so the comparison is honest, not cherry-picked.
Coverage is public AIS only; ports with weak coverage show fewer closed voyages. We'd rather show a small honest sample than a big misleading one.