Ski Jump Takeoff Velocity: An IMU Protocol for Isolating the Athlete's Vertical Drive

A coach standing at the knoll of a K90 hill watches two jumps land within a meter of each other and assumes the athletes did roughly the same thing at the table. They didn't. One carried more in-run speed off a slightly better wax job and contributed less leg drive; the other came off the table slower but drove harder into the transition and got more vertical pop for it. Jump distance is a downstream number — it folds in-run velocity, wind, table geometry, and the athlete's own extension into a single figure, and by the time you're reading it off a tape measure, you can no longer tell which part came from the hill and which part came from the athlete's hips and knees. If you're trying to build takeoff power over a season, that's the exact variable you need isolated, and it's the one distance can't give you.

Vertical takeoff velocity — the upward velocity component the athlete generates during the roughly 250-300ms window of ground contact on the takeoff table — is the biomechanical variable sports scientists track when they want to know whether an athlete's leg extension is improving independent of hill speed. Virmavirta et al. (2009) measured this directly with force plates embedded under the takeoff table at an Olympic-level competition and found vertical velocity generated at takeoff in the 2.3-2.7 m/s range among finalists, correlated with jump distance even after in-run speed was accounted for. A force-plated takeoff table isn't something most programs have access to. This guide lays out a field protocol for approximating that measurement with a single body-mounted IMU: mounting, axis alignment against a body tilted 35-40° forward through the in-run, windowing the ~0.3-second takeoff event out of a longer signal, and the benchmark bands and signal errors that come with doing this outside a lab.

Jump distance is the outcome everyone scores on, and that's exactly why it's a poor training variable. It's sensitive to wind gusts on the day, wax and ski selection, table setting, and in-run speed — none of which the athlete controls through their own physical qualities. Two jumps separated by wind alone can differ by several meters with identical takeoff mechanics underneath them. Video overlay helps a coach see body position, but a camera at 30 or 60fps captures maybe 8-18 frames across a 250-300ms takeoff, which is enough to see the shape of the extension but not precise enough to compute a velocity number a strength program can track week to week.

What a program actually needs is the piece of the outcome the athlete's own musculature is responsible for: how much upward velocity did the legs generate in the brief window they had to generate it, independent of how fast the hill delivered them to the table. That's a physical quality trainable in the gym and on the hill, and it's the number that should move as a takeoff-strength block progresses — even in a stretch where jump distances stay flat because wind or wax conditions worked against the athlete.

Through the in-run, an athlete travels at roughly 85-95 km/h in a deep aerodynamic tuck, torso close to horizontal over the skis. At the transition onto the takeoff table's curved radius, they extend the hips, knees, and ankles through a window of roughly 250-300ms, converting stored elastic and concentric leg power into an upward velocity component while horizontal ski velocity carries over largely unchanged. Vertical takeoff velocity is that upward component alone, measured in the global (gravity-referenced) vertical axis at the instant the skis leave the table — not the resultant velocity, and not anything measured relative to the body's own long axis, which is still tilted forward at that moment.

That distinction is the entire technical challenge of measuring it with a wearable sensor. An IMU reports acceleration in its own local axes, and during the in-run those local axes sit 35-40° away from true vertical because the trunk is pitched forward in the tuck. Ettema, Bråten, and Bobbert (2005), modeling in-run dynamics in a simulation study, showed this posture is a genuine trade-off: a deeper, more prolonged aerodynamic crouch reduces drag and preserves in-run speed, but the later and more abruptly the athlete extends out of it, the less time remains to apply force before the skis leave the table — in-run posture and takeoff force-generation time are mechanically linked, not independent choices a coach optimizes separately. Isolating vertical takeoff velocity cleanly requires rotating the raw accelerometer signal out of the tilted in-run frame into the global vertical before any velocity integration happens — skip that step and the number is contaminated by however steep that athlete's particular crouch happened to be.

Mount the IMU at the sacrum, on a low-profile belt worn under the suit at the small of the back — close to the body's center of mass, the standard placement borrowed from countermovement-jump testing, and clear of the limb-segment noise a boot or shin mount would pick up during simultaneous ankle, knee, and hip extension.

Calibration Sequence

1. Standing-neutral hold (3 seconds): athlete stands upright in ski boots, establishing the zero-reference orientation against gravity before the in-run tuck is assumed.
2. In-run tuck reference: athlete holds the full aerodynamic tuck for 3 seconds, giving the app a second orientation reference so it computes this athlete's actual forward-tilt angle rather than assuming a fixed 35-40° figure — tuck depth varies by 8-10° across athletes and by hill.
3. Three practice runs at reduced speed confirm the takeoff-detection algorithm correctly flags the transition-to-liftoff window; a green check marks success per run, and two consecutive misses call for a belt-tightness recheck, since a loose sensor rotates independently of the pelvis and corrupts the tilt reference.
4. Sampling and sync check: confirm 800Hz sampling is active and, where available, sync the IMU timestamp against a photocell at the start of the table radius — the accelerometer alone approximates the window, but an external reference removes ambiguity about exactly where it starts.

Run the full in-run through landing for every recorded jump rather than isolating the takeoff live — the extraction happens in post-processing, with the full acceleration and orientation trace available to work with.

Processing Steps

1. Rotate the raw acceleration signal from the sensor's local frame into the global vertical using this athlete's own tuck-angle reference from calibration, not a generic assumed tilt.
2. Locate the takeoff window using the characteristic signature: a sharp rise in vertical acceleration as the athlete drives into the table's radius, lasting approximately 250-300ms, ending at a distinct drop to near-zero as the skis leave the surface (free-fall onset).
3. Integrate the rotated vertical-axis acceleration across that window only, from a vertical-velocity value of zero at window onset — the radius is built to redirect horizontal speed, not deliver vertical velocity before the athlete extends.
4. Discard flagged jumps — wind-affected starts, a missed gate sync, or a detection window under 200ms or over 350ms, which usually mean a mistimed extension or a sensor artifact rather than a real takeoff.
5. Log vertical takeoff velocity per jump against session mean and SD, building a personal baseline from at least 5-6 clean jumps before treating any single session's number as meaningful.

A single jump's number means little alone. Track the session mean rising across a training block and the session SD tightening — together they show more vertical velocity, generated more repeatably.

These bands are field-practical categories built around the range reported by force-plate research at competition level, not a strict cutoff — a wearable IMU protocol carries more measurement noise than an embedded force plate, so treat the boundaries as approximate.

Hill size matters more than most coaches initially assume: a K120 table redirects the body through a shallower, faster transition than a K90, which changes the force-time characteristics of the extension even for the same athlete. Compare an athlete's numbers across sessions on the same hill before comparing across hills.

Field IMU protocols on a moving winter-sports course fail in a handful of predictable ways. Knowing the signature saves a coach from training a bad number for weeks before catching it.

Vertical takeoff velocity is most useful tracked as a rolling trend across a training block rather than a single trackside readout.

• Every on-hill session: log vertical takeoff velocity for every clean jump, building the session mean and SD.
• Weekly: compare the week's session means against the prior 2-3 weeks — a flat or declining trend across a takeoff-strength block, even while jump distances look fine, is worth reviewing against gym-side triple-extension and reactive-strength numbers.
• Before hill changes: re-establish a fresh baseline on the new hill size rather than comparing raw numbers across a K90-to-K120 jump, given the force-time differences the transition radius introduces.
• Windy or marginal-condition days: weight these sessions less heavily in the trend, since in-run stability noise inflates session SD independent of the athlete's actual mechanics.

Key References

• Virmavirta, M., Isolehto, J., Komi, P. V., Schwameder, H., Pigozzi, F., & Massidda, M. (2009). Take-off analysis of the Olympic ski jumping competition (HS-106m). Journal of Biomechanics, 42(8), 1095-1101.
• Ettema, G. J. C., Bråten, S., & Bobbert, M. F. (2005). Dynamics of the in-run in ski jumping: A simulation study. Journal of Applied Biomechanics, 21(3), 247-259.