Washington waters · tested Sep 26, 2026

How accurate is Ebb & Flood?

At 133 of 191 current stations, you see NOAA's own prediction, unchanged. At the other 58, NOAA publishes only slack and max times, and Ebb & Flood estimates the speeds in between. When we tested this method against NOAA's full predictions, 97.3% of estimates were within 0.3 knots.

Tested against NOAA predictions for Sep 26, 2026 to Sep 25, 2027. Other states are tested the same way before they're added.

Predictions, not observations. NOAA predictions come from astronomy and each station's history. They don't include wind, river runoff or air pressure, which can change real currents and tides, sometimes a lot. Ebb & Flood is for planning, not navigation.

Where the numbers come from

NOAA133 of 191 current stationsNOAA publishes speed and direction every 6 minutes. You see NOAA's numbers, unchanged.

Estimate58 of 191 current stationsNOAA publishes only slack, max flood and max ebb. You see those exactly; the speeds in between are estimated.

Estimate121 of 184 tide stationsNOAA publishes only highs and lows. You see those exactly; heights in between are shaped from the nearest full tide station within 40 km.

"Washington waters" means Washington plus Oregon and Canadian stations on shared water: the Columbia River, Haro Strait, Boundary Pass and the Strait of Juan de Fuca. NOAA lists some stations once per water depth; Ebb & Flood counts each location once, at the depth nearest the surface. NOAA also lists 20 current locations with no predictions, mostly where currents are weak and variable.

Current estimates

Between slack and max, the estimate reaches 50% of the max speed a third of the way through and 90% two thirds of the way. Legs over 4.5 hours build more slowly (45% and 85%), and legs over 5.5 hours more slowly still (35% and 75%).

We tested it at all 133 full-prediction current stations, every 30 minutes for a full year: 2,329,178 readings.

97.3%within 0.3 kn of NOAA

0.03 kntypical miss (median)

0.22 kn95% of estimates miss by less than this

Nonetendency to read fast or slow

By time from slack to max

Slack to max	Share of time	Within 0.3 kn	95% within
Under 2.5 h	15%	100.0%	0.07 kn
2.5–3.5 h	36%	99.0%	0.17 kn
3.5–4.5 h	29%	96.9%	0.25 kn
4.5–5.5 h	12%	94.7%	0.31 kn
Over 5.5 h	8%	91.0%	0.40 kn

Station by station

Under 90%9

90–95%16

95–98%20

98–99.5%22

99.5–100%66

Stations by share of estimates within 0.3 kn. Half of all stations are at 99.4% or better.

Long legs are least certain. When slack and max are more than 5.5 hours apart (8% of the time), 91% of estimates are within 0.3 kn.
The strongest passes vary most, though misses there are small next to the speeds involved. Deception Pass (Narrows): 75% within 0.3 kn, typical miss 0.14 kn, at a station that runs up to 8.7 kn; Port Townsend Canal: 77% within 0.3 kn, typical miss 0.12 kn, at a station that runs up to 4.8 kn; The Narrows, North End (west side): 85% within 0.3 kn, typical miss 0.08 kn, at a station that runs up to 4.9 kn.
Direction: flood vs. ebb matched NOAA in 99.999% of readings where the current ran 0.3 kn or faster.

Tide estimates

We treated 62 full tide stations as if NOAA published only highs and lows, shaped the heights in between from the nearest other full station, and compared with NOAA's real heights: 1,085,707 readings over a year.

0.08 ftaverage miss

0.34 ft95% of estimates miss by less than this

91.6%within 0.25 ft

Mixed-tide days. On a few days a year, when the tide lingers near high or low, the shape can be off by more than 1 ft for a few hours. 19 of 62 test stations had at least one such reading in the year; the worst was 3.6 ft. Any tide station can be switched to its nearest full station in the app.

How far ahead you can plan

NOAA returned predictions for every year we tested, through 2050. Because predictions come from astronomy, a trip next summer is predicted the same way as one tomorrow. Ebb & Flood keeps 12 months ready, refreshes them every 90 days in case NOAA revises a station, and saves them for offline use.

How we test

We can't check estimates at the stations that need them, because NOAA publishes nothing to compare against there. So we test the method where NOAA does publish full predictions: we hide the full prediction, estimate from slack and max times alone, and compare every 30 minutes for a year. NOAA's own slack, max, high and low values are always shown exactly. We rerun these tests whenever the method or data changes, and for each new state before it's added.