The Gold Standard of Yachting | Correcting Wind Measurements for Motion at an Offset Sensor

Imagine a rigid platform with a labelled reference point O and a wind sensor S. The sensor is fixed to the platform, but it is ten metres east of O at the instant described by our first drawing.

A velocity assigned to O describes motion at O. It does not automatically describe motion at S when the platform rotates. Two points on a rotating rigid body can have different instantaneous velocities even though their distance apart remains fixed.

This distinction is easy to miss when a record contains one column called boat velocity. Ask which physical point that velocity represents, how it was measured and whether it has already been transformed to another point. An instrument’s location label is part of the mathematical input.

For the initial example, use a local horizontal plane and ignore deformation of the support. All supplied velocities refer to the same instant. We will add three-dimensional motion later. The first task is not to choose a correction formula; it is to put O, S and the direction from O to S on a labelled sketch so each supplied quantity has a location.

Contents · Next chapter

Use east, north and up as a right-handed coordinate system. A vector written (east, north, up) points towards the direction of motion. Thus (4, −3, 0) metres per second means motion eastward and southward.

This is different from a meteorological direction naming where wind comes from. A wind-from bearing must first be converted to a towards-velocity vector before entering the equations here. The existing sailing guide develops that conversion; we will use components directly so it does not distract from sensor motion.

Define positive horizontal angular velocity as counterclockwise when viewed from above. That convention is not the same as a clockwise increase in a compass heading. A heading-rate column cannot be inserted with an assumed sign.

On the sketch, draw a short counterclockwise turn and watch where the east-offset sensor initially moves. It moves north. That physical check will determine the sign of the rotational term more reliably than remembering a cross-product symbol without a direction convention. Every later numerical answer should pass the same simple-turn check.

Contents · Previous chapter · Next chapter

Let W be the local air velocity relative to the fixed ground axes. Let vS be the sensor’s velocity in those same axes. The relative air velocity at the moving sensor is a = W − vS.

Rearranging gives W = a + vS. The operation adds the motion of the point that actually made the relative measurement. If the measured vector is expressed in sensor axes, it must first be rotated into the ground axes before this addition.

For this lesson, a is initially supplied already in east–north–up components. It is a vector measurement in an idealised undisturbed flow, not a raw cup count or a compass angle. Actual instruments require their own calibration and response account.

The sign can be checked with still air. If W is zero and the sensor moves east at one metre per second, a is one metre per second westward. Adding the sensor’s eastward velocity returns zero. Subtracting it again would double the spurious westward component. This boundary check establishes the correction’s direction before rotation introduces another term.

Contents · Previous chapter · Next chapter

First suppose the platform does not rotate. Let the reference-point velocity be vO = (2, 1, 0) metres per second. A rigidly attached sensor has the same velocity because there is no relative rotational motion between the points.

If W = (6, −2, 0), the measured relative air vector is a = (4, −3, 0). Adding vO reconstructs W exactly. The fact that the sensor is ten metres away makes no difference in this translation-only case.

This baseline explains why an incomplete correction can appear to work in some records. If the angular velocity is zero, or if the sensor lies on the relevant rotation axis, the omitted term vanishes. Agreement in one such condition does not test performance during another motion.

Keep the air vector and reference-point velocity fixed for the next example while adding a turn. That controlled change isolates the missing cause. Changing the wind, translation, sensor position and angular velocity all at once would make it harder to see which part of the reconstructed result came from sensor motion.

Contents · Previous chapter · Next chapter

At the stated instant, the sensor is ten metres east of O. Let the angular velocity be 0.10 radians per second counterclockwise about the up axis. The instantaneous tangential speed associated with that offset is angular speed multiplied by radius: 0.10 × 10 = 1 metre per second.

The direction is north at this instant. The sensor’s rotational velocity relative to O is therefore (0, 1, 0). Add that to vO = (2, 1, 0), giving vS = (2, 2, 0).

The offset is a position vector at one instant, not a statement that the sensor remains east of O while the body turns. It rotates with the body. The velocity formula is instantaneous and must use the offset direction corresponding to the same time as the angular velocity.

A paper circle helps make this distinction visible. Mark a point on its east edge and move it a very small angle counterclockwise. Its displacement is initially northward. No operational manoeuvre is needed to establish the geometry, and the invented angular rate is not a recommended vessel motion.

Contents · Previous chapter · Next chapter

For a horizontal offset r = (x, y, 0) and an up-axis angular rate omega, the rotational velocity is (−omega y, omega x, 0). This is the horizontal form of omega-vector cross r.

One derivation writes the rotating point as x = r0 cos(theta), y = r0 sin(theta), with theta measured counterclockwise from east. Differentiating with respect to time gives dx/dt = −omega y and dy/dt = omega x when the distance r0 is fixed.

This derivation explains two useful checks. The velocity is perpendicular to the offset, because their dot product is x(−omega y) + y(omega x) = 0. Its speed is the magnitude of omega times the offset distance.

For r = (10, 0, 0) and omega = 0.10, the result is (0, 1, 0), matching the sketch. If the sensor were ten metres north instead, the term would be (−1, 0, 0), pointing west. These tests help catch a reversed cross-product order: r cross omega has the opposite sign.

Contents · Previous chapter · Next chapter

Keep W = (6, −2, 0) and vO = (2, 1, 0). With r = (10, 0, 0) and angular velocity (0, 0, 0.10), the rotational term is (0, 1, 0), so sensor velocity is (2, 2, 0).

The ideal measured relative air vector is therefore a = W − vS = (4, −4, 0). Adding only the reference translation would give (6, −3, 0), leaving a one-metre-per-second southward error.

Adding the full sensor velocity gives (4, −4, 0) + (2, 1, 0) + (0, 1, 0) = (6, −2, 0), restoring the stipulated air vector.

Every row uses metres per second in common axes. The table is an account of one instant, not an average across a turn. Its purpose is to show which physical contribution repairs the missing component.

Contents · Previous chapter · Next chapter

Suppose the full model is W = a + vO + omega cross r, but an analyst reports Westimate = a + vO. Subtracting the true stipulated W gives estimation error Westimate − W = −omega cross r.

In the first example, that error is (0, −1, 0). The direction and size are predicted by the omitted motion term. This is more informative than saying that the wind estimate is simply noisy.

A residual matching that pattern is evidence consistent with a missing correction, not proof that no other error exists. Wind variation, axis misalignment or flow distortion could also affect a real record. A controlled synthetic test is valuable because the intended input and correct output are known.

For a useful learner check, reverse omega while leaving the instantaneous geometry fixed. The omitted-term error should reverse direction. Double the offset distance and it should double in magnitude under the rigid-body model. A claim that the error stays unchanged under both changes is inconsistent with this proposed explanation and deserves a return to the account.

Contents · Previous chapter · Next chapter

An angular velocity of 0.10 radians per second is about 5.73 degrees per second. If an input is instead 0.10 degrees per second, converting to radians gives about 0.001745 radians per second. At ten metres, the tangential speed is about 0.01745 metres per second, not one.

That is a factor of roughly 57.3, produced by a unit mismatch. The formula omega times distance expects an angular rate in radians per unit time because the derivation uses arc length equal to radius times angle in radians.

Sign needs a separate check. In a horizontal east–north drawing, a clockwise heading increase viewed from above corresponds to negative up-axis angular velocity under our convention. Do not convert the magnitude correctly while retaining an incompatible sign.

A reproducible record stores the rate’s units, axis direction and positive-rotation convention. Labels such as yaw rate or heading rate are useful starting points, but they do not fully specify those conventions across every sensor or software package. Confirm the definition instead of inferring it from a familiar name.

Contents · Previous chapter · Next chapter

Choose a new rigidly attached reference point Oprime four metres east of O at the same instant. Its velocity is vOprime = vO + omega cross d, where d = (4, 0, 0). Thus vOprime = (2, 1.4, 0).

The sensor is now six metres east of Oprime, so rprime = (6, 0, 0). Its rotational term relative to that point is (0, 0.6, 0). Adding vOprime and this term still gives sensor velocity (2, 2, 0).

The reconstructed W is unchanged. Algebraically, rprime = r − d and vOprime = vO + omega cross d. Substituting makes the two d terms cancel.

This invariance is a powerful check. A correct physical reconstruction should not change merely because the analyst relabels the reference point and consistently transforms all associated quantities. If moving the origin changes the answer, inspect whether the translation was updated along with the offset.

The velocity transformation applies to two points fixed on the same ideal rigid body. A reference point moving independently relative to the body would require an additional relative-motion term.

Contents · Previous chapter · Next chapter

Two separate operations can be needed. First rotate the measured relative vector from sensor axes into ground axes. Then add the sensor’s velocity. The coordinate rotation changes how a vector is represented; the motion correction changes which relative velocity is being described.

Let R be the rotation matrix taking sensor components into east–north–up components. If the measurement is asensor, the complete ideal expression is W = R asensor + vO + omega cross r, with the last two terms already in ground components.

For a simple horizontal check, let the sensor’s first axis point north and its second point west. Then a sensor vector (−4, −4, 0) corresponds to ground components (4, −4, 0). With the earlier translation and offset term, it reconstructs (6, −2, 0).

The axis transformation alone would not remove the northward sensor motion. Nor would adding a ground velocity to an unrotated sensor vector be meaningful. Write the coordinate frame beside every vector and test R on its unit axes. That catches an axis error before it becomes hidden inside a final speed.

Contents · Previous chapter · Next chapter

Place a second sensor at r2 = (0, 10, 0), while the first remains at r1 = (10, 0, 0). Keep the same angular velocity and reference translation. Their rotational velocities are respectively (−1, 0, 0) and (0, 1, 0).

If the local air velocity is identical at both sensors, the second relative-air vector is (5, −3, 0), while the first is (4, −4, 0). Their difference is (1, 1, 0), exactly the negative of omega cross (r2 − r1).

Different relative readings therefore need not imply different ground-relative air motion. The sensors themselves can be moving differently. Correcting each with its own velocity restores the same stipulated W.

The uniform-air condition is essential. Real spatial gradients and disturbed flow can make the local air vectors differ. Two sensors cannot automatically separate those effects from motion or calibration. The comparison supplies a model check only when its assumptions are independently justified or when the record is a controlled synthetic example like this one.

Contents · Previous chapter · Next chapter

For general ground-frame vectors omega = (ox, oy, oz) and r = (x, y, z), the rotational velocity has components (oy z − oz y, oz x − ox z, ox y − oy x).

This is a vector cross product, not multiplication of corresponding components. It includes horizontal motion caused by rotation around horizontal axes as well as vertical motion caused by an offset from those axes.

Take r = (10, 0, 8) metres and omega = (0, 0.05, 0.10) radians per second. The rotational velocity is (0.4, 1, −0.5) metres per second. The north-axis rotation contributes eastward motion at the elevated sensor and downward motion at its eastward offset.

The calculation assumes the body is rigid and omega, r and vO refer to the same instant in the same axes. It does not require identifying a particular vessel motion by a colloquial name. If a system supplies body-axis angular components, transform them consistently before using this ground-frame expression.

The physical check remains local: each rotation contribution is perpendicular to its axis and to the appropriate radial displacement. The three-dimensional formula adds those contributions in a common coordinate system.

Contents · Previous chapter · Next chapter

Use the rotational term (0.4, 1, −0.5) from the previous chapter. Stipulate reference translation vO = (2, 1, 0.3) metres per second. The sensor velocity is therefore (2.4, 2, −0.2).

Let the true local air velocity for the synthetic record be W = (6, −2, 0.2). The sensor’s relative measurement is W − vS = (3.6, −4, 0.4). Adding the reference translation alone gives (5.6, −3, 0.7), which is incorrect in all three components.

Adding the full rotational term restores (6, −2, 0.2). The vertical result is particularly instructive: a positive measured relative vertical component does not necessarily mean that the surrounding air has that same upward velocity. The sensor’s own downward motion contributed to the relative measurement.

This remains a kinematic reconstruction inside a stipulated record. It does not establish atmospheric turbulence or a vertical flux from one sample. Such quantities require their own sampling, response and analysis definitions. The calculation shows why correcting a mean horizontal translation is not automatically enough to interpret every component from a moving sensor.

Contents · Previous chapter · Next chapter

The equation uses the local air velocity at the sensor. If the platform, support, sails or surrounding structures alter that local flow, adding the sensor velocity does not recover an undisturbed ambient wind automatically.

Motion correction addresses relative velocity and coordinate representation. Flow distortion addresses a change in the air being sampled. A perfectly executed kinematic calculation can still describe disturbed local air rather than the environmental quantity the analyst wanted.

Primary research on spatially distributed motion measurements documents the importance of alignment and motion information for ship-based wind measurements. The NOAA-hosted autonomous-surface-vehicle study likewise discusses motion compensation in a specific research setting. Neither provides a universal correction for every yacht installation.

For a useful evidence question, ask what independent observation would distinguish a motion error from a local-flow bias. Changing only the coordinate calculation cannot establish the flow around a real structure. That needs suitable calibration, comparison or modelling evidence. Preserve this boundary so a successful synthetic reconstruction is not presented as validation of an entire physical measurement system.

Contents · Previous chapter · Next chapter

Suppose r is recorded while the sensor lies east of O, but omega is taken from a later point in the turn. Combining them can create a rotational velocity that never existed at either time.

A sensor vector also needs the orientation appropriate to its own measurement time. Applying the current orientation to an older relative-wind sample is not the same operation as transforming the original simultaneous sample.

The existing guide to aligning position observations taken at different times develops that time-boundary problem for positions. Here the same discipline applies to several instantaneous vector inputs rather than two location records.

Interpolation is a model of what happened between samples. It may be useful, but its assumptions should be visible. Abrupt motion, filtering delays and unequal sampling rates can limit what the supplied records establish.

A compact synthetic check deliberately shifts one input by one row while leaving the others aligned. Compare the reconstructed residual with the correctly aligned result. The exercise demonstrates sensitivity to timing without claiming an operational sampling rate or a universally adequate delay tolerance.

Contents · Previous chapter · Next chapter

For aligned samples, a mean reconstructed wind is the mean of each complete vector account: mean(R a) + mean(vO) + mean(omega cross r). The operations inside each sample matter.

In general, mean(R a) is not mean(R) multiplied by mean(a), because orientation and relative measurement can vary together. Similarly, mean(omega cross r) is not necessarily mean(omega) cross mean(r).

Consider two equally weighted horizontal samples. In the first, omega = 0.10 and r = (10, 0); in the second, omega = −0.10 and r = (−10, 0). Both rotational terms are (0, 1). Their mean is (0, 1). Yet mean(omega) is zero and mean(r) is zero, whose cross product is zero.

These are two valid instantaneous rigid-body states in an invented record; no continuous trajectory between them is specified. The example proves the algebraic distinction without claiming how quickly a real platform could move between the states.

Averaging speed magnitudes introduces another distinction. The magnitude of a mean velocity vector is not generally the mean of individual speed magnitudes. Decide which quantity the report intends before reducing the component record to one number.

Contents · Previous chapter · Next chapter

Suppose only a and vO are known. The equation W = a + vO + omega cross r contains an unknown motion term. Many combinations of air velocity and rotational sensor motion can reproduce the same supplied relative measurement.

For the horizontal example, a = (4, −4) and vO = (2, 1) are compatible with W = (6, −3) if the rotational term is zero. They are also compatible with W = (6, −2) if the rotational term is (0, 1).

The relative-wind record alone cannot choose between those explanations. More samples of the same incomplete variables may still leave the distinction unresolved unless additional structure is justified.

Useful extra information includes a known rigid offset and simultaneous angular velocity, or a directly established sensor-point velocity in the required axes. A suitable independent air-velocity reference can test the reconstruction. Each route supplies information about a different missing part of the account.

Do not infer an angular correction by forcing the output to look meteorologically smooth and then present that smoothness as independent validation. The same output preference would have helped determine the correction being evaluated.

Contents · Previous chapter · Next chapter

If the angular-speed magnitude is bounded by 0.02 radians per second and the sensor offset magnitude by eight metres, then the rotational-speed magnitude cannot exceed 0.16 metres per second under the rigid-body model.

This follows from the cross-product magnitude: the angular speed times offset distance times the sine of their angle. The sine cannot exceed one. The bound may be conservative when the offset is nearly parallel to the rotation axis.

If the offset lies exactly on that axis, the rotational contribution is zero even with a nonzero angular rate. An elevated sensor is therefore not automatically affected by every component of rotation in the same way.

The 0.16 figure is a conditional geometric bound, not a confidence interval and not a statement of instrument accuracy. It omits uncertainty in the supplied bounds, translation, orientation, local flow and sensor response unless those are separately included.

For a repair exercise, ask a learner to distinguish “the omitted rotational term is at most this magnitude under these inputs” from “the final wind error is at most this magnitude.” The second statement requires an account of every other error source and their possible combination.

Contents · Previous chapter · Next chapter

Set aside the east-offset example. A new sensor lies at r = (0, 6, 0) metres. The angular velocity is (0, 0, −0.20) radians per second, reference translation is vO = (−1, 2, 0) metres per second, and measured relative air is a = (3, −1, 0).

Calculate the rotational term, sensor velocity and reconstructed air velocity. Then compute the result obtained by omitting the rotational term, and state the vector error of that incomplete result.

For an independent reference-point check, move the origin two metres north along the same rigid body. Derive both the new reference translation and the new sensor offset. The final wind must remain unchanged if all transformations are consistent.

Do not begin by copying the signs from the first example. Draw the north-offset sensor and a clockwise turn under the stated convention. Its initial rotational motion is eastward. That sketch supplies a fresh physical check before arithmetic.

For an extension, ask what changes if the sensor is no longer rigidly fixed but slides relative to its mount. Identify the missing velocity term in words before attempting a new formula.

Contents · Previous chapter · Next chapter

The rotational term is (−omega y, omega x, 0) = (1.2, 0, 0). Sensor velocity is (0.2, 2, 0), and reconstructed W is (3.2, 1, 0) metres per second.

Omitting rotation gives (2, 1, 0). Its error relative to the full reconstruction is (−1.2, 0, 0), westward, as predicted by the negative of the omitted term.

For d = (0, 2, 0), omega cross d = (0.4, 0, 0). The new reference velocity is (−0.6, 2, 0). The sensor offset from that reference is (0, 4, 0), whose rotational term is (0.8, 0, 0). Their sum remains sensor velocity (0.2, 2, 0).

A sliding sensor needs its relative motion with respect to the rigid mount, expressed in the same axes, in addition to the rigid-body contribution. If its offset changes in body coordinates, the simple fixed-offset rule is incomplete. Naming that condition shows transfer beyond memorising a cross product: the learner understands which physical motion the original model assumed away.

Contents · Previous chapter · Next chapter

A useful record identifies the measured air vector, its units, axis directions and whether it points towards motion or towards a wind source. It identifies O and S physically, gives the offset direction, and states the reference point associated with the supplied translation.

Angular velocity needs units, component axes and a positive-rotation convention. Orientation needs a clear mapping between sensor and ground components. Each quantity needs a timestamp and a description of any averaging or filtering already applied.

For a synthetic example, retain the stipulated W so residuals can be checked directly. For a real observation, distinguish reconstructed local air velocity from an independently validated environmental measurement. Record known exclusions such as non-rigid motion, flow disturbance and response delay.

The output should include components before a speed or direction summary. Components expose a sign or frame error that a single magnitude can conceal. If an independent reference-point transformation is available, include that invariance check too.

A complete record does not need every possible technical detail. It needs enough specific information to reproduce the stated operation and understand its limits. A missing frame or physical point is often more consequential than another decimal place in a reported speed.

Contents · Previous chapter · Next chapter

Consider this invented claim: “We added the platform’s GPS velocity, so the wind is fully corrected.” The missing question is which point the GPS velocity represents and whether the sensor has additional motion relative to that point.

A more precise explanation might read: “The relative-air vector was rotated into ground axes and combined with translation at the stated reference point plus the rigid-body rotational velocity at the sensor. The result describes local air velocity under the stated alignment, timing and rigidity assumptions.”

That sentence still does not establish undisturbed ambient flow or instrument accuracy. Those require their own evidence. Keeping the claim at its supported level makes the calculation more useful, because another reader can see what to test next.

For practice, give the learner three records: pure translation, rotation about an axis passing through the sensor, and rotation with a perpendicular offset. Ask where the additional term vanishes and where it does not. The repair should follow the geometry in each record rather than add a correction merely because a sensor is mounted high or far from a labelled point.

Contents · Previous chapter · Next chapter

For an early vector route, draw a circle with a reference point and mark sensors east and north of it. Use a small counterclockwise rotation to identify instantaneous directions. Add translation arrows only after those relative directions are clear.

For a secondary mathematics route, derive (−omega y, omega x), check perpendicularity with a dot product and solve the changed geometry. Introduce radians by comparing arc distance with angle, keeping the units beside every calculation.

For an advanced route, implement a three-dimensional cross product and an orthonormal coordinate rotation for synthetic records. Test pure translation, zero offset, an offset parallel to the rotation axis, reversed rotation and a changed reference point. Each test should have a known physical answer, not merely reproduce the same implementation twice.

Repair means returning to the first uncertain sign, frame or point label. Stabilisation means obtaining consistent results across changed geometries. Extension means adding a genuinely new motion or uncertainty source and stating what further evidence is required.

These exercises can be completed with drawings, cards or a spreadsheet. They develop measurement reasoning without becoming a substitute for supervised maritime training or an operational wind-estimation system.

Contents · Previous chapter · Next chapter

Is the correction always angular speed times mast height?

No. The relevant contribution is a vector cross product with the displacement from the stated reference point. An offset parallel to a rotation axis contributes zero for that rotation; a perpendicular offset gives the largest magnitude for its distance.

Can I use heading rate as angular velocity?

Only after confirming its axis, units, sign and relationship to the model. General three-dimensional angular velocity is not supplied by one horizontal heading-rate value alone.

Does changing the reference point change the wind?

A consistent rigid-body transformation changes both translation and offset terms while preserving sensor velocity and reconstructed wind. Changing only the offset creates an inconsistent account.

Does motion correction remove flow distortion?

No. It changes the relative-velocity account. It does not establish that the local air at the sensor equals undisturbed air away from the platform or its structures.

Can two sensors identify the air velocity without other information?

Not automatically. Their difference can reflect motion, spatially varying air, alignment or response. The uniform-air comparison in this guide is a stipulated teaching assumption, not a general property of two physical installations.

Contents · Previous chapter · Next chapter

Return to the Yachting subject guide for the wider subject. For prerequisite vectors, use sailing, apparent wind and tacking angles. For related paper-based inference, read shared current from reciprocal water-relative velocities and position observations taken at different times. The separate guest-journey boundary guide addresses service responsibility rather than sensor kinematics.

Primary research context includes Brooks's 2008 spatially distributed motion study and Assessment of Atmospheric and Oceanographic Measurements from an Autonomous Surface Vehicle, whose manuscript reached final form on 26 September 2022. The FSU research explanation of true-wind calculation, dated 14 April 1997, provides historical context for translation and coordinate definitions. These studies concern their documented research systems. The geometries and numerical checks here are original synthetic examples. Links checked 23 October 2026.

Connect the reasoning to How X Works and The Gold Standard of Education: learning transfers into life. The portable skill is to locate a measurement physically before combining it with motion described somewhere else.

Contents · Previous chapter · Return to Yachting