Thursday, August 25, 2011

Wednesday Noon ride and Strava timing accuracy

After working at my job for close to year, I decided it was finally time to indulge in the Wednesday Noon Ride this week.

The weather was perfect: warm but not hot, a so-very-welcome liberation from the tights and long sleeves which have been part of virtually every ride I've done this year thanks to San Francisco's persistent "marine layer". And Matt has done a great job of championing the often-neglected Wednesday ride which has climbed Old La Honda since the Egyptians were first domesticating cats.

I'd done two of the Friday rides so far and the experiment was generally a success: out the door by a drop-dead time of 11:30, back before the 1:30 pm time when the cafeteria shut down its main lunch service. Obviously it's not something to do every day but since my typical lunch break is a line-dependent 7-12 minutes it takes to go to the cafeteria, buy something, then get back to my terminal, it'd still be doable to go even once per week and still average less than a half-hour per day of lunch time. I figured adding the Old La Honda climb - 84 descent and cutting out a missing portion of the Friday loop would add around 20 minutes to this, assuming I didn't wait for the traditional regroup at the top. That's an extra 4 minutes per day.

So there I was, and what a turn-out! Once we'd picked up Greg, who'd ridden from San Francisco, with Kieran also in the group we had three riders who'd done sub-16:15 times. Matt in particular had been on a tear, getting sub-16:05 PR's each of the previous two weeks. So it was going to be an interesting climb.

No way I was going to PR, though. I've barely been climbing this year, a "long climb" for me being over two minutes. I'd done a few north-gate Diablos, which went surprisingly well (not great times, but not suffering my usual Juniper parking lot fade). But other than this my two Kings Mountains and three Old La Hondas had been on the north side of slow. Indeed, when I'd left home Wed morning I'd brought my steel Ritchey Breakaway rather than my much lighter, weenied-out Fuji SL/1. It wasn't worth risking the carbon on the Caltrain commute to work, or for that matter race tires on the glass-strewn Central Expressway which was the quickest route to and from the ride.

I was impressed by the pace to the climb: none of the attacks or surges which virtually always characterize the nooner, no leg-breaking suffer-fest up Alpine Road. It was clear under Matt's leadership this was all about Old La Honda. He was posting a list of times on Facebook, and there was no credit given for a good Woodside town line sprint.

So here was the main point of this post: times on the Facebook post were self-timed, old school stopwatch. Linda, a long-time noon-rider, had suggested that self-timing was obsolete thanks to Strava. I responded that Strava still has significant errors. As I documented here with data recorded by Alan Weatherall using two Garmin GPS units mounted to the same handlebars, times typically differ by up to 4 seconds, which is twice the default Garmin sampling time (Garmin also offers 1-second sampling time). But sometimes differences are larger.

As we hit the base of the climb I hit my lap timer right on schedule. Okay, so I did it at the trailing edge of the bridge, while some claim the leading edge is the official start. Public confession: I time from the trailing edge. I figure the leading edge convention some proclaim is sufficently obscure that many if not most don't follow it. "Bridge to stop sign" is all most people know.

I'll spare the details of the climb. I felt good, though: a good pace for me feels easy at the start slowly increasing in difficulty as the climb progresses. However, drinking from my mostly-empty bottle left me in gasping for air, a good sign I was still working close to my aerobic limit. And indeed the difficulty did increase. At the end, I tried to drop the rider I was with, but he repassed me, and I followed his wheel to 100 meters from the finish, when he gapped me. I tried to close, but couldn't quite recatch his wheel.

I hit the Garmin Edge 500 lap button again right at the stop sign. The last time I'd seen on the lap timer was 18:08, so add a few seconds to that. I then continued on, more slowly of course, up the remaining short climb on Skyline Boulevard before beginning the descent to the Hwy 84 junction where I stopped for water.

Despite my rush to get back to work I took an indirect route so I could chat with Greg (who'd joined me on the way to the store). I still made it back with just six minutes to spare before my self-imposed 2 pm deadline. Within two minutes of reaching my desk I'd uploaded my data and Strava had reported my segment times.

On the plus side, I'd gotten overall second places on two of my favorite segments associated with my normal commute: the Central Sprint and the Ringstorf to Moffitt Sprint. But the focus today was on Old La Honda. Sure enough, I had a Strava PR for the climb. But the claimed time was a bit shocking: 18:22. Whoa -- had really that much time passed from my last glance at the clock as I began my final sprint to the stop sign?

So then I checked my lap time: 18:14.55. Even that was longer than I had figured, but I always seem to underestimate the passage of time at the end of climbs. But that was 7.45 seconds faster than Strava estimated. As I previously noted, if a segment is defined too close to a congregation point, for example the Old La Honda - Skyline junction, then when a rider stops Strava may include waiting time in the segment timing. However, I'd never stopped here: I'd ridden on past the junction in my rush to get back to work. To see the problem you need to look at the data tracks.

Here's the start of the climb, comparing the lap to the segment. They agree fairly well:

But the top of the climb is a different story. Here I overlay the two plots in a GIF animation (reload or click on the image if the animation has stopped). The error in the GPS signal is obviously substantial. It's so substantial, in fact, that it's notable that Strava even gave me credit for the segment. The end of the lap, which is within a sampling time of the stop sign, and the identified end of the segment are way off:

So the result of this little "experiment": if you want a credible Old La Honda time, still time it yourself. The Strava-Garmin combo simply isn't up to the task.

From the work perspective I think the occassional Wed nooner can be worked into the schedule and still be well under the 5 hours per week "standard" for lunch. And on the cycling side that 18:14.55 wasn't bad at all given my lack of climbing this year and all the extra weight I had on the climb yesterday. So even if it did mean I was left eating the Cream of Rice I had at my desk as an emergency ration for lunch, it was worth it.


Mark S. said...

Dan - Thanks for this interesting post and the excellent blog. Here's a question - have you ever tried to quantify, either theoretically or empirically, the effect of drafting on Old La Honda times? I've poked around on the web a bit for a good drafting calculator but come up empty. I know you once wrote that even light wind effects could have about a 20 second impact on OLH times. I'm guessing a good draft is worth about 10 seconds on an 18-20 minute time, but that's just a guess. Do you have any thoughts?

djconnel said...

If you assume drafting saves 30% of power, and if I'm riding Old la Honda about 43% the speed I could ride on the flats with zero-resistance tires in the same position, then I'm saving around 30% of 8% = 2.4% of power which would save 2.1% of speed or 22 seconds per 18 minutes multiplied by fraction of time drafting, assuming the other rider is going the perfect pace.