Laying track on a DEM?

[steamboateng]

From you I'll take it as a compliment. Thanks!
Regards

 

[JCitron]

I agree with Andy. Mike this is one of the best threads we've had here in some time. It reminds me of the old days on the forums.

I too have had moments like yours in areas such as Newburyport where the drawbridge track lead separates from the street. I spent more time fussing and fiddling to get the right height there because of the height issue causing Merrimac Street to blend into the trackbed. This is the same issue I described down in Bradford and Haverhill.

My route-chunk isn't quite as extensive, but I too have had none of the geo=referencing issues as described either. It's amazing how the topographic maps plop right down on the DEM. I was so excited seeing the hills come to life along the Merrimack River between Haverhill and Bradford. I can see the streets as they climb Silver Hill and up near me on Golden Hill.

I too can't wait to see what Dr. Z has to say on this.

John

 

[steamboateng]

Thank you, sir.
I often quit the forums for a spell. Sometimes to avoid conflict; sometimes because the cheif mate demands a life too; sometimes because I get into other stuff!.
I always come back.......been doing it nearly for 10 years, now.
Mostly, simply because I like what most of you like..........
Trainz,
I am honored to be among you.........in a very simple way.

 

[geophil]

I'll try to explain what kind of error sources we are dealing with, geo data, TransDEM geo transformations, TransDEM export to Trainz and smoothing, and the human user of course, possibly this evening when I have a bit more time. But discrepancies between a 1/3 arc USGS NED DEM and the 1:24k USGS topo map in the 20m range are quite probably down to human error of some kind. Particularly the older topo maps are NAD27, not NAD83/WGS84. That has to be observed when georeferencing the sheets.

 

[steamboateng]

Blutorse4792, please forgive us for grabbing onto your thread to get some TransDEM issues resolved. The isues are relevant to your question and an open discussion may benefit us all.

 

[geophil]

Potential sources of error in the TransDEM process:

• Geo data
  • DEM Accuracy:
    • Vertical: Depending on source (orbital/terrestrial) and horizontal resolution. Satellite DEMs most often do not filter out buildings or forests. Terrestrial NED DEMs include earthworks and dams, but no construction above ground and no forests. Note: not all nominal 1/3 arc sec NED DEM are truly 1/3 arc sec. Some positive example I remember: Truckee (Donner Pass), Columbia River at Bonneville Dam. Negative examples were Royal Gorge or the Virginia and Truckee RR. The latter look like resampled 1 arc sec data.
    • Horizontal: No error noticed with NED DEMs. Some with Viewfinder Panoramas or first generation ASTER (long before GDEM).
  • Topo maps:
    • Feature detail resolution: Due to map scale most features on the map are drawn wider than in nature, e.g. the second track of a double track railroad may be too far away from the first track. 1:24,000 is too small to show individual tracks in full precision.
    • Feature accuracy: Not all map features are accurate. But topo maps issued by state agencies like the USGS usually are quite good. They often serve as official reference.
• Georeferencing in TransDEM. Internally, TransDEM is always UTM/WGS84. All geo data not in UTM/WGS84 will have to be reprojected. TransDEM provides all the formulas.
  • Map Projections: The formulas used in TransDEM give1 mm mathematical precision. Which means reprojection does not lose noticeable detail by the algorithm itself. Naturally, resampling will introduce some "blur" which can normally be neglected. However, each projection is only as good as its parameters. In practice, TransDEM aims for 1m accuracy which is usually far better than the geo data itself. There were a few issues with some coordinate system in TransDEM in earlier versions, one of the French and Russian systems was slightly wrong.
  • Georeferencing process: For automatic georeferencing, WMS or Map Tiles, we have to rely on the provider. For manual georeferencing it's all down to the correctness of the projection specified and of the reference points themselves. For slightly distorted maps, TransDEM 2.3 introduced extended georeferencing with a network of triangles. This may help with certain scanned paper maps but is more complicated. Generally, the biggest source of error is the human user here, making mistakes in the manual georeferencing process.
• TransDEM export to Trainz: TransDEM maps its internal UTM/WGS84 system 1:1 to Trainz world coordinates. Both systems are Cartesian metric systems.
  • Interpolation: Trainz terrain is defined by ground vertices in a 10 or 5 m raster. DEM and map data usually has a different raster. Export is done by bilinear interpolation (same algorithm as used with reprojection/resampling).
  • Smoothing: TransDEM offers a Gaussian low-pass filter which will smooth the DEM slightly to prevent terracing with lower resolution DEMs. It can be switched off by setting it to 0. When switched off, all elevation values remain untouched and should be identical in Trainz and in TransDEM. Smoothing, if active, does not shift any data sideways, but it does indeed raise or lower individual points, reducing the vertical difference to its neighbours.

To track down georeferencing errors, compare different geo data sources, topo map, one or more map tile providers, aerial images, even other DEM sources (SRTM/ASTER). If there are discrepancies and the main map data source is one that has been georeferenced manually, human error is the mist likely cause.

 

[steamboateng]

Thanks for info geophil. We can assume then, with maps accurately georeferenced, and terrain baseboards generated in TrainzDEM with a smoothing factor of 0 (zero) that any hight errors are pretty much inherent in the initial DEM data downloaded from the web, and also thsose small errors introduced via interpolation when converting data to the Trainz map.
The USGS states that its 1/3 arc sec data located on the NED site has an error of about 2.44 m (8 ft.).
Well JCitron, we can blame all our elevation troubles on the US Government..............but hey, they're used to it!
Regards, Dr. Z; and thanks again for taking the time to explain.

 

[JCitron]

I concur, Mike. The US Government is at it again.

Thank you, Roland for taking the time to explain this.

The 2.4m discrepancy does make that difference in complicated urban areas such as Haverhill at Washington Street near the depot and on the other end of the bridge. The tracks have been reworked quite a lot since the topo maps were created, and today the alignment on both ends of the bridge are so different which doesn't help the situation either.

http://binged.it/WBaSmI

http://binged.it/Wicy5i

John

 

[steamboateng]

John, I han't got to Havehill or Lowell yet. I'm gonna wait until you finish it and then ask you how you done it!
I've been thinking about Roland's thread. My route was generated with the 0 smoothing setting, and is much more representatrive of the actual landscape than a setting of 3. However that shallow bowl effect is still noticable at waterbody shorelines. I will assume that the error, although minor, is associated with the interpolation algorithms.
Thanks to Blutorse4792 for letting us ramble on in your thread.
Regards all

 

[JCitron]

John, I han't got to Havehill or Lowell yet. I'm gonna wait until you finish it and then ask you how you done it!
I've been thinking about Roland's thread. My route was generated with the 0 smoothing setting, and is much more representatrive of the actual landscape than a setting of 3. However that shallow bowl effect is still noticable at waterbody shorelines. I will assume that the error, although minor, is associated with the interpolation algorithms.
Thanks to Blutorse4792 for letting us ramble on in your thread.
Regards all

Gee thanks, Mike.

There are issues here beyond what I've discovered with the DEM. Oh so much is gone and has been "revised" and urban renewed over the years. The Merrimack River bridge has become an issue too. I had to fake one using splines. It's wrong, but it works. I may ask the Bridge Man Ben to make me one if I can get some of the dimensions for the bridge. The same with the stations with both Haverhill and Bradford on very slight curves which are difficult to duplicate with the station assets we have, and I'm not in a position to make the models anymore. My patience level for things like this is about 20ms on good days.

I too have been thinking about his thread. I may very well pull what I have done, which isn't much anyway, and redo it again. Oh well for the last 20 hours I've spent on this. I'm not doing much anyway, so there's nothing but my time lost here and I have plenty of that on my hands now.

The water issue maybe due to the lack of height information for the bottom of the water bodies so they're flattened out.

John

 

[Dermmy]

The same with the stations with both Haverhill and Bradford on very slight curves which are difficult to duplicate with the station assets we have, and I'm not in a position to make the models anymore.

John

Hunt out andi06's station kit, the assets are all on the DLS but you can get a single DLS link to every part of the kit from his site. The kit follows British practice but if you sunk the platforms into the ground a bit I would think you could get a pretty good representation of US practice, or just use one of his invisible platforms below a US spline of your choice. The kit enables just about any station configuration you could want....

 

[steamboateng]

John, I think the US DEM data presented is radar data, thus water body levels would be incorporated into the dems. The USGS does do seabed surveys, but I think that data is not relevent on these forum.
I understand your frustration at trying to reconcile a DEM baseboard map with 'historical' data. However, we must keep in mind that the landscape has changed. We are working with maps generated in the 40's And god only knows where they got that info from. Most of the maps we work with are referenced to NAD27! Any NED data we download now, is relevent to now. A bit of 'Kentucky windage" should be taken into acount when interpreting between the two. I believe it's a mistake to believe that Dr. Z is pesenting us with a perfect program. He's only providing us with a 'workable' program that fulfills the neeeds of most Trainzers......................!
It is grossly overestimating Dr. Z's capabilities to believe his program can meet all of our DEM needs, regardles of how specific
Quit watching the Fox Channel; it's bad for your heart..

 

[JCitron]

Hunt out andi06's station kit, the assets are all on the DLS but you can get a single DLS link to every part of the kit from his site. The kit follows British practice but if you sunk the platforms into the ground a bit I would think you could get a pretty good representation of US practice, or just use one of his invisible platforms below a US spline of your choice. The kit enables just about any station configuration you could want....

Yeah, I could do that Andy. I never thought of it actually. I've used the kits before, and I won't bother lowering them because we now use high platforms at the commuter stations. The two stations I'm working with were updated during the early part of 2000s when the Downeaster service was started, and Haverhill Station was rebuilt. The Downeaster doesn't stop at Bradford, but they upgraded that station anyway. When the area was upgraded, they dug up a lot of the existing tracks that ran from the old Bradford yard to the Groveland branch. which was abandoned anyway, along with the cross-over, to what was left of the yard. The yard became a parking lot with two storage tracks, and the platforms were moved over to the right on what was the main line. They also extended the platforms for quite a distance too perhaps from 50-100 feet to 820 feet. (250 meters).

Anyway, Mike is right. We're using topo maps that date back to the 1940s and probably before. If I recall, I've seen some that are earlier than that somewhere. If anything maybe I'm being too picky about what I have to work with, and should settle on a compromise. I'll still see if Mr. Bridge Man can make me the bridge though. the splines are almost there, but not quite right. This is what happens when you live in the area you're trying to model.

John

 

[steamboateng]

Many of the UNH maps date before the 40's. I suspect that any revisions were larghely cosmetic until after the War; .i.e 1950's .
I also suspect that tail gunner Joe Macathy wasn't preaching 'transparency' at that time
I would conclude that the maps had a bit of 'Kentucky windage'.
But this is all supposition, I really don't know.
And as a student of American history I find it 'veddy, veddy interesting!'

 

[Dermmy]

This is what happens when you live in the area you're trying to model.

John

I gave up! The route I'd most like to model is a local Australian route and I just know it too well. On it I can't live with the compromises I accept without question on the routes I know next to nothing about...

 

[steamboateng]

Off topic......a lot!
Dermy, while flying across the the Kokoda Trail, Port Moresby and into Townsville, then down to Brisbane, I noticed a country with a lot of horny kangaroos and lotsa frogs. (Yep, you can see 'em jumpin' around from the air). Is that you guys. I was taught you folks walked upside down and your toilets swished in the opposite direction! Is that true?

Enquiring People want to know!

 

[geophil]

The best thing is to give an example what can be achieved with USGS data.

My very first encounter with 1/3 arc sec DEMs was Truckee, CA (Donner Pass), back in 2006. The 1:24k topo map is on top, WMS or manually georeferenced should not matter. The DEM has been resampled to a 10 m raster.

Have a closer look at the sewage ponds. (No, they don't smell here. ):

And here we have the result in TS12, 5m grid. I'd say the map is spot on.

I also discovered that the smoothing filter has no effect here. I tried with both the default setting of 3 and switching it off = 0. No difference. I think that has been the original intention. Smoothing is not needed for hi-res DEMs.


Regarding NAD27: Yes, many of the topo maps in use are in NAD27 coordinates. The conversion from NAD27 to NAD83/WGS84 is not without error, not because of the formulas but because of the original surveying, but that error should be in the 1-2 m range.

 

[steamboateng]

I think more information must be presenterd to substantiate that claim.. No bones and no pun intentded!

 

[Dermmy]

This is my only TransDEM route ever - the Tralee & Dingle. Irish prototype, which has a non-standard map referencing system which TransDEM handled without issue. I posted this shot many years ago on a bulletin board dedicated to the route and the location was immediately recognized by the locals on account of the profile of the hill. The overlay is from scanned paper maps and the contour lines fit the hills exactly over the entire route...

@steamboateng - we are I assure you entirely right-side-up, but the toilets do spin in the opposite direction...

 

[steamboateng]

I am comforted by your assurance. Physics has it's own way of workin' stuff out!
Geophil, that is not a fair assment of your program.