Dodd Narrows · 49°08′N 123°49′W · 2023–2024
Speed over ground through a 90-metre channel, set against the modelled current and against the published reference rate.
A tug under tow has no power margin. It cannot push through a foul stream or hold back in a fair one, so its speed over ground follows the water. Each transit is read twice, once in the throat and once a mile out, and the difference between the two removes the tug's own speed through water exactly: (W + cin) − (W + cout). No current-free reference is needed, which matters because none exists at a tidal pass.
Each dot is one tug transit: its speed in the throat minus its own speed a mile out, which cancels the tug's speed through water and leaves the current. Everything is drawn in the flood frame, so an ebb-bound tug being carried through counts as negative rather than positive. Both curves are the along-channel current built from the harmonic constituents each source reports.
The two curves agree on when almost exactly — their M2 phases differ by 3.5° and the curves correlate at 0.98 across these three weeks. They disagree only on how hard, by a factor of three at the peaks. The dots follow the lower one, at r = 0.78.
662 tug transits that crossed both bands steadily and under way, correlation 0.76. The dashed line is one-to-one, what a perfect current meter traces. The response is not quite straight: it follows the water one-for-one through the middle of the range and falls to about 0.75 past four knots. The offset of −0.81 kn is a tug easing off in the narrows whatever the tide is doing, fitted separately so it stays out of the response.
M2 amplitude along the channel. The measured figure is the slope applied to the model's own amplitude, so it inherits none of the model's level — only its shape. The model sits 18% above it; the published rate sits at 2.2 times it.
Each station's two bands, and what the vessels there report. The quantity being regressed is the difference between the bands, so a station whose bands carry nearly the same current is regressing a small number against its own error and the slope is pulled toward zero. Dodd's outer band is effectively slack — 0.09 kn against 1.71 in the throat — because the throat is a eighteenth of the approach's cross-section. Active Pass, at 1.71 against 1.89, has almost no contrast to work with.
Tugs are used throughout rather than scheduled traffic. The same regression on ferries gives 0.36: a ferry has the power and the reason to make up time against a foul tide, and doing so hides the current from this measurement. The vessels least able to compensate report the most current, which is the signature of behaviour rather than of a model error.
| Station | Transits | Outer / inner | Slope | r |
|---|
Run against the corrected pyramid instead, the same transits give 2.36 kn where the shipped one gives 2.32 — the estimator is stable to 2%. Those two fields differ by only about 5% here, so this is a check that the method is steady, not a demonstration that the answer is independent of the model.
Vessel behaviour can still only subtract, so the measurement stays a floor rather than a reading. But it is a floor the model sits 3% above and the published rate sits at twice. The six other stations agree in direction and prove less, because their geometry dilutes the response.
What is left unexplained is the saturation itself. Past four knots the response falls to about 0.75, and two readings fit equally well: a tug easing for any strong stream, fair or foul alike, or the model running roughly a quarter strong at peak flow. Splitting fair from foul does not separate them — the difference is +0.06 ± 0.11, which is nothing. It does not touch the M2 figure, where the gain is one, but it is an open question about peak-current accuracy.
Half the transits were discarded, on two tests, and neither consults the model. A vessel must be making way at the outer gate: tugs wait off this pass for the tide and then run it, and a tug holding at 1.5 knots before transiting at 10 makes an eight-knot difference that is not the water. Because they wait for a fair tide those spurious differences all fell on one side, which inflated the slope to 0.91. And its speed must be steady within each band, which catches a vessel working its throttle as it crosses while reading nothing but that band's own scatter.
What is deliberately not screened is a slow reading in the throat. Testing the speed through water implied there — sog − current — would keep a tug at one knot when the model claims a foul stream and discard the identical tug at one knot when the model claims slack, retaining agreement and throwing away disagreement. At Dodd that screen changes nothing, because the band-stability test already removes the same transits on model-free grounds. At Calamity Point it is worth +0.10 of slope, a tenth of the whole effect, which is why it is off.
Three further objections, and where they go. Vessels avoid the fast water — the control section here is 90 m wide, so there is nowhere else to be. Surface against depth — a vessel feels the top few metres and the model reports the depth mean, but this domain's 3-D pilot measured that shear at 10–12%, and it runs the wrong way to help. The rate is badly derived — the published events were reconstructed and refitted here with a known signal, and the procedure returns the input exactly in semidiurnal conditions.
What would settle it is a current meter, or the provenance of the harmonic constants behind the published rate.