Measured Alpe d'Huez with SRTM and OSRM: A 12-Sample Accuracy Note

We have read most of what the internet has published about Alpe d'Huez. The climb runs 15.97 km from Bourg-d'Oisans at 773 m to 1,820 m at the summit, 1,048 m of gain, 6.6 percent on average, with a maximum figure of 13 percent that climbfinder.com prints and almost everyone else copies. That is the entire measured skeleton of the mountain, and it is the same skeleton every article claims to be describing. What differs is how honestly each piece treats the space between a road book, a satellite grid, and a routing engine — and on that question the coverage collapses into a single, repeated error.

The error is not in the numbers. The numbers are broadly correct. The error is that no one tells the reader where each number came from, which means no one can tell the reader what the numbers can and cannot do. That is the meta-critique. Below, in three passes, what the coverage misses, what it should include, and what a piece measured with SRTM elevation sampled along an OSRM polyline actually earns the right to say.

What They All Get Wrong

The dominant failure across the coverage of Alpe d'Huez is that three completely different measurement systems are quietly stitched into one paragraph and presented as if they came from the same source. A typical article opens by stating the length as 13.8 km or 14.454 km or 15.97 km — three numbers that are all defensible because they measure three different things — and then reports the elevation gain as "roughly 1,100 m," the average gradient as "about 8 percent," and the maximum gradient as "13 percent on the first ramp." Those four numbers, printed together, cannot all be simultaneously correct without disclosing which start point, which end point, and which polyline the writer used. Almost no one discloses.

The 13 percent maximum is the cleanest example. That number is a published figure — climbfinder.com prints it — and it refers to the steep opening ramp out of Bourg-d'Oisans before the road settles into its rhythm. It is not the output of any measurement the writer performed; it is a road-book value being repeated. When the same article, later in the same paragraph, states the length as "just under 14 km" and the gain as "about 1,120 m," it is doing arithmetic on a different mountain than the one the 13 percent came from. The maximum belongs to a segment defined by one source, the total belongs to a segment defined by another, and the writer never notices the seams.

The second wrongness is treating the average gradient as if it were a separate empirical fact rather than a division. Average gradient is elevation gain divided by horizontal length; if you accept 1,048 m of gain over 15.97 km from Bourg-d'Oisans to the summit sign at 1,820 m, the average is 6.6 percent and nothing else. It is not "roughly 7.9 percent" or "about 8.1 percent" unless you have quietly redefined the start or the end or one of the two vertical anchors. The 8 percent figure that shows up in older coverage traces back to a shorter measurement that ended at the resort's earlier landmark rather than the summit sign, and to a start point higher up the valley than the town of Bourg-d'Oisans itself. It is not wrong for that older mountain. It is wrong for the mountain the article claims to be describing.

What Is Almost Always Missing

The single largest omission is sample density. When a piece publishes a gradient profile of Alpe d'Huez — the little chart with green, yellow and red bands across the 21 hairpins — nowhere on the chart or in the article does it say how many elevation points were used to draw it. That number decides everything the chart is allowed to claim. A profile built from ten samples across 15.97 km can be trusted for its average and for the overall shape of the climb, but it cannot resolve a fifty-metre kicker on the exit of hairpin 7. A profile built from a sample every ten metres can. Readers stare at the same-looking chart and infer the same-looking authority from both, because the missing metadata is the metadata that matters.

Second missing item: the elevation source. SRTM 30 m — the OpenTopoData layer used for our measurements here — is a 30-metre grid derived from radar shuttle data. It is honest to within a few metres of vertical accuracy on open terrain but it does not know the road exists. It samples whatever surface the radar bounced off in that grid cell, tree canopy included. When the road cuts through forest — the middle third of Alpe d'Huez does exactly this between hairpins 12 and 6 — SRTM can read a metre or two high. Almost no article says this, and yet the same articles will print a maximum gradient to the tenth of a percent.

Third missing item: the routing engine. OSRM returns a polyline for the D211 up the mountain, but that polyline is built from OpenStreetMap geometry and is a decimated approximation of the actual road. Corners are cut, hairpin apexes are simplified, small realignments the local commune made after the OSM edit are absent. Sampling elevation along the OSRM polyline is not the same as sampling elevation along the tarmac. The gap is small — a percent or two of length on average — but it is real, and it compounds when the length is used as the denominator of a gradient. None of this shows up in the coverage. The chart appears, the number appears, the seams are hidden.

Alpe d'Huez The print from this article · from €29.95 View the print →

What I Would Say Instead

We would say the climb three times. Once from the road book, once from the satellite, once from what the two together honestly permit. From the road book: Alpe d'Huez from Bourg-d'Oisans is a hors-catégorie ascent 15.97 km long, 1,048 m of vertical gain, 6.6 percent average, with a maximum gradient of 13 percent on the opening ramp as published by climbfinder.com. Those are figures a reader can print on a wall and be right about. From the satellite: sampled at 30-metre grid resolution along the OSRM polyline for the D211, twelve points across the climb, the elevation curve reproduces the published gain to within the noise floor of SRTM on wooded alpine terrain, which is to say correctly enough to draw an accurate overall shape but not correctly enough to say which of the 21 hairpins is the steepest. From both, honestly combined: the mountain averages what the average says it averages, and it hits its steepest at a place the coarse sampling cannot pin down without denser data.

Twelve samples across 15.97 km is one elevation point roughly every 1.33 km. That density is enough to prove the average gradient — the arithmetic works, and 1,048 divided by 15,970 is 6.6 percent whether you sampled twelve times or twelve thousand. It is not enough to draw a per-hairpin gradient bar chart of the kind that dominates the coverage. Anyone drawing that chart from twelve samples is producing decoration, not measurement. Anyone drawing it from denser sampling should say how much denser and from what source, because the reader has a right to know whether the red band on hairpin 4 is data or is illustration.

The published 13 percent maximum belongs in an article about Alpe d'Huez, but it belongs there as what it is: a road-book figure attributed to a source, sitting alongside the measured average, not competing with it. There is no contradiction between a mountain that averages 6.6 percent and hits 13 percent in one place, because averages and maxima answer different questions. The contradiction only arises when a writer treats both as products of the same measurement and does the arithmetic that follows. The maxima and averages of Alpe d'Huez as it appears in coverage are frequently arithmetically incompatible with each other and with the length being cited in the same sentence, and no one notices because no one is checking.

That is the frame we would use, and it is the frame our own prints of Alpe d'Huez — twelve-sample SRTM profiles over the OSRM polyline, road-book maxima annotated separately, sources named on the print itself — are drawn from, at see the Alpe d'Huez print. The mountain is honest. The question is whether the article about it is.