How Accurate is TransDem?

[gearhead2578]

While taking a needed break from my Skookumchuck route, I went back to work on Tacoma (building Milwaukee's Pacific Sub). While modeling some things downtown I noticed the hill doesn't seem steep enough - How accurate is the elevation in transdem? Where does the program get it's information?


Transdem rendering looking down from 23rd & Yakima toward BNSF yard toward tide flats


gearhead2578_20120109_0000 by welder3078, on Flickr

Pic taken by me from the same place (out my old living room window in 2005)


HPIM0154 by welder3078, on Flickr

The hill is steeper in real life - enough that you can't drive up it in winter until they sand the crap out of it

 

[stovepipe]

To keep the amount of sampling data sensible, there is an inteval of 10's of meters between sample points. This means the very top of hills and the very bottom of valleys are often missed, reducing the height difference and hence steepness of a slope.

On a DEM I'm working on the max height of a particular hill, was 15m less than it should have been.

 

[Rail4Pete]

Transdem just convert the data you feed it with. So it's the data source you have to accuse whether it is accurate or not. If you look at an orange from 10 feet distance, you will not notice the small irregularities on its surface, but if you are at 1 feet distance you will see much more details. This is how the data accuracy depend from....the distance it has been taken.

Rail4Pete

 

[gearhead2578]

Ah that does shed some light - I was under the impression that trandem was much more accurate - but even a 15m difference isn't much. What I'm looking at is the difference between 5% and 12%+ gradient on that hill. Just as an example, when I was in my 20's I used to get a kick out of jumping my car over the cross streets on the way up.... I don't miss the jail time

A few examples of the hills


tacoma139 by welder3078, on Flickr


tacoma045 by welder3078, on Flickr


tacoma087 by welder3078, on Flickr

Seems waaay off to me

 

[RRSignal]

There are NED sets available from USGS seamless server that are accurate to 1/9 degree arc. I think that corresponds to a resolution of 10 feet. Unfortunately, they are very much incomplete, whereas I think 1/3 degree arc (30-foot sample intervals) cover the conterminous U.S. There are maps of coverages at the seamless server, but I've found holes in even the stated 1/9 coverage so beware.

Another thing to be aware of is that DEMs can be huge in terms of memory. My current SEPTA R3 DEM came in at about 250Mb after some careful trimming; I couldn't get much from the 1/9 NEDs but it looks like that would have been well over 1Gb had I been able to get the data. Mind you, this is for what's basically a pair of 30-mile-long routes separated by a few miles at their farthest points.

 

[Dermmy]

There is no issue whatsoever with TransDEM and it's implied impugning simply illustrates total ignorance of the process used. TransDEM implements the best data available. If that's not good enough pretty sure you can guess the rest...

 

[okhiu]

There is no issue whatsoever with TransDEM and it's implied impugning simply illustrates total ignorance of the process used. TransDEM implements the best data available. If that's not good enough pretty sure you can guess the rest...

And what would be the best data available for Europe? I tried to create a DEM of a Romanian rail line, a defile, that is very very narrow.

( here's a movie from there http://www.youtube.com/watch?v=fndgEdKpPtQ )

I used some hgt files from here http://dds.cr.usgs.gov/srtm/version2_1/SRTM3/Eurasia/ but the elevation is not very well made, i mean there is a big mix between the river (which is at the lowest point), the road and the rail track. And also there are big height differences between close areas...I have the rail at +220 in one place and in 200-300m the same rail is at about +190......I suppose this is the fault of those .hgt files ... if so, do you know any better source ?

 

[RRSignal]

And what would be the best data available for Europe? I tried to create a DEM of a Romanian rail line, a defile, that is very very narrow.

( here's a movie from there http://www.youtube.com/watch?v=fndgEdKpPtQ )

I used some hgt files from here http://dds.cr.usgs.gov/srtm/version2_1/SRTM3/Eurasia/ but the elevation is not very well made, i mean there is a big mix between the river (which is at the lowest point), the road and the rail track. And also there are big height differences between close areas...I have the rail at +220 in one place and in 200-300m the same rail is at about +190......I suppose this is the fault of those .hgt files ... if so, do you know any better source ?

You may want to poke around the Transdem forums located here. The Resources forum should have the info you're looking for. I believe SRTM only provides 3 degree resolution (90 meters) so you might be able to find something better.

 

[backyard]

Look Here...

TransDEM, Trainz Edition v2.2 now available.

 

[steamboateng]

The USGS NED site lists resolutions as:
1 arc-sec data = 30 meter vertical resuution
1/3 arc-sec data = 10 meter vertical resolution
1/9 arc-sec data = 3 meter vertical resolution

I just checked elevation data in Surveyor route recently installed from TransDEM maps, in all cases the Surveyor map was within 10-12 feet of the noted elevations on the usgs quad map!

 

[JCitron]

The USGS NED site lists resolutions as:
1 arc-sec data = 30 meter vertical resuution
1/3 arc-sec data = 10 meter vertical resolution
1/9 arc-sec data = 3 meter vertical resolution

I just checked elevation data in Surveyor route recently installed from TransDEM maps, in all cases the Surveyor map was within 10-12 feet of the noted elevations on the usgs quad map!

I think too that a lot of clutter doesn't help much either. By clutter I mean buildings and such that actually distort the height of the ground as seen by the satellite. I ran into this issue with Downtown Haverhill. The rail line is about 6 meters above the road on a bridge at one point, but according to the DEM, they are at grade. There's not even a bump between the road below and the tracks above.

So my thinking is the buildings reflected back the surface and altered the height information sent to the satellight or high-flying aircraft capturing the information.

John

 

[gearhead2578]

I think too that a lot of clutter doesn't help much either. By clutter I mean buildings and such that actually distort the height of the ground as seen by the satellite. I ran into this issue with Downtown Haverhill. The rail line is about 6 meters above the road on a bridge at one point, but according to the DEM, they are at grade. There's not even a bump between the road below and the tracks above.

So my thinking is the buildings reflected back the surface and altered the height information sent to the satellight or high-flying aircraft capturing the information.

John

Probably so I didn't even think of that. Wow guys-the (intelligent)response to this one is outstanding
I don't think I've ever been able to light up the forum like that before. Thanks all for the input - quite a bit of good info. I think the best solution for this one though is just going to be to hand-tweak the elevation. I ran around in that city off and on all my life so it shouldn't be too bad to remember how the ground lies even though I'm 1500 miles away now.

- And yes Dermmy, I was ignorant to the process used therefore was slightly impugning the software as I had figured it was a bit more exact (like maybe within 100' elev. but in this case it isn't). Most of the point of this thread was just to point out that you just don't know how far off it is until you go to model an area that you're extremely intimate with and could remember nearly every pothole, gritty alley, building, etc.

 

[Vern]

I have always found Transdem to draw a pretty good representation of the terrain from the DEM data. What doesn't always match up though is the mapping, whether scaling raster maps or using one of the built in tile servers. Not unusual to find these off by 150 - 200 metres in either plane, so you have rivers on the side of hills etc. Roland (Geophil) and I have discussed this a bit, just seems to be inconsistency between the various datasets - as I've encountered the same issue with DEM in Railworks and markers plotted in Google Earth.

 

[steamboateng]

I think i'll have to disagree with some of you on this one. I have checked the data from Trainz terrain maps installed, against the usgs maps installed as UTM objects. In all cases the the the installed DEM matches the corresponding usgs map markers within 10 or so feet (vertical) on a !/3 arc sec generated map. I agree that transitory slope data may vary somewhat; but I have found that valley lows and hill top highs are in close correspondence with usgs data. Please note that I'm matching data directly to underlying map data. Also note that I'm matching data from the same source.
Accurate georeferenced maps are the key to to determining how accurate the DEM data is. I have noted that the DEM may be off 12 or 15 feet in the 'x' or 'y' location, But that is a factor I can deal with, and modify accordingly. My maps aren't always the most perfect in georeference placement, considering the scale involved!
My conclusion is that TransDEM is a fairly accurate program, subject to the data it's fed.

Regards to all

Edit:Actally the DEM plaement is more accurate than the georeferenced map!

 

[geophil]

TransDEM itself maintains an overall accuracy of 1 m internally. However, as the others have written, it all depends on the geo data you feed it with.

Now, coming back to the first posting in this thread. As it seems, this is not TransDEM, but MicroDEM/HOG. And here, things are slightly different. The MicroDEM/HOG process, most often used in combination with TIGER vector data, as in the example, has a number of limitations.

1. TIGER data. Per definition, this type of vector data is only 1:160,000, good enough for the US census, which is the original purpose, but quite often not good enough for route building in Trainz. It's a less than ideal horizontal accuracy due to relatively low scale. Rivers flowing uphill are one of the possible consequences.
2. Terracing, caused by image transfer. Normally, elevation data is transferred from MicroDEM to HOG using a colour-band raster image. Each colour represents one elevation value. The number of colours in such an image is very limited. The typical image type can only represent 128 different colours, meaning only 128 different elevations. The result on steeper terrain will look like rice paddies on a mountain slope.
3. Smoothing. HOG applies a low-pass filter to the terrain it creates. While some sort of terrain smoothing will often be necessary, the one in HOG is quite excessive, possibly to compensate for terracing effects. This means vertical detail will suffer.

For clarification: This is specific to the MicroDEM/HOG process and has nothing to do with TransDEM.

 

[steamboateng]

I would like to add, that the dem data processed with TransDEM was taken within the last several weeks. The usgs maps I used to correlate the data were 60 to 70 years old!
10 to 12 ft. error........
C'mon man................!

 

[Vern]

Yeah I think the USGS maps probably get quite close. Where I've found issues is if you take data from Google Maps or Open Street Map there is often up to 150m discrepancy. Another candidate was the old Poehali Russian topo maps, particularly when hitting the lower resolution sets - 1 to 100,000.

However there are tools in Transdem to recover from this by shifting the DEM, takes a bit of trial and error and a couple of route re-exports to get about right, after which the imaginarium takes over!

 

[geophil]

I would like to add, that the dem data processed with TransDEM was taken within the last several weeks. The usgs maps I used to correlate the data were 60 to 70 years old!
10 to 12 ft. error........
C'mon man................!

In October I played around with Prussian cadastre coordinates. The system was introduced in the 1870s and in use with Prussian State Railways, German State Railways after 1924 (Deutsche Reichsbahn Gesesellschaft) and early Deutsche Bundesbahn. It took me a while to figure out the details of this system. It's a Cassini-Soldner projection and it had 40 "zones", all based on what later became "Potsdam Datum". And I had this old 1:1000 track plan, apparently with such coordinates. As I wrote in the TransDEM forum, once georeferenced it did fit to the modern 1:5000 map and ortho-photos.

Yes, our forefathers did a good job, even without the aid of GPS gadgets. Map making history says, modern surveying started in France during the Age of Enlightenment, and the Cassinis producing the "Carte de Cassini", the first one based on triangulation.

 

[steamboateng]

geophil, your a historian at heart; you love those old maps...............

 

[backyard]

Yes, he does...

Another factor to think about is that even if the elevation is correct, most "hills" (under 4000ft-elev.) have a tree-line over the top, accenting the mind's concept of the grade profile.

Therefore, 10-meter elevation being within ten-meter would be a mute point after trees are installed...