Mountain Bike Specific Training: Part 2 – Measuring bumpiness


Mountain Biking is really bumpy. And most riders think the bumpier the gets, the better. I agree. I like my races technical. This creates a problem for those that train with power. Putting down power over bumpy terrain is a different beast than doing so on a smooth road or trainer. Rather than just needing well tuned prime movers to get the power down, you need accessory muscle activation to stabilise yourself.

Measuring this instability, and then recreating it in training has thus far been impossible. Most mountain bikers ride off-road as much as possible, but when that’s not an option, we have to resort to road riding. If we could correct power data to account for the terrain over which it was produced, we’d have a much better idea of how hard a race or training effort was.

Dave Schell and I have begun trying to measure this instability. We’re currently using the Wahoo TickrX, a heart rate monitor that also measures trunk angle and upper body movement using accelerometers. It’s designed for runners, but we’re trying to hack it to be useful for mountain biking.

Here’s the theory: when riding along a smooth road, your upper body is almost completely still. The bumpier it gets, the more your upper body will move. Using this, we begin to estimate the bumpiness of a ride, and use that number to correct power data accordingly.

ΔTrunk Angle

IF [(trunk angle) at t(n)] ≠ [(trunk angle) at t(n+1)] THEN count = 1)/total time in seconds

This gives us a number between 0 and 1 that we can use to score the “roughness” of the ride. For example, a score of 0.6 could be translated to a ride that was 60% rough and 40% smooth.

It’s reasonably crude, but it works for creating a ride ranking that can easily compare rides. The things it’s missing: it does not factor in magnitude of change in torso angle. Bigger bumps, drops and step ups should be weighted more heavily. It doesn’t factor in stopped time, when torso angle will likely be 90 degrees to the ground and thus be a big change from riding position. Or reaching for water bottles and shoes adjustments etc. I’m thinking that these actions will be few enough over the course of a typical ride that they should muddy the data too much.

To account for magnitude of change, I subtracted one time point from the next to give an actual degree of change. This was the second attempt:

When Wtn ≠ 0, ΔTA/time = [(trunk angle)tn-1 – (trunk angle)tn]^2) / (total time in seconds)

ΔTA: Change in trunk Angle

Wt: power in watts at time t

This then gives us an average degree of change of trunk angle. This metric is unbounded, unlike the number above, so it’s a little harder to use, but still gives a good comparision. It has to be squared to account for the negative changes in torso angle. In theory, matching the torso angle data to power meter data should allow us to use only data points where power is not zero. Then we could see the changes in torso angle while pedaling, which is what we’re really getting at: how interrupted was your power output?

Correcting the raw power numbers would be pretty simple. If we assume that a road ride would end up giving a Bump Factor of 0, power could be multiplied by ((BF/100)+1) to give terrain normalised power (TNP)

TNP = Power * (BF/100+1)

for example:

300 watts on a smooth flat road, where the bump factor = 0

TNP = 300 * ((0/100) + 1)
TNP = 300 w

300 watts on a mountain bike trail where bump factor = 4

TNP = 300 * ((4/100)+1)
TNP = 312 w

Thus, the terrain could be said to cost an extra 12 watts. A first trial run of riding Heil Ranch outside of Boulder gave a BF of 4, so that’s a good figure to start with This is likely an underestimate. Once the final metric has been calculated, we can begin working out what the actual correction needs to be.
This is a work in progress… there’s a lot more data that the TickrX can capture, and we’re hoping that some of the metrics it’s already recording could be hacked into a riding version. My coach Dave Schell is working with the people at Wahoo (who make the TickrX) to figure out how we can get raw data from their accelerometers to determine what movements are actually made during MTBing, and which ones are important. Like I said, this is a VERY EARLY work in progress….