Elevation data... Which is the most accurate?

Last weekend, RoadSkater & I were talking about elevation data while on the scenic 51-mile "Hannah" route...  I unfortunately lost my data from that skate due to a hard drive crash, so I can't compare against his.  But it did get me thinking...

Which elevation data is most accurate?

1) Polar's barometric sensor

2) Garmin's GPS sensor

3) NASA's 30m elevation data (i.e., on gpsvisualizer.com)

My initial thought regarding the NASA data was that it didn't sound very "granular".  However, I did some rough calculations the other day, and just now put some numbers into a calculator and came up with this...

If your watch/gps is set to record data every 5 seconds, then if you're travelling at 6 m/s then you're capturing elevation data at the same granularity of the NASA data.  6 m/s = 13.4 mph.  So if you're going slower than 13.4 mph (and sampling every 5 seconds) then you are (in theory) capturing data at a higher resolution than NASA.

You can increase the resolution by taking more samples.  My Polar HRM can record every 1s, 2s, 5s, 15s or 30s.  So in my case, I set it to sample every 2 seconds on my last skate.  That means I can achieve a higher resolution than the NASA data if I'm skating less than 15 m/s, or 33 mph.  

That assumes, of course, that the whatever sensor you're using (barometric or gps) is as accurate at measuring elevation as whatever method NASA used.

With all that in mind, I graphed my elevation data in Polar's software (which uses the barometric sensor) and gpsvisualizer.com.  Polar doesn't save the elevation data in the gpx file, so the only way to draw an elevation profile on gpsvisualizer.com is to use the NASA data.  The results were mildly surprising...

Polar (barometric)	gpsvisualizer (NASA)

Looking at the charts, the NASA data appears to be more granular than the data I captured, despite the fact that I was capturing many more data points.  My average speed was 13.9 mph and I took samples every 2 seconds.  So that, in theory, works out to sampling every 12 meters - more than double that of the 30m NASA data.

So either the NASA data is artificially "noisy", or the Polar data is "smoothed".  If I had to guess, I'd say the barometric sensor just isn't all that accurate, so it can't pick up the slight changes in elevation when you're sampling more frequently.  But the elevation profile looks roughly the same as the one generated on gpsvisualizer.com, so it's probably "good enough".

I wish I had data to compare from a Garmin unit.  Even a comparison of the captured elevation data versus the NASA data would be useful.  I also wonder what options the Garmin units have for sampling rates...

- SM -