Why Your Race Has Three Different Elevation Gains

You've finished. Your watch says: 2,640 meters of gain. You upload that same recording to Strava — it says 2,280. You open the race website — the organizer promised 2,500. Three numbers, one race, the same mountains.

The first instinct is to find someone to blame. The watch overcounted, Strava undercounted, the organizer calculated it their own way. The instinct is understandable, but it's pointing you in the wrong direction. Nobody lied.

The truth is less convenient: there is no single correct elevation gain. It's a property of the measurement problem itself — and it's worth understanding once.

01A question with no right answer

There's a classic question from geography: how long is the coastline of Britain? The answer depends on your ruler. Measure it on a map with a hundred-kilometer step, and you get one number. Use a one-kilometer step, and the line "grows": bays and headlands appear that the coarser ruler never saw. A one-meter step makes it longer still. There's no limit: the finer the ruler, the longer the coastline.

Elevation gain is the same problem, turned vertical. A course doesn't climb in a straight line: it's all small jumps — a two-meter scramble up a rock, a one-meter drop back down, a step made of exposed roots. Do you count those? Count everything a centimeter-scale ruler can see, and the gain runs off toward infinity right along with the length of the British coastline. Smooth it out instead, and the only question left is by how much. And every choice you make gives you a different number.

And this isn't just a nice analogy — the same thing has actually been measured on a bike. In a 2010 paper, physicist Dennis Rapaport took a single recording of a 42-kilometer hilly route and ran it through varying degrees of smoothing. The resulting elevation gain from that one track ranged from 680 to 1,410 meters — more than double — changing only with how heavily the curve was smoothed. His conclusion, verbatim: the question of which estimate is "correct" — "like measuring the coast of Britain — has no single answer."

We've seen this ourselves. The same GPX file for the Igualeja Trail race produces anywhere from 1,419 to 1,658 meters of gain in our engine — only the smoothing scale changes, the file itself never does.

02What elevation is even measured with

Before services can even start arguing about smoothing, the elevation has to come from somewhere. There are three sources, and they differ a great deal in quality.

GPS. Satellites are good at telling you where you are, and noticeably worse at telling you how high up you are: the whole satellite geometry works against vertical accuracy, because satellites are only ever visible from above. Garmin's own documentation puts GPS elevation accuracy at ±120 meters — and that's not a weak-signal scenario, it's the typical figure. That's not a rounding error you can wave off: any single point in the track can be off by tens of meters.

Barometer. Decent watches measure elevation from air pressure, and that's tens of times more accurate: per Garmin's spec, ±3 meters under stable weather and proper calibration, ±15 meters in ordinary conditions. But the barometer has its own quirks: a storm front rolls in, and your "elevation" drifts even while you're standing still; the temperature swings, and your calibration goes off.

Digital elevation model (DEM). You can skip trusting the device entirely and just ask the map: "what's the elevation at this exact point on Earth?" At Pacevine we use Copernicus GLO-30 — a terrain model captured by orbital radar and officially validated against satellite laser measurements: Copernicus GLO-30's officially stated accuracy is on the order of 4 meters vertically (LE90), and in practice, where it's been independently checked, it's often 2–3 meters. For comparison, the older SRTM model that many services still rely on has a stated accuracy of 16 meters.

Already, these three sources alone will give you three numbers. But the accuracy of a single point is only half the story: even a perfectly accurate barometer will over-count the total if you sum up every tiny wobble. The real spread comes less from the sources themselves than from the filters applied on top of them — more on that next.

03Invisible thresholds: every service has its own ruler

A raw track is noisy. Every service has to filter that noise out somehow — and each one does it its own way, and silently.

Strava states directly in its documentation: a climb only counts if it continues for more than 2 meters straight for tracks with a barometer, and more than 10 meters for tracks without one. ITRA, which certifies trail races, applies its own smoothing with a rolling 3-meter threshold — on top of whatever your watch already did. And ITRA separately requires race organizers to record their track at "one point per 10 meters of distance," because recording "once per second" gives you a different point density depending on how fast you're moving.

Our own engine has its own parameter too — we smooth the profile with a window and cut the course into segments using 18 meters of hysteresis: an elevation change only counts if a smaller up-down wobble than 18 meters doesn't "eat" it first. The number is different because the goal is different: we need a profile you can actually pace off of, not the most detailed possible record of every bump in the trail.

Notice there's no bad intent here, and no "correct" threshold. There's an engineering choice that each party made in its own way — and that almost nobody shows the user. Run the same file through Strava, ITRA, and a watch, and it's bound to produce three numbers. And it does.

04Where the organizer's number comes from

It looks like the number on the race website most often comes from an entirely different pipeline: it was calculated once, while the route was being designed — in mapping software, off a different elevation model, sometimes years ago — and it hasn't been checked against anyone's watch since. If the race is ITRA-certified, its D+ has been recalculated from the organizer's GPX using the method described above; if it isn't, that number could be almost anything, right down to a figure the course designer came up with off the top of their head.

So there's simply no basis for expecting your watch to match the race website: these are two different measurements, taken with different instruments and under different rules.

05What to do about it

Accept this once: elevation gain isn't an absolute quantity — it's the result of whichever counting method you chose. Asking "how much gain is there?" without specifying the method is like asking "how long is the movie?" without specifying whether that includes the credits.

The practical consequences are simple:

Compare like with like. Comparing your own training sessions against each other is fine: the same device, the same counting method. Comparing your number against someone else's — only with the understanding that a 10–20% gap on mountain terrain is normal, not anyone lying.

For planning, the profile's shape matters more than the total. The final D+ is one line for the race poster. Pacing is built on the shape of the profile: where the climbs are, how steep, in what order. The shape of the profile is far more stable across different methods than the sum is.

Demand the method, not just the number. A service that shows you an elevation gain without saying how it calculated it is showing you a number you can't compare against anything or double-check.

That's exactly why we built Pacevine this way: a course's headline D+ is computed from a digital elevation model (Copernicus, that same 2–3 meters of practical accuracy), not from your watch's track, where small noise adds up into an inflated total; pacing is computed from the smoothed profile — the one you actually run on; and the method is described openly, right here in this article. Our number has no obligation to match the organizer's figure. It has an obligation to be reproducible — and explainable.

Three numbers at the finish line aren't three versions of the truth. They're three rulers. In practice, there can be even more than three different numbers — exactly as many as there are methods in use. Now you know what sets them apart.