Results 1 to 16 of 16
  1. #1
    Rep Power: Infinity
    Reputation: NateHawk's Avatar
    Join Date
    Dec 2003
    Posts
    12,698

    Better elevation corrections coming?

    A new DEM (digital elevation model) worldwide dataset looks like it'll be coming online in 2014. This one will have 12m resolution, which isn't the best resolution out there right now, but that other (higher resolution) stuff is only available for relatively small areas. This one will have global coverage, so I can see it getting used by data sharing websites for elevation corrections.

    WorldDEM Coming in 2014: Most Accurate Yet Global DEM | Slashgeo.org

    The vertical accuracy is interesting. The new dataset will supposedly have 2m relative accuracy and 10m absolute accuracy. That tells me that for nearby pixels, they should be within 2m vertical of each other. And for any two pixels in the dataset, they should be within 10m of each other. That's not bad at all.

  2. #2
    FloridaKeys Fishing Guide
    Reputation: OscarW's Avatar
    Join Date
    Feb 2010
    Posts
    2,823
    The amazing thing is that with my somewhat older Furuno GPS on my boat I used to have a 5 Ft accuracy across the board when using the differential antenna that picked up on the old Loran stations( ground control) and combined this into a fix in Lat Long and altitude.
    The Milspec GPS units are the only ones not degraded by the DOD and thus have pinpoint accuracy, whereas the commercial units are for an, obvious, reason not that accurate.
    It will be interesting to see how this develops.
    Current ride(s) 2011 Santa Cruz Blur LT

  3. #3
    Rep Power: Infinity
    Reputation: NateHawk's Avatar
    Join Date
    Dec 2003
    Posts
    12,698
    Quote Originally Posted by OscarW View Post
    The amazing thing is that with my somewhat older Furuno GPS on my boat I used to have a 5 Ft accuracy across the board when using the differential antenna that picked up on the old Loran stations( ground control) and combined this into a fix in Lat Long and altitude.
    The Milspec GPS units are the only ones not degraded by the DOD and thus have pinpoint accuracy, whereas the commercial units are for an, obvious, reason not that accurate.
    It will be interesting to see how this develops.
    Huh? Selective availability hasn't been on for a dozen years.

  4. #4
    FloridaKeys Fishing Guide
    Reputation: OscarW's Avatar
    Join Date
    Feb 2010
    Posts
    2,823

    ... and if we just ...

    Yes it has, but you need something like this unit I have on my boat...
    Product Detail - GP1650DF
    Current ride(s) 2011 Santa Cruz Blur LT

  5. #5
    Rep Power: Infinity
    Reputation: NateHawk's Avatar
    Join Date
    Dec 2003
    Posts
    12,698
    I have used handhelds that picked up on the ground stations. Also have done post processing on regular gps data to correct it with ground station data. Nothing restricted about it. Just expensive.

    I am talking about elevation corrections on fitness tracking websites, anyway. A lot of fitness gps receivers don't have alitmeters and even if you do have one sometimes an incoming weather system destroys the utility of your elevation profile. Better spatial resolution and vertical accuracy on the DEM will mean for better elevation corrections

  6. #6
    mtbr member
    Reputation:
    Join Date
    May 2006
    Posts
    11,333
    Quote Originally Posted by NateHawk View Post
    I have used handhelds that picked up on the ground stations. Also have done post processing on regular gps data to correct it with ground station data. Nothing restricted about it. Just expensive.

    I am talking about elevation corrections on fitness tracking websites, anyway. A lot of fitness gps receivers don't have alitmeters and even if you do have one sometimes an incoming weather system destroys the utility of your elevation profile. Better spatial resolution and vertical accuracy on the DEM will mean for better elevation corrections
    In the mountains eleveations gain/loss can easily be out 25%....between units.

  7. #7
    FloridaKeys Fishing Guide
    Reputation: OscarW's Avatar
    Join Date
    Feb 2010
    Posts
    2,823
    My mistake as I thought GPS in general. Yes the fitness models are lacking some features in comparison to (aero) nautical units. It is all about the $$, as my Edge 800 does have a barometric altimeter built in but it was pricey too. I found it to be accurate enough for my purpose although a GPS based calculation should be more accurate.
    Current ride(s) 2011 Santa Cruz Blur LT

  8. #8
    Rep Power: Infinity
    Reputation: NateHawk's Avatar
    Join Date
    Dec 2003
    Posts
    12,698
    Quote Originally Posted by jeffscott View Post
    In the mountains eleveations gain/loss can easily be out 25%....between units.
    you're missing the point. elevation corrections applied by Garmin Connect, for example, use the currently widely available 30m DEM data.

    The 12m pixels in the new dataset cover 84% less area than the 30m pixels. That gives them more detail, for one. I don't know what the vertical accuracy of the 30m data is, but it's probably lower than the new dataset. The higher spatial resolution alone will reduce the vertical error in an elevation profile. It will allow the profile to capture smaller changes in the terrain that don't show up in 30m pixels.

    I didn't say the new data will enable removal of all error. I postulated that it might result in better corrections. Of course those corrections will be determined based on the GPS data itself, which has its own accuracy issues.

  9. #9
    mtbr member
    Reputation:
    Join Date
    May 2012
    Posts
    10
    Wow...fascinating

  10. #10
    mtbr member
    Reputation: smilinsteve's Avatar
    Join Date
    Jul 2009
    Posts
    7,151
    So, How accurate is the DEM that Topofusion uses?

    It seems pretty accurate to me. I know when I do an out and back using my phone with Back country navigator, the elevation error is obvious, as the out and the back profiles look very different. when I correct the profile using the Topofusion DEM, the out and back seem to match perfectly.

  11. #11
    Rep Power: Infinity
    Reputation: NateHawk's Avatar
    Join Date
    Dec 2003
    Posts
    12,698
    It's not so much the accuracy of the DEM. And if you recall, Topofusion's Climbing Analysis compares several different methods.

    Most widely available DEM's have 30m resolution, which is quite a bit less than 12m (square that number to see the area occupied by a single pixel). There is 10m stuff out there, but it doesn't have the coverage of the 12 dataset so it's only going to be local.

    As far as the accuracy of those datasets goes, it'll vary somewhat. And you won't know until you find out who generated it and then figure out what their standards were for generating the dataset. The accuracy will depend somewhat on the Geoid or Spheroid used as a model of the earth. A DEM generated using an older, less accurate model may have high accuracy when compared to that model. However, a new Geoid was calculated recently that's reported to be the most accurate to date, and that will change those accuracy numbers.

    The accuracy of the corrections applied to your GPS track will still depend on the accuracy of your GPS track. It's entirely possible that the accuracy of corrections applied using higher spatial resolution DEM's will actually REDUCE the accuracy of the elevation calculations.

    Why?

    When correcting with a DEM, the location matters. With 30m pixels, anything inside that pixel gets the same elevation. More often than not, GPS accuracy is better than 30m, so you'll probably get the same elevation or close to it. With the new DEM dataset at 12m, now you're flirting with the accuracy of the GPS receiver. 12m accuracy from the GPS receiver is frequently about 12m, and sometimes more. So when you do an out-and-back trail and you see a little bit of track separation, the chances will be greater that your track will fall on a different DEM pixel. It will still "look" like a mirror image, but it's possible that you'll also see slight discrepancies on your out vs. your back segments.

    Now, that will also be countered by capturing better fine scale undulations in the trail with the increased resolution, which I think will probably result in an overall improvement. But there are still likely to be times (when you have a bad day for GPS reception) that you'll get poor results.

  12. #12
    mtbr member
    Reputation: smilinsteve's Avatar
    Join Date
    Jul 2009
    Posts
    7,151
    Great post Nate.
    Here is what I wonder about. Suppose you are riding along the edge of a mesa, so if you look down, the elevation 20 feet to your right could be 300 feet lower. I think Topofusion does some kind of calculation where it looks at the average elevation from point to point and throws out values that are out of whack (for example you have points at elevations of 6125, 6130, 6120, 5820, 6120, it will throw out the 5820). Am I right about that? And for your elevation graph it will average points on either side of a point to insure there is smooth transition as you find in real life, not jagged ups and downs.

    So it seems (in my probably incorrect view) that the algorithms can pretty well correct for the DEM model inaccuracies or the GPS location inaccuracy.

    As you said, the improved DEM might not give you improved accuracy of your elevation profile.

    But going back to my out and back example, its a pretty good quick and dirty check to look at the elevation of your starting and ending points to see if they are the same. Then you can look at the shape of the elevation profile. Since I'm in Colorado, these variations are not small, so i am usually surprised to see how the Topofusion DEM correction will always show start and end to be the same elevation within maybe 10 feet or so, maybe 20, and the profiles to match almost exactly.

  13. #13
    Rep Power: Infinity
    Reputation: NateHawk's Avatar
    Join Date
    Dec 2003
    Posts
    12,698
    I wouldn't be surprised if Scott did write some kind of algorithm to find and remove outliers.

    I haven't seen any details on his method, but I do see that if you run the climbing analysis tool that you will wind up with some widely different numbers between methods so there are definitely differences between the methods.

  14. #14
    mtbr member
    Reputation: smilinsteve's Avatar
    Join Date
    Jul 2009
    Posts
    7,151
    Quote Originally Posted by NateHawk View Post
    I wouldn't be surprised if Scott did write some kind of algorithm to find and remove outliers.

    I haven't seen any details on his method, but I do see that if you run the climbing analysis tool that you will wind up with some widely different numbers between methods so there are definitely differences between the methods.
    I'm sure you will find this interesting, if you haven't seen it already:

    TopoFusion.com | GPS Mapping Software for Windows

  15. #15
    Rep Power: Infinity
    Reputation: NateHawk's Avatar
    Join Date
    Dec 2003
    Posts
    12,698
    thanks, I had not read that page before.

  16. #16
    mtbr member
    Reputation:
    Join Date
    Feb 2004
    Posts
    545
    If I remember right the original spec on STRM mission was a max vertical error of 15-16 meters absolute, generally USGS quotes it as being around 10 meters, which is probably a good average. I've read before that the "noise" level of the STRM mission was around 5 meters vertical plus or minus.

    One thing I'd be concerned about is why their absolute to relative is such a big jump, typically for other platforms of various resolutions you don't see a 5x degradation from relative to absolute, unless they are just trying to be very conservative on their absolute specs.

    I'd say it will be a potential improvement partly due to field of view resolution and party due to the increase vertical accuracy and it's big plus of course is worldwide coverage. An interesting experiment would be which is more important for use in a typical say 20 mile trail ride, having high vertical absolute accuracy or high relative accuracy.

    As Nate said you have to factor in the GPS track log positional errors, even if you have great 10 meter vertical elevations if your gps can't consistently keep your tracklog within the correct 10 meter pixel for the whole ride. I can see track log positional and "averaging" errors creating more elevation problems over a long ride than the lack of accuracy of the actual DEM elevation values. I think this is the bigger issue, until we get more accurate consumer GPS units with better tracklog capabilities which I don't see a real push for anytime soon especially with more and more people using their smart phones for their GPS needs.

    Example, lets say you ride an average of 12 mph on your ride, that means you cover ~5.8 meters a second, so if you add some random "fudge" factor you'd want a tracklog point at least every second to try and ensure you having a tracklog point in every DEM pixel horizontally, preferably every half second to try and better place you in the fringes of pixels your track might cut across. In addition to that you need a GPS that will get you at LEAST +/- 5m accuracy reliably for every track log point, again preferably better because that still can put you right at the fringes of every 10 meter pixel, preferably you'd like to see +/- 2.5 meter accuracy. As Nate said 30 meter horizontal pixels might actually do a better job because the GPS units out there right now are more likely to put you in the correct pixel. Which scenario you'd get more overall error with would be an interesting experiment.

    If you don't need global data, the NED 1/3 arc dataset (10 m FOV pixels) still provides higher vertical accuracy levels for the entire CONUS, somewhere around 2.5 meters absolute, and around 1.7 meters relative by most testing, it's typically what the USGS uses for newer 7.5 topos and they quote 7 meters vertical absolute conservatively. More and more 1/9 arc data is showing up (3 meter) but probably only about 1/5 of the CONUS is covered right now, maybe less.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •