When Strava elevation looks wrong it is usually because altitude is the hardest quantity a bike computer has to measure. Position comes from satellites and is checked against several at once. Height comes from GPS altitude, where the geometry is weakest, from a barometric sensor the weather disturbs, or from a terrain model the platform holds.
Strava elevation is therefore an estimate standing on top of another estimate, and elevation gain compounds the problem further by summing only the upward steps it sees. GPS altitude and a barometric reading fail in different directions and for different reasons, which is why a single climb can yield several defensible totals.
Barometric sensors against GPS altitude
A barometric altimeter infers height from air pressure, which falls predictably as you climb. It is precise over short intervals and responds quickly, which makes it good at capturing the shape of a climb. It is also fooled by weather fronts, by wind hitting the sensor, and by riding into a building.
Satellite altitude needs no sensor at all and drifts with no weather, but it is inherently less accurate vertically than horizontally. The geometry is unfavourable: satellites sit above the horizon and never below it, so the vertical component of any position fix carries several times the error of the horizontal one.
| Barometric | Satellite | |
|---|---|---|
| Responds to | Air pressure | Signal geometry |
| Short-term precision | High | Low |
| Long-term drift | Weather dependent | None |
| Typical failure | Pressure change mistaken for a hill | Noisy vertical scatter |
| Affected by tunnels and buildings | Strongly | Signal loss only |
Why elevation gain inflates, not deflates
Elevation gain sums upward movements and ignores downward ones entirely. That asymmetry means sensor noise can only ever add to the total. If the sensor jitters by a metre either side of the truth a thousand times across a ride, the downward jitters are discarded and the upward ones are counted as climbing.
This is why a genuinely flat ride can report two hundred metres of ascent. Nothing went wrong in any single reading, and no component failed. The summing rule simply converts random noise into a one-directional total, and a longer ride accumulates more of it than a short one over the same terrain.
Why Strava elevation gain differs between two riders on one climb
Strava elevation gain differs between riders because different sensors, smoothing and sampling rates produce different totals from identical terrain. A device recording every second sees more noise than one recording every five. A barometric unit in a jersey pocket reads differently from one mounted in clear air on the bars.
Platforms also apply their own corrections. Some replace device altitude with a terrain model looked up from the coordinates, which removes sensor noise and introduces map error instead. On a road that crosses a bridge or a tunnel, the terrain model is confidently wrong in a way no sensor would be.
What can be fixed after the ride
Elevation correction re-derives height from a terrain database using the recorded positions. It is the right choice when a barometric reading has obviously drifted across a long ride, and the wrong one wherever the route leaves the ground. Neither source is authoritative, so the honest approach is knowing which one produced your number.
Where one isolated spike is the problem, cropping the file is far more precise than correcting the whole of it. A GPX audit locates the affected points directly, which turns a vague sense that the total looks too high into a specific list of moments where altitude moved without the bike moving with it.
Elevation questions
Why is my Strava elevation gain higher than my friend's on the same ride?
Almost always a sensor and smoothing difference, not a routing one. Barometric units capture more genuine detail and more noise; satellite-derived altitude is smoother and flatter. Since gain only sums upward movement, the noisier device reports the larger total on identical terrain. Comparing the two totals tells you about the devices, not about the road you both rode.
Should I enable elevation correction?
It helps when a barometric reading has drifted across a long ride, and hurts on routes with bridges, tunnels or multi-level roads, where a terrain lookup returns the ground and not the surface you rode. Judge it per ride rather than setting it once. A long tunnel or a multi-level bridge is the clearest case for leaving it off.
Does weather really change recorded climbing?
Yes, measurably. A pressure system moving through during a long ride shifts the barometric baseline, and the altimeter reads that shift as slow, continuous ascent or descent. On a six-hour ride in changeable conditions the error can reach tens of metres before any noise is counted. It accumulates in one direction for as long as the front takes to pass.
Why does a flat ride report several hundred metres of climbing?
Because gain counts upward movements and discards downward ones. Sensor noise oscillating around the true height contributes only its upward half to the total. Multiply a metre of jitter by a few thousand samples and a genuinely flat route acquires a substantial and entirely fictional climb. Smoothing the altitude channel before summing is what removes most of it.
Which figure should I trust for a long climb?
For a single sustained climb, the barometric figure is usually closer, because the pressure change is large relative to the noise and the terrain model has no opportunity to misplace you. For total gain across a rolling ride, treat any figure as approximate. The error there is dominated by how often the device sampled, not by the terrain itself.
Related: the GPX audit reports raw and smoothed climbing side by side, GPS accuracy covers the horizontal equivalent, and the VAM calculator turns a climb into a rate.