Garmin Connect "Corrected Elevation" accuracy

donymo

I rode the Augusta bike course on Sunday and Garmin Connect reports an elevation gain of 4,400+ (uncorrected) and 1,800+ (corrected).  The Ironman website's bike elevation is 1,017ft.

So, SOMEBODY is wrong here!

What are your experiences with the elevation gains reported by Garmin Connect vs reality?

zwuba

Garmin Connect elevation correction is not that bad actually ... it usually adds a few meters but it has never been that far off whenever I used it.
TrainingPeaks elevation correction is completely off the track und also Strava (which is actually correcting all workouts automatically) is normally too high.

My favourite site to look up the REAL elevation profile is www.gpsies.com - this one gave me the most accurate results so far so I use it as my benchmark

William Jenks

The "corrected" elevation gain is almost always smaller than the "uncorrected" gain.

In theory, the corrected one should be a lot closer, but I don't know of any super-reliable data to prove this. The corrected data are based on US Geological Survey data where the elevations of the points are "known". So, it basically ignores your original elevation data and replaces it with the known elevations at each point. (At least, this is how I understand it to work. I am sure someone will correct me if I am wrong.) You can see that if this data set is sufficiently refined and accurate, you should get an essentially accurate answer.

The uncorrected data contain all the "noise" of however your Garmin is collecting the data (GPS or barometrically). Either has its flaws. If you just lay the Garmin on the sidewalk for an hour and download the data, you'll see some scatter in the elevation data, even if the net change is essentially zero. (In my experience, the barometric one can also have a long term drift, although it's usually not large.) If the elevation starts at 1000 and ends at 1000, it still might have 100 feet of climbing because the reading was bouncing up and down between 995 and 1005 for the hour that you were collecting data. In the "corrected" version, you would have essentially no climb...the only climb would be related to any small horizontal errors and whatever changes elevation the USGS thought those points had...which would be much smaller unless you were right on the edge of a cliff. :-)

The data for corrected and uncorrected that you report (factor of 2-3 different) are not out of line with what I have seen for fairly long rides. I have no idea which is more accurate: the corrected data or the IM Website report...after all, where did they get THEIR figure?

zwuba

For me the "corrected gain" was NEVER small than the "uncorrected" - never ever on every site/system I used so far.



I think the problem is that for example your Garmin Edge 500 does collect the barometric information every X seconds so it basically filters a lot of small rolling hills. If you correct a ride like that by the information William talked before you will see all these "gain" coming back -> increasing your elevation gain more or less depending on your terrain you rode.



When you to this "elevation correction" stuff in TrainingPeaks you can clearly see what I'm talking about

zwuba

For this particular ride shown above (it was last Saturday) I used my Edge500 and the weather (air pressure) was perfectly stable and I absolutely KNOW that I would not be able to ride 3350hm within less than 3,5 hours!

William Jenks

Stefan would be correct for sure if the elevation data were over-smoothed or under sampled and you were going up and down a LOT...but I'll be honest that his report surprises me. Seems like the sampling would have to be something like every 30 sec to a minute to have that kind of problem, whereas at least the old data formats where it was easy to look at the elevation showed it taken at every point.

To me, it looks like there is a heck of a lot of noise in Stefan's "corrected" elevations. Stefan, do you really think the down-spikes about half way up (and down) your mountain that peaks at 25 km and the spike at about 72 km are real? I honestly think you have crappy looking "corrected" data!

zwuba

There is no such down-spike at 25k in reality At this point there is a short decent of maybe 2-3m.
As I said the corrected data ifrom TrainingPeaks is shit ... because EVERY workout I put in there will come back with that noisy elevation profile. For example my absolutely totally flat 5k testroute will be corrected by TP to 100 meters.

William Jenks

So the point I was trying to make before is similar to what you are saying: noisy data leads to greatly increased apparent total climbs and descents, because you are climbing and descending the noise from point to point. You are correct about over-smoothing leading to data with too little total climb.

To be honest, I've not paid attention to the TP corrected elevations before, and have only used the Garmin corrected elevations. It's possible that one data set is better than the other and it's also possible that there is (for whatever reason) better data for the USA than there is for Austria.

Mark Maurer

I don't use the TP corrected data for the very reason that Stefan outlined. It's always adding way too much. If I want to justify why I feel so bad after a particular ride I do the elevation correction in TP just to see what it says.

I have seen an elevation gain while swimming using my Garmin 910. That always amuses me.