How accurate is elevation on Garmin GPS/Garmin Connect?

freehub

How accurate is it?

http://connect.garmin.com/splits/82095075

I did this ride today, and split 11 it reckons there is 0 elevation gain and whilst the route really really flat, there is frequent drags and one bit where you go uphill and when you're at the top you're probs climbed a height about 2 houses high


Any ideas? Is elevation correction set to disabled better?


Cheeers.

bompington

Don't know about this one specifically, but I've always found GPS elevations to be highly suspect: I once gained 150m on a 5km ride in northern Holland. The next day, our caravan was 50m below sea level, worrying even for the polders.

freehub

I know sometimes it is deffo wrong like, apparently once I think I did 50 miles at around 50,000ft, my average speed must have being alot higher because no airliners passed me

schweiz

Simple answer- not very!

Google GPS Height Accuracy, it will explain more about the WGS84 geoid and the way GPS height is determined.

The barometric altimeter in some edge models (iirc not the 605 and 205), smooths out the height but basically the edge calibrates itself for the local pressure and the local GPS height when it initialises. The pressure changes detected are then used to calculate the height differences with the edge recalibrating itself during the day to account for pressure changes caused by weather fronts rather than change in altitude.

It's a rough guide, but by no means accurate. My commute has a different height gain and loss for each journey and each day.

Rule74Please

garmin data (falls under rule 74 anyway) is on a barometric sytem for elevation so expect it to vary unless you calibrate it. The connect data uses the OS map or similar to set your heights on computer screen. again it is only as accurate as the data input. meant to be a guide not a substitute for riding by feel

danowat

Not very accurate, much better if you run it through an external correction program, like athe elevation correction plugin for sport tracks.

Secteur

my garmin connect website has an option for "elevation correction" - try it, it does make a difference.

meanredspider

I always disable the correction in Connect as it seems to make it worse not better (my commute goes 50m below sea-level with correction Enabled). Whenever I've climbed a Munro, I've found the elevation on my Garmin Forerunner 305 absolutely spot-on on the device itself - matches the OS map within a meter. Where it struggles more is where it doesn't have a good wide "view" of the sky and the elevation accuracy loses its resolution.

rich164h

Essentially the GPS heights will be wrong, therefore you should keep the elevation correction switched on so that the software will compare your horizontal position against a Digital Elevation Model (DEM) and pull out that as your real height. What that means is that the elevation accuracy is limited by the accuracy of the DEM that's being used and free DEMs are coarse in spatial resolution therefore also inaccurate if used for local changes in height. For example I created a global DEM as part of my PhD which had cells/pixels of ~1km x 1km. Globally that's ok and each cell would have had an accuracy of only a few metres, but clearly there can be a lot of variation within each 1km cell which means that the DEM height won't give you the correct height for specific points within the cell. For climate modelling for example it's fine though. At the other end of the spectrum, high resolution DEMs such as those produced by aircraft mounted laser scanners (sub-metre resolution) are only available for small parts of the country (normally flood plains and major urban areas) and costs thousands of pounds. It's very unlikely that Garmin has bought all of that data and even if they did they'd still be limited to the coarser datasets in the gaps.

My guess, and I say that because I'm not sure I've ever seen listed what DEM Garmin use, is that they'll be accessing the free STRM 90m NASA data with will have an average relative accuracy of ~10m at best, much worse in hilly/mountainous areas. Remember though that this is still an average height across the 90mx90m pixel so if there is a lot of variation in that cell (i.e. a road running along a steep hillside) you will still encounter the sub-pixel problem I described above.

Whilst all that means that the garmin DEM heights are likely to be inaccurate, at least the shape of the surface should be fairly well constrained so that the average over a trip will be more or less correct. The GPS elevations will appear almost random (i.e. their accuracy is probably a few hundred metres rather than a 10-20m) and essentially useless.

Hope this helps.

Rich

schweiz

meanredspider wrote:

Where it struggles more is where it doesn't have a good wide "view" of the sky and the elevation accuracy loses its resolution.

+1


however if the weather has been particularly changeable during a ride (i.e. a presure front has moved across you) then you may find a corrected altitude provides a more accurate profile but altitude correction is only as good as the altitude model that is being used. Basically these models split up the whole of the world's surface into tiles and then assign an altitude to that tile. If you are in that tile, then that is your altitude, no matter what the profile of the land is within the tile itself. The higher the resolution of the tiles (i.e. it covers a smaller area) the more accurate the model should be. However, for example, with the altitude model google use with google maps, when using bikeroutetoaster I have come across tiles that have completely the wrong value in (usually 0 or something rediculously high).

freehub

Hmmm, I think I sort of understand that.

One thing I notice is, when coming back into Manchester, I look at the elevation gain on some of my rides, with elevation correction on, it shows me going up in the city center and I know for a fact that is wrong, totally wrong, it's got like 2 drags not even that.

rich164h

I can't speak for Manchester data as I've never used it, but if Garmin are using the STRM90 data it is derived from a radar payload that was carried on the shuttle back in 2000. Radar will not be able to see through buildings, indeed urban areas are particularly poor for radar as edges of roofs etc are very strong reflectors so it's difficult to see what's going on (a bit like when you look towards an aircraft in the sky that's flying close to the sun, it's difficult to see the detail of the plane itself). It also means that the signal will be for the building roofs (which vary greatly of course) and will also be averaged for that 90x90m pixel so I'm not surprised that there is errors in that sort of place.

What is the size of alleged hills in Manchester according to Garmin?

springtide9

@rich164h : Interesting post. Thanks for sharing

Secteur wrote:

my garmin connect website has an option for "elevation correction" - try it, it does make a difference.

+1

Might also be worth exporting the GPX in Connect file and importing into something like http://ridewithgps.com and doing the elevation corrections there (and comparing)

meanredspider

schweiz wrote:

meanredspider wrote:

Where it struggles more is where it doesn't have a good wide "view" of the sky and the elevation accuracy loses its resolution.

+1


however if the weather has been particularly changeable during a ride (i.e. a presure front has moved across you) then you may find a corrected altitude provides a more accurate profile but altitude correction is only as good as the altitude model that is being used. Basically these models split up the whole of the world's surface into tiles and then assign an altitude to that tile. If you are in that tile, then that is your altitude, no matter what the profile of the land is within the tile itself. The higher the resolution of the tiles (i.e. it covers a smaller area) the more accurate the model should be. However, for example, with the altitude model google use with google maps, when using bikeroutetoaster I have come across tiles that have completely the wrong value in (usually 0 or something rediculously high).

Well, as far as I'm aware, my Garmin has no barometric correction - only GPS. I know some Garmin devices do but not my Forerunner 305.

My understanding - maybe incorrect (though my reading suggests it isn't) - is that elevation is derived from the GPS signal calculations from the satellites. Hence a "big sky" helps elevation accuacy. The correction in Connect then compares location and smoothes out accordingly.

schweiz

meanredspider wrote:

schweiz wrote:

meanredspider wrote:

Where it struggles more is where it doesn't have a good wide "view" of the sky and the elevation accuracy loses its resolution.

+1


however if the weather has been particularly changeable during a ride (i.e. a presure front has moved across you) then you may find a corrected altitude provides a more accurate profile but altitude correction is only as good as the altitude model that is being used. Basically these models split up the whole of the world's surface into tiles and then assign an altitude to that tile. If you are in that tile, then that is your altitude, no matter what the profile of the land is within the tile itself. The higher the resolution of the tiles (i.e. it covers a smaller area) the more accurate the model should be. However, for example, with the altitude model google use with google maps, when using bikeroutetoaster I have come across tiles that have completely the wrong value in (usually 0 or something rediculously high).

Well, as far as I'm aware, my Garmin has no barometric correction - only GPS. I know some Garmin devices do but not my Forerunner 305.

My understanding - maybe incorrect (though my reading suggests it isn't) - is that elevation is derived from the GPS signal calculations from the satellites. Hence a "big sky" helps elevation accuacy. The correction in Connect then compares location and smoothes out accordingly.

Sorry, I read Edge 305, which does have baro correction.

In your case the accuracy, as you say, is completely dependent on how many satellites are being received and their positions relative to you (overhead, on the horizon etc.) and the ability of the garmin to triangulate accurately.

meanredspider

And, just for interest, I've been comparing how Garmin Connect displays the vertical "shape" of the Kessock Bridge with Correction Enabled and Disabled

The profile of the bridge is much more realistic (smooth - correct profile) with Correction Disabled than it is with Correction Enabled.

The height of the peak of my ride (also on the OS map) is within a few feet both in Corrected and unCorrected modes.

My guess is that correction works well for smoother routes where the "noise" is significant. On bumpy routes like mine (where flat is a fictional notion) the noise is irrelevant.

schweiz

I'm surprised a man made bridge is in the altitude database of either google or WGS84. Seeing as you have no baro altitude, i would have though you would have just got the profile of the sea i.e. flat.

e.g. this tunnel on the route of the Alpen Challenge isn't in the database google/bikeroutetoaster use



It is downhill all the way but the altitude database uses the land profile above it

meanredspider

schweiz wrote:

I'm surprised a man made bridge is in the altitude database of either google or WGS84. Seeing as you have no baro altitude, i would have though you would have just got the profile of the sea i.e. flat.

I just have the GPS altitude measurements/triangulations which seem to be more accurate than the derived profile when "corrected".

The bridge is the "arc" towards the lefthand end of my route shown below - taking me down to almost sea-level. It's about 1km long

carl_p

Hmm not sure I understand much of that. If we are assuming Garmins do not record altitude correctly, can the same be said for distance? For example if you are riding up (or down for that matter) a hill you are effectively travelling along the hypotenuse of a triangle which we all know is the longest side. Does this mean that because a Garmin doesn't know you are riding an incline it is understating distance as well?

CarleyB

freehub wrote:

I know sometimes it is deffo wrong like, apparently once I think I did 50 miles at around 50,000ft, my average speed must have being alot higher because no airliners passed me

I'm such a child, this did make me giggle.

meanredspider

Well - from what I read, GPS altitude is 1.5x less accurate (in best conditions) than horizontal positioning. I think it's fair to say that the horizontal bit is pretty accurate most of the time (within a few meters max). Given that with even steepish gradients (in cycling at least) of say 10% - if it doubled the height travelled (ie 100% error) the distance travelled wouldn't change that much (14% error). Realistically the gradients are rarely consistently 10% and the error in altitude in my experience is rarely more than 10% and, even then, that's +/- 10% so probably cancels out.

Put another way - my Garmin measures my commute to better than 1% variation on distance. I'm happy with that.

John.T

I think Garmin Connect ignores a lot of the recorded points so you seldom get the same amount of climb and descent. The correction tab sometimes improves things but often gives lower readings. Ridewithgps.com usualy corrects to a higher reading but is seldom more than a couple of feet difference in climb and descent. They say "For the US and territories, we have elevation every 10 meters, with an accuracy of +/- 7m. This is considered accurate, and depending on if you have a barometric pressure based altimiter, may be more accurate than the uploaded data from your GPS unit.". I don't know what data they use for the UK but they do have Google mapping which is better than Bing. While not perfect the site is generaly more user friendly than Connect giving gradients etc. You can select portions of a ride for closer analysis. The only problem I have with this is the speed curve is not 'smoothed' so jumps about a bit. The site currently takes the GPS speed rather than the GSC10 one. They are working on this for the future.

magoo289

So the elevation with Garmin devices is suspect at best?

I never doubted the reading on my FR305 or Garmin Connect, until today. Always knew there was +/- tolerance in the reading. I followed a course from another user from Garmin Connect today, and every elevation reading on my device was about 30m lower than the other user's readings.

jlgt

magoo289 wrote:

I followed a course from another user from Garmin Connect today, and every elevation reading on my device was about 30m lower than the other user's readings.

Not saying that this is definitely the cause, but if every reading was off by roughly the same amount, then it could be that one of the two devices has been (perhaps wrongly) calibrated to a particluar altitude, overriding the "default" reading. If every time it was 30m out, it at least shows that the two devices are registering changes in elevation consistently.

John.T

If you are not using a pre-set 'elevation point' then you need to switch the unit on at least an hour before you start your ride for it to calibrate itself. Or calibrate it yourself if you know the height. Mine can show -30mtr at home on switch on but sets to 70mtr as soon as it finds the elevation point.

forward_loop

Haven't had much time to digest the technical stuff but:
In the very first firmware realises for my edge 500 the altitude was spot on - I mean within a metre of known levels (eg sea level up to 1600msummit then back down). I was surprised just how accurate it was. (particularly around the equator - condusive conditions/satellites ?))

At some point down the line - a couple of updates later - the accuracy went awol, by perhaps as much as 50 metres.

I also find that when I have a set a pre set elevation point - sometimes the unit wont find it and it will show as .01km away or something - my satellite accuracy shows around 6 metres

itboffin

springtide9 wrote:

@rich164h : Interesting post. Thanks for sharing

Secteur wrote:

my garmin connect website has an option for "elevation correction" - try it, it does make a difference.

+1

Might also be worth exporting the GPX in Connect file and importing into something like http://ridewithgps.com and doing the elevation corrections there (and comparing)

Spot on advice here ^

Garmin connect is such a pile of crap, you'd have to work pretty hard to create a website so slow and poorly thought out esp. given the amount of data your garmin is collecting.

That said holy crap dude how flat is that route :shock: I would kill for 30 miles with only a couple of hundred feet of ascent.

itboffin

Forward loop wrote:

Haven't had much time to digest the technical stuff but:
In the very first firmware realises for my edge 500 the altitude was spot on - I mean within a metre of known levels (eg sea level up to 1600msummit then back down). I was surprised just how accurate it was. (particularly around the equator - condusive conditions/satellites ?))

At some point down the line - a couple of updates later - the accuracy went awol, by perhaps as much as 50 metres.

I also find that when I have a set a pre set elevation point - sometimes the unit wont find it and it will show as .01km away or something - my satellite accuracy shows around 6 metres

Isn't that because the 500 & 705 models have barometric altimeter which the 605 doesn't?

kettrinboy

After loading the 2.2 firmware on my Edge 800 and after 10 consecutive rides, it is now within +/- 2-3 m of the true reading all through a ride, by true reading i mean the spot heights on the OS 50k maps which should be pretty close i assume, on versions 2.0 and 2.1 it could be as much as +/-15-20 m off, and that was after waiting for the unit to settle for a while after switch on and then setting the correct start elevation, so after 100 hours use on the 800 i am finally pretty pleased with elevation readings on it, at least now it is as good or better than my 305 which it wasnt before version 2.2.

schweiz

Carl_P wrote:

Hmm not sure I understand much of that. If we are assuming Garmins do not record altitude correctly, can the same be said for distance? For example if you are riding up (or down for that matter) a hill you are effectively travelling along the hypotenuse of a triangle which we all know is the longest side. Does this mean that because a Garmin doesn't know you are riding an incline it is understating distance as well?

It makes next to no difference.

On a 10% gradient then over a kilometer the hypotenuse distance is 1005m, which is 0.5% error, which is probably more accurate than the average car odometer.

GPS is also far more accurate in 2D position than in 3D position.

Also, if you have a GSC-10 speed/cadence sensor and use it with an Edge 705, 500 or 800 then the speed comes from that as long as you've set up the bike profile correctly.

schweiz

Today I rode home through through a thunderstorm, the front passed over me so it was a dry - wet - dry ride if that makes sense. The air pressure changes played havoc with the barometric altimeter in my 705.

Below is a 'normal' profile (200m climb, 70 m descent, 30km)



and today's profile (182m/106m)



and today's in Garmin Connect without altitude correction (223m/146m)



and today's in Garmin Connect with altitude correction (529m/388m)



So whilst the corrected profile 'looks' better the altitude values are way out!