GPS altitude measurments

[Eric Jeppsen]

Last week, Garmin updated the online Connect website. At first, I thought the change was only cosmetic (layout, etc.), but today I noticed that the updated site uses an elevation correction feature. In Garmin's words:

"Elevation Corrections cross reference the horizontal position (latitude/longitude) provided by the GPS with elevation data that has been acquired by professional surveys. When corrections to elevation data are made, each trackpoint of your activity now contains the elevation from the web service, not the elevation provided by your GPS device.

"Garmin Connect selectively applies corrections to depict a more realistic representation of your elevation experience. Activities recorded from devices without a barometric altimeter are enabled with Elevation Corrections by default. Alternatively, activities recorded by devices with a barometric altimeter generally contain accurate elevation data and therefore Elevation Corrections are disabled by default. For those users who are familiar with the MotionBased Gravity service, this is the same service."

What I'd like to know is whether the corrected or uncorrected altitudes are more accurate.

Two weeks ago, I ran my usual climb out of City Creek twice in a row. According to my old and uncorrected GPS report, I gained 689 feet on the first ascent, lost 579 feet on the first descent, gained 606 feet on the second ascent, and lost 612 feet on the second descent. (Those measurements come from an out and back on the same singletrack, so clearly at least three are inaccurate.)

This morning I ran the same climb once. The new and corrected GPS report showed 764 feet on the ascent and 760 feet on the descent (uncorrected, 681 on the ascent and 675 on the descent). The corrected results are more consistent than those on the old report (although, interestingly, so are the uncorrected results). But what really caught my attention is that the corrected results add 100 feet of climbing over 1.2 miles.

Are there any GPS or topography geeks (and I use the term affectionately  ) who can shed some light on GPS accuracy and tell me which measurements I should trust? Obviously, I'd like to think that my climb is 760 feet, but I don't want to pad my numbers, either.

[Paul Petersen]

I'm not a GPS geek, but am a topography geek. The web service data they use to correct with will be more accurate, but less precise. It's likely the USGS 10-m or 30-m elevation model. In general, any elevation data you get from a small garmin device will be horrendous, and mostly meaningless.

[Eric Jeppsen]

Thanks, Paul. Could you elaborate on the USGS elevation models?

And by your response, do you mean that both reports (corrected and uncorrected) are mostly meaningless?

[Paul Petersen]

Actually, I didn't read your email closely enough. Apparently they are using some other web service for the "correction". I suppose the "corrected" is more accurate. After all, why would they correct to something less accurate?

But my earlier statement is still true: any elevation and derived elevation stats you get from a garmin running device will not be very good. That's just the nature of a handheld, rec-grade unit.

[Steve P]

Paul, do you have a ballpark idea of how inaccurate it is? I've noticed that the altitude on my Garmin varies a lot when I'm going up or down a hill, but overall it seems to be in the right ballpark. I'm just wondering if we're talking that it might be 10 ft off at a given point in time? 25 ft? 100 ft?

[Joe]

I have a bit of experience with GPS accuracy from work and I would agree that you want to trust the map data more, even though it's probably not coming from the best DTED.  I'm guessing that's what Paul means by less precise.  I don't know what exactly the web service would be using but basically the elevation they'd use for a given point is actually an average in a sample for a small square of land within which that given point is contained, if that makes any sense.  GPS is awesome for horizontal accuracy (even crappy home versions) but not great for vertical/elevation - I'd say anywhere from 2-4 times worse than horizontal accuracy (ballparking that from experience).  I have the 405 so I don't know if all Garmins have this reading but if your Garmin says ~20 ft accuracy (as mine usually does when it has had time to acquire the optimal satellites), you could expect 40-80 ft accuracy for elevation.

[Paul Petersen]

I agree with Joe. For recreational GPS units, horizontal accuracy is usually within 15m, or within 3m if WAAS enabled. Vertical accuracy (z-values) will be several times that amount. If you ever watch your elevation reading, you will see it jump all around. Yes, it gives you a precise reading (within a foot, or tenth of a foot), but is very inaccurate. The digital elevation model, which does some averaging, is more accurate, but is less precise. DEM's will tend to cut of the top of a peak or fill in a valley, but at least won't give you really weird values like a GPS (assuming you did your mapping correctly).

[Adele Kimbrough]

I don't have a garmin, but have found this discussion interesting because I have been wondering how accurate garmins are at measuring distances when when running up/down significant hills. 

For instance, if you are on flat ground, I would think a garmin just takes the x-y coordinates periodically to calculate the distance.  But if you are going up an incline, I think the garmin would have to calculate the distance based on the x-y coordinates and the difference in elevation.  i.e., from Pythagorean's Theorem: (actual distance)squared = (horizontal distance)squared + (vertical difference)squared.

Is this how garmins calculate distances? If so, then if the garmin uses an incorrect elevation gain/drop, then the "calculated"  actual distance travelled would also be off. 

Also interesting is how the elevation is taken from the average of a smaller square of land.  If you are going up/down a big paved road, it would probably be pretty accurate.  But if you are going up/down steep switchbacks, or similarly un-uniform terrain, the recorded gain/drop in elevation could be off substantially?

[dave rockness]

I'm way in over my head in "technical talk", but have found my garmin to be very close to accurate on very hilly, certified race courses.  My most recent race was a 10k with 7-8 hills and my garmin measured the course at 6.25 miles (which probably accounts for my poor "tangenting" skills).  Just over a month ago my half-marathon was within .05 of nailing the distance as well (Garmin 305).  It my not be perfect on hills, yet I'm finding it to be pretty close and a very helpful tool. 

[Calvin]

Just from an observational point of view, the Garmins seem to do a remarkable job in measuring distances, (especially given the published 15m accuracy rating.)  I tend to repeat the same courses on a regular basis and the measured distance usually comes within plus/minus .02 miles over say a 7.5 mile course.  Not good enough to be certifying races, but a darn good tool for distance training.

Additionally i've used my Garmin 201 and 205 to do a fair amount of geocaching.  Typically following the GPS coordinates gets me to within 10 feet of a cache.

Just one more note about elevation changes and the pythagorean theorem.  On a 6% grade, (which is about a 316 ft climb over 1 mile) the difference between the hypotenuse (inclined distance) and the uncorrected horizontal distance is only about 10 feet.  Probably not a significant source of error in the distance measurements when compared to the inherent x-y accuracy rating of the GPS unit.  (SQRT((5280)^2-(316)^2)=5270)

[Paul Petersen]

Calvin -
I agree. I run with a garmin 205 almost every day. It's pretty consistent measuring the same course, and I find it a valuable training aid. At least it errs wrong the same way every time. I do not care if I'm a few hundredths of a mile off on my run distance.

Regarding the effects of z-change on planimetric measurement, I've done the math as well, and got very similar results. The difference between the hypotenuse (elevation-affected distance) and the planimetric distance is basically nothing. I once figured out how much "extra" I ran on the St George Marathon course, and it pretty much zip. I suppose over a 100-mile ultra race, a 6% grade could add on 1000' extra? Big whoop. But if you're doing 100-miles at 6%, an extra 0.19 miles is the least of your problems!

A few other notes of things that will mess up your garmin distance:
*taking sharp corners: always causes mine to overestimate distance and speed. You would lose a ton of accuracy over switchbacks. My garmin is pretty accurate over straight-line distance, but poor on curves.
*running in canyons.
*running in forests
*dropping the garmin in a river.

Happy running. Don't be too much of a tech geek!

[Dale]

Most current civilian GPS receivers have an altitude error specification that's about 1.5 times the horizontal error specification, or about ±23 meters (75 feet) 95% of the time.  You can find pilots that'll use GPS to navigate pretty much anywhere, but I doubt you'll find one that'll use GPS altitude measurements to conduct their approach to landing in instrument flight conditions.
Here's a good primer on GPS and runners: http://members.cox.net/sd_run_maps/gpsprimer.htm.  
I did checkout the Garmin Connect website corrected versus uncorrected altitude graphs and must admit they are pretty good, much better than standard GPS altitude graphs.  It even took the marathon I just ran and projects an altitude graph pretty dang close to official elevation profile on their website.  Even TCXConverter and the CourseTool had issues coming up with the corrected altitude, probably because the road it's run on cuts thru a canyon with steep sides....at some points I was running nearly under the rock overhangs...so the elevation graphs were very skewed.  The Garmin Connect measurement was much saner.