Fixing Barometric Altimeter Drift That Skews Jump Height Readings Indoors

Forty minutes into a return-to-play testing block, the numbers stop making sense. The first three countermovement jumps of the session sit at 51-53 cm, tight and repeatable, exactly the scatter you expect from a healthy athlete on a normal day. By the last block, before anyone has touched a squat rack or added a single extra rep, the same athlete is reading 59-61 cm on the identical jump. Nobody gets 15 percent more explosive in the span of one testing session. What actually happened is quieter than fatigue or a coaching cue landing well: the rooftop HVAC unit above the turf cycled on twice during that block, the loading-dock door opened for an equipment delivery, and the barometric pressure sensor feeding the device's jump-height algorithm drifted its own zero point right along with the room. The jump did not get taller. The floor the device measures from moved.

A barometric altimeter estimates height indirectly, by reading local air pressure and converting a change in pressure into a change in altitude using the simple physical fact that pressure falls as you rise: roughly 12 pascals per meter near sea level. Most jump wearables that carry one are not using it as the primary jump-height sensor. The accelerometer still does that work in real time, integrating acceleration twice over the flight phase to get displacement for each individual rep. The barometer's job is narrower and more specific: it acts as a slow, low-frequency anchor that keeps the accelerometer's double integration from drifting across a session, since any accelerometer, no matter how good, accumulates small integration errors that compound into a growing bias over dozens of reps if nothing ever corrects them back toward a known reference.

That correction only works if the reference stays put. The algorithm treats the ambient pressure baseline captured near the start of a session as a fixed stand-in for ground level for the rest of that session. Outdoors, on a calm day, that assumption mostly holds. Indoors, in a mechanically ventilated building, it frequently does not, and the same feature built to stop one kind of drift becomes the entry point for another.

The pressure change a real jump produces is tiny relative to what indoor air pressure does on its own. At sea-level density, the barometric formula puts the pressure drop across a jump's peak height at roughly 1.2 pascals per centimeter of rise, so even a strong countermovement jump barely moves the sensor.

A signal that sits at only one to three times a sensor's own noise floor cannot be trusted rep by rep on its own, which is exactly why no device tries to read jump height straight off the barometer. It is only usable once it is heavily filtered and blended with the accelerometer over many reps. That filtering is also what lets a slow, real change in ambient pressure slip past undetected, since a low-pass filter built to smooth out per-sample noise treats a genuine multi-minute drift as a legitimate, slow-moving signal rather than as an error to reject.

Indoor ambient pressure is not a constant. Research into barometric sensors for indoor positioning has documented exactly the mechanisms at play in a gym or testing room. Muralidharan et al. (2014, presented at HotMobile) tested smartphone-grade barometers for floor-level detection in office buildings and found that HVAC operation and door movement could push apparent altitude readings by an amount large enough to be mistaken for a full floor change within minutes, purely from ambient pressure swings with no real elevation change involved. Their setting was floor counting in multi-story buildings rather than sport wearables, so the exact magnitudes will not map one-to-one onto a single weight room, but the underlying mechanism, uncontrolled pressure change driven by mechanical ventilation, is the same one degrading a jump-height reading.

Xia et al. (2015, Sensors) studied the same problem from the correction side, showing that a lone barometer drifts enough from ambient changes to generate false floor transitions, and that referencing a second, stationary barometer to cancel out the common ambient pressure change substantially reduced that drift-driven error compared to a single uncorrected sensor. Their validation used stairwells and multi-floor buildings over minutes-long windows, not the sub-second flight time of one jump, so their exact averaging intervals do not transfer directly to a jump algorithm. What does transfer is the fix itself: a stationary reference point that separates real device movement from the room's own pressure changing underneath it.

The scale of the mismatch is the whole problem. HVAC fan cycling alone commonly moves ambient pressure by a few tenths of a hectopascal over a 10-20 minute cycle, a door opening near a sensor can produce a sharper swing on the same order, and neither has anything to do with how anyone is jumping.

Set beside the 0.036-0.090 hPa a real jump produces, even the mild end of ordinary HVAC cycling dwarfs the actual signal several times over.

This is the detail that makes the artifact so easy to mistake for real adaptation or real fatigue: it is not noise. Noise scatters a reading up and down, rep to rep, in a way that a trend line and a bit of statistical sense can shrug off. Drift moves in one direction, smoothly, for as long as the room's pressure keeps moving in that direction, and a smooth one-directional creep is exactly the shape a coach expects from genuine improvement or genuine decline across a session.

The mechanism runs through the same correction that was supposed to prevent accelerometer drift. As ambient pressure falls indoors, the fusion algorithm reads that as the device rising relative to its stored baseline, and nudges the accelerometer's height output upward to match. It has no way to tell the difference between the athlete's mounting point actually moving and the room around it losing pressure for reasons that have nothing to do with sport. Every jump computed after that point inherits the same upward nudge, and if the room keeps drifting, the next correction stacks on top of the last one. That is precisely the pattern in the opening scenario: 51-53 cm early, creeping to 59-61 cm late, with the HVAC cycling twice in between and nothing changing in the athlete's legs.

Run this on any barometer-fused jump device feeding height into a testing or return-to-play decision, and repeat it whenever the testing room, HVAC schedule, or session length changes meaningfully.

1. Confirm the device combines a barometric sensor with the accelerometer for jump height, and check the settings menu for a manual or automatic baseline re-zero option.
2. On a normal training day, let the device log jump height continuously for 40 minutes with no re-zero, having the athlete perform 3 consistent countermovement jumps every 10 minutes at a fixed sub-maximal effort.
3. Plot the readings against elapsed time. A steady creep in one direction, with no matching change on video, is the drift signature; scatter with no consistent direction points to a different problem entirely.
4. Cross-check at least two time points against an independent measure: 240 fps side-view video, using flight time squared times gravity, divided by eight, to get an independent height, or a contact mat if one is available.
5. If drift is confirmed, set the device to re-zero every 8-10 minutes, or every set, during the natural 3-5 second stationary window between reps, instead of relying on one baseline captured at session start.
6. Repeat the 40-minute diagnostic with re-zeroing switched on and confirm the trend now sits flat, tracking the video-based reference within your device's normal per-rep noise band.

A collegiate volleyball athlete performed a fixed sub-maximal countermovement jump every 10 minutes across a 40-minute testing block in a gym with rooftop HVAC on a 15-minute cycle. The table shows the device's reading with no re-zero, the same session repeated the following week with re-zeroing every 10 minutes, and an independent 240 fps video reading for both sessions.

Without re-zeroing, the device climbed 8.4 cm over 40 minutes against a video reference that never moved more than half a centimeter from its own average. With re-zeroing every 10 minutes, every device reading landed within 0.5 cm of the video number, well inside normal rep-to-rep noise for a countermovement jump. Nothing about the athlete changed between the two sessions; the only variable was how often the baseline got reset against the room.