Sandra Vi going for record across America

scam_watcheroo wrote:

So far today wrote:

One is that, for Sandy and Mimi, it almost always measured longer than a straight down the road path measured by Google.

Google directions will typically be shorter because it uses a path finding algorithm that can assume perfectly straight lines. Even if we exclude the side stops a bit beside the road, in reality, a GPS track measured by a watch will give a longer distance because of the uncertainty in exact position from the GPS that accumulates as we add up the distance between each pair of GPS coordinates. This is evident by zooming into what looks like a straight line on the GPS track and you will notice it actually wobbles back and forth a few meters throughout the path.

Except that the close up of the Strava path is not a real route. Sandy's final mile at City Hall is enough to show that running through fences and buildings is not possible. In one of the wobbly images it shows she jumped into the Hudson River. Mimi swam across a river just off the a bridge she crossed. Anyway, the Strava curve looks very detailed but the points are not exactly where you actually run. There are a lot of these wobblies in both Sandy's and Mimi's data sets, and they may be in other Strava data sets as well.

So far today wrote:

Scam, you might be interested in two things I noticed about the Strava data. One is that, for Sandy and Mimi, it almost always measured longer than a straight down the road path measured by Google. That is not so odd on it's own because they both measure distance differently, but Sandy's final run route measured 133 miles per Google while her Garmin recorded 139.6 miles.

Part of the difference in miles might be something I am calling data wobblies. The extra miles are partly due these wobblies.

https://imgur.com/a/GYavx

That is exactly that I was thinking.

GPS running data is highly incorrect because every data point can be +/_ so many feet away from the actually ran course, which creates a very wobbly course. You can try that yourself running on a track and look at the Strava map.

133 google miles against 139.6 Strava miles is a 5% difference. Actually a lot more than I thought it would be.

So far today 2 wrote:

Except that the close up of the Strava path is not a real route. Sandy's final mile at City Hall is enough to show that running through fences and buildings is not possible. In one of the wobbly images it shows she jumped into the Hudson River. Mimi swam across a river just off the a bridge she crossed. Anyway, the Strava curve looks very detailed but the points are not exactly where you actually run. There are a lot of these wobblies in both Sandy's and Mimi's data sets, and they may be in other Strava data sets as well.

It also depends on what type of filtering is applied to the raw GPS coordinates data before using it for distance or speed calculations. Strava's smoothing algorithm is proprietary so we will know what they do exactly.

For my analysis files, I just leave the GPS coordinates where they are and smooth the distances between successive points with a 10 point rolling average before using that for speed and stride length calculations. Makes thing very simple and is easy for everyone to understand. I get about 138 miles using this method instead of the numbers you have, so in between the two extremes.

If I wanted to step up the smoothing to something more advanced, I also have a low pass filter available designed specifically to smooth out noise in GPS coordinates but requires more parameters to be specified. It all builds down to whether you can get rid of noise without also risking losing real features in the GPS coordinates data.

She took a wrong turn or two (or crossed streets multiple times in high turn areas, etc) that likely added 2-3 miles over the duration of the final push in populated areas, coupled with the circling of City Hall at the end. She seemed to be about one mile high each day, or about 2% in general. Pete K was right on Google distance for the most part at 3067, probably because of a less sensitive Garmin 225

Markus wrote:

That is exactly that I was thinking.

GPS running data is highly incorrect because every data point can be +/_ so many feet away from the actually ran course, which creates a very wobbly course. You can try that yourself running on a track and look at the Strava map.

133 google miles against 139.6 Strava miles is a 5% difference. Actually a lot more than I thought it would be.

Google map data is not exact. This can be seen by comparing satellite imagery to what Google draws down. Very often they don't align and Google measurements are not perfect. So it's not really a 5% error. The real noise free mileage is likely in between what Google says and what the GPS watch says.

As another comparison point, Pete's mileage for PA+NJ/NY, per his Garmin data that he provided us, was 328.6 + 88.2 = 416.8. The Google measure smooth-down-the-road route was 413.7 miles. That is 0.8% greater which makes more sense. If you look back at all the daily splits I provided you will notice that nearly every day Sandy's Strava miles were longer by 1-2%.

My best guess is that there is a difference between how Pete and Sandy calibrated their Garmins. It also might be that the Sandy's newer Garmin tries to provide better path accuracy by estimating what happens between data points. But, as you note Markus, try running a known marked course on a trail and then look at the route your GPS says you took. Sandy and Mimi evidently both jumped into rivers and ran through buildings and ran across traffic lanes according to the fine-level Strava.

scam_watcheroo wrote:

It also depends on what type of filtering is applied to the raw GPS coordinates data before using it for distance or speed calculations

It might even be estimating rather than filtering. Filling in gaps to make a smoother looking run path. But, as Markus notes, the accuracy of the actual GPS points has some error so that actual run path will always have those errors.

To me it seems using a measured course, even if it is Google Maps, at least provides something to compare against. You can take all the side trips you want during a marathon but the course is 26.2 miles and not the 26.2+ miles you likely ran according to your GPS.

finish track wrote:

Pete K was right on Google distance for the most part at 3067, probably because of a less sensitive Garmin 225

Maybe even because of an attempt to increase accuracy in the newer Garmins. Or how each Garmin was calibrated.

Pete K said he used his Garmin from last year and was planning on comparing his results to Sandy's on the day he ran with her. That comparison would be interesting--two different Garmins along the same path; well maybe not exactly the same path because pit stops might be different.

trying to help wrote:

Well, first I'm sure many today can relate, sometimes in their life, ....

^^^I endorse the above message (after all, I wrote it). :-)

Here is another video Zwitty linked to earlier in the thread, for those of us that couldn't seem to get enough of this reality-tv drama:

Thank you for that raw footage Zwitty, and for the person that filmed it. If there are any others available, please post!

Here is an article on Transcon Record Holder, Pete Kostelnick:

https://ultrarunning.com/features/interviews/damn-son-interview-with-transcon-record-holder-pete-kostelnick/

Assuming then, Pete was cranking 75 miles/day in his Transcon, and running 12.5 hours per day, that is a 10min/mile pace, on average, including any short breaks during the 12.5 hour running period.

So scam_watcheroo, would it not be interesting to do an analyses of Pete's Transcon data, similar to what you performed on Sandy's?

There is a lot of interesting discussion here about GPS trackers. They are not as accurate as some of us think they are. My GPS often thinks I cut the corner to shorten my run. I know I didn’t that.

this is amazing, no? wrote:

So scam_watcheroo, would it not be interesting to do an analyses of Pete's Transcon data, similar to what you performed on Sandy's?

There is another runner lady planning to do a transcon across the US in a few months? The splits and summaries and data analysis were great. I hope the thread regulars do the same again.

How about transcons in other countries? Are others planned?

When I was using a Garmin Forerunner 305 watch years ago, I thought it did a great job for my needs. Sure, using it at a track, and it would be a little funky on results (though still not too bad), and performance improved once you got away from running in circles on the track. That was well over five years ago; hard to imagine they have actually gotten worse....

Maybe trackers wrote:

There is another runner lady planning to do a transcon across the US in a few months? The splits and summaries and data analysis were great. I hope the thread regulars do the same again.

I was completely out-of-the-know, unaware when Pete Kostelnick was doing his Transcon. I must have tuned out any thread on this forum, after tuning out the R. Young fiasco. I'm starting to digest another article on Pete's that I just came across:

this is amazing, no? wrote:

So scam_watcheroo, would it not be interesting to do an analyses of Pete's Transcon data, similar to what you performed on Sandy's?

I've download Pete Kostelnick's Strava data but there is a problem with where Strava does not provide a starting time for each run like I could grab for Sandra. For each day, there are 2 or 3 runs and they all appear to start at the exact same time because there is no start time information so that really screws my analysis tool up when it comes to adding together the durations that I show in my graphs and summaries. Everything else works in my analysis tool, it's just the total run durations that are wrong because of this. I will have to create a workaround for that in the near future to handle other Strava accounts that don't show starting times.

For now, at least the summary graph showing how often Pete goes through a run-walk cycle is accurate:

The yellow line shows that Pete's average walk/rest duration is 2-5 minutes for each run-walk cycle. The blue line shows that Pete's average run duration is mostly between 10-21 minutes for each run-walk cycle depending on which day we are looking at. The black line shows Pete transitions between running and walking from 30-115 times in a day.

This means that on average, Pete spends more time running in each run-walk cycle and transitions much less between running and walking than Sandra did, which is the expected result given their speed differences.

You can see all the other detailed per run graphs (for cadence, speed, and digits distribution) for Pete here (upload in progress):

Wow, SW, amazing you crunched that so fast!

I came across this video of the Pete Kostelnick finish, as recorded by Charlie Engle:

https://www.facebook.com/charlie.engle.12/videos/713881452095749/

I haven't gotten completely through the video, but I haven't seen any 'wipeout' by Pete mentioned earlier in this thread.

"Pete's average walk/rest duration is 2-5 minutes for each run-walk cycle...Pete's average run duration is mostly between 10-21 minutes for each run-walk cycle

^^^Wow, this is a very interesting BULLET POINT. So he runs 15 min, walks 2 min while scarfing food down? Is that pretty much how these things go? Or does he interject a walk, for energy conservation, whether he gets any extracurricular activity (eating, elimination, etc.) done during the brief break, or not?

Sandy will be at NYC City Hall steps at noon today for photo ops.

If you're in the area go say hi!

this is amazing, no? wrote:

Here is another video Zwitty linked to earlier in the thread, for those of us that couldn't seem to get enough of this reality-tv drama:

https://www.youtube.com/watch?v=UzLFK2gbjgM&feature=youtu.be

Nice video of finish. Thanks for reposting.