Adding to my collection of profiles for climbs I've done, here's one for the Axalp climb, approximately 10 km west of Meiringen, Switzerland, on the south shore of Lake Brienzersee.
First the profile. I used my Garmin Edge 500 for position and altitude. On the climb up, I lost GPS signal for a bit, which resulted in a data gap. Since I take the data straight from the FIT file, I rely on Garmin's distance determination, which would normally be good if my Powertap had been functioning, but the super-cheap alkaline LR44 batteries I got off Amazon don't last as long as silver oxide 357's with which the Powertap ships. As a result, I needed to rely on the Garmin's GPS distance determination, which doesn't interpolate across gaps, unlike Strava's distance determination. It was easy enough to convert latitude and longitudes into distance, but the Garmin's smarter than that: local variation in position turn straight paths into zig-zags, and the result is a persistent overestimation of total distance. Honestly I don't know how Garmin does its position->distance conversion, but all I know is it doesn't handle gaps well.
But I was saved. I descended the same hill I'd climbed, so I was able to use the data from the descent, reverse the order of the points, and presto-magicko, I have climb data. Of course this isn't strictly accurate because I climb the opposite side of the road as I descent, so switchback insides become outsides and vice-versa, but on the narrow roadway the difference is small.
Using the descending data offers an additional advantage, which is that if barometric pressure changes due to weather, or temperature changes, corrupt the altitude data, these changes will likely be a lot less at descending speed than at much slower climbing speeds.
Anyway, here's the profile:
My profiles all end up looking fairly similar because I use the entire plot from lower left to upper right. I then take characteristic subsets of the curve and do nonlinear regressions to fit a gradient and an offset to the segments, providing a feel for how the gradient of vairious subsets of the climb behave. But it's also useful to consider the grade extracted on a local basis. For this I first do a smoothing operation using a convolution function with characteristic length of 50 meters:
Yes, this one is steep, peaking out in the middle at over 15%, but fairly consistenly hanging out at 10% or steeper. There was some relief through and beyond Axalp itself. I continued to a T-intersection with a street sign:
It rated out at 160% of Old La Honda using my formula, as I rode it. However, the climb can be ridden a bit further, taking the right-hand option, until it turns to gravel/dirt.
But the light was fading and I wanted to reach the bottom of the descent before I needed my headlight. So I didn't dally too long, or explore the way further. I did, however, stop long enough to take this photo, not far from the top: