Data Quality in Strava Segments: GPS, Length and Elevation
When a segment reads 1.78 km and −0.3 percent, both numbers are estimates. For predictions you need to know how large the error can be.
Typical error sources
Segment lengths come from the original recording and can be one to three percent off. Elevation is worse: barometric data drifts, GPS altitude is noisy, and correction models smooth small undulations away.
How to spot problems
Be suspicious of segments whose elevation profile does not match the terrain, whose average gradient differs strongly from the start-to-finish height difference, or where your own rides produce very different times at similar power.
Impact on predictions
One percent of length error is one percent of time error — nearly two seconds on a three-minute segment. A 0.5 percentage-point gradient error hits flat segments even harder.
This is exactly why an existing best time is such a valuable anchor: systematic data errors cancel out when you work with differences instead of absolute values.
In short
Trust segment metadata only roughly, and work with differences against your own times.
Frequently asked questions
Should I create my own segment?
If you want a cleaner profile and clear start and finish points, yes. The downside is an empty leaderboard with no reference times.
Does a better head unit improve the data?
Yes, multi-band GNSS reduces positional noise considerably — but the segment definition itself stays as it is.