GPX Elevation Spikes: Why a Route Feels Wrong on Your Trainer
When an indoor route has a sudden wall or a climb that isn't there in real life, the cause is almost always the elevation recorded in the GPX, not the trainer. GPS and phone altitude carry metres of noise, and gradient is height change divided by distance, so a 5-metre error over 50 metres reads as a 10 % ramp. Cycloscape cleans every file on upload (it discards impossible heights, fills gaps, smooths the profile and measures gradient over about 30 metres, capped at 35 % up and 30 % down), but a badly recorded file can still need fixing at the source.
Why a small error becomes a big ramp
Gradient is rise over run. Five metres of rise over 50 metres is 10 %; the same five metres over 10 metres is 50 %. Elevation errors don't need to be large to matter, only short. That is why spikes show up as brief walls, while a genuine climb builds over hundreds of metres.
Where the error comes from depends on the recording. GPS is less accurate vertically than horizontally. A barometric altimeter drifts with the weather over a long ride. And files whose elevation was corrected from a map can dip under a bridge or in a tunnel, because the map gives the height of the valley floor, not the deck.
What Cycloscape does to your file
Every GPX goes through the same clean-up before it becomes a route. It doesn't invent a better route; it removes what can't be real.
- Heights outside what's physically possible (below the Dead Sea shore or above Everest) are treated as missing.
- Missing heights are filled in by distance between the nearest known ones.
- Lone points far from both neighbours, such as a cold-start GPS fix, are dropped.
- The elevation profile is smoothed with a short moving average to damp recording noise.
- Gradient is measured over about 30 metres, not point to point, and capped at 35 % up and 30 % down.
- A file with no elevation at all becomes a flat route.
How to spot the problem
After uploading, look at the elevation profile before you ride. A spike is a narrow needle or a saw-tooth pattern; a real climb is a slope. Then compare with what you know: does a road you've ridden really have a 20 % section there? Does the total climbing look plausible for the area?
Check the route's line on the map too. A track that cuts across a switchback, jumps between two recordings or joins two rides changes the distance, and a wrong distance distorts the gradient as much as a wrong height.
Fix it at the source
The cleanest elevation usually comes from a planned route rather than a recorded ride. Route planners in Strava, Komoot and Garmin Connect take heights from an elevation map, so the profile is smooth and repeatable; a ride recorded on a phone carries every wobble of its altimeter. If a recorded activity is the only file you have, see whether the platform offers an elevation correction for it before exporting.
Keep the original file before changing anything. After a correction, compare the distance, the total climbing and the shape of the main climbs with the original, and ride the questionable section again.
Don't flatten real climbs
More smoothing is not more accuracy. Heavy smoothing removes genuine short ramps, the kind that make a route what it is. Fix data that is wrong and leave the rest alone. If the real climbs are simply too hard on your trainer, lower Trainer difficulty instead: the route stays true and your rides on it stay comparable.
And if the same harsh feeling appears on several routes you trust, the data is not the problem. Look at the difficulty setting, your gearing or an ERG workout instead.
Frequently asked
My GPX has no elevation. What happens?
Cycloscape builds the route flat. Export a planned route from Strava, Komoot or Garmin Connect instead: route planners add elevation from a map.
Why does Cycloscape show different total climbing than Strava?
Each platform smooths elevation its own way, and total climbing is very sensitive to it: small wobbles counted on every point add up. A difference of a few percent is normal; a difference of double is a data problem worth checking.
Does matching total elevation prove the gradients are right?
No. Two files can report similar total climbing while placing the climbs differently or squeezing them into shorter distances. Look at the profile, not only the total.