19 replies to this topic

[Martin Gamache]

Attempting to re-interpolate a new DEM from a topo map (DLG/DRG) would be an excercise in futility- this is how the lower quality NED data was produced in the first place, and any such product would be likely be inferiour to even the NED data.

Not necessarily.

NED has lots of problems due to legacy USGS DEMS with all kinds of production artifacts such as scan lines, seams, quilting patterns etc... Sometimes it is useful to regenerate from the original DLG data to get rid of those artifacts, especially if they happen to fall over complex terrain where fudgin it in Photoshop would be difficult and time consumming.

Dylan, as you seem to know alot about this would you happen to know which interpolation tools / methods were used to generate USGS DEMs in the first place. I know about how the data was collected (GESTAL Photo Mapper vs, digitizing, vs. Photgrammetry) from the USGS documentation, but have found very little info on how the DEM themselves were interpolated.

In your opinion, given zero knowledge of GRASS how long and difficult would it be for someone to become familiar with it enought to learn to use to be able to run the RST algorithm on some vector data and export a DEM?

Thanks,


mg

[dylan]

Hmm... the BBcode doesn't seem to be working for me today...

----Quote: Alpine Mapping Guild ----
Not necessarily.

NED has lots of problems due to legacy USGS DEMS with all kinds of production artifacts such as scan lines, seams, quilting patterns etc... Sometimes it is useful to regenerate from the original DLG data to get rid of those artifacts, especially if they happen to fall over complex terrain where fudgin it in Photoshop would be difficult and time consumming.
----Quote: Alpine Mapping Guild ----


Good point. There may be cases where the NED is so bad, that a re-interpolation from the DLG contours will give a better result. Even the current NED data can have serious "banding" like errors (for lack of a better word). However, interpolation from contours is tricky, and the results are highly method dependent. Some common methods would include ANUDEM, TAPES-C, Topo-something in Arc, and v.surf.rst .

----Quote: Alpine Mapping Guild ----
Dylan, as you seem to know alot about this would you happen to know which interpolation tools / methods were used to generate USGS DEMs in the first place. I know about how the data was collected (GESTAL Photo Mapper vs, digitizing, vs. Photgrammetry) from the USGS documentation, but have found very little info on how the DEM themselves were interpolated.
----Quote: Alpine Mapping Guild ----

From what I have read, many of the USGS DEM products are created via a work flow like you mentioned: stereo photos -> contours via a mechanical/automated system -> (simple spline-like interpolator) -> an interpolated, gridded final product. The end result is a nice gridded dataset, but with various types of artifacts. As depicted in the attached image, contour heights are over represented in the gridded DEM, and calculated aspect angles seem to suffer from over represented values in the 8 cardinal directions. In addition, the spline interpolator can introduce "waves" or "ripples" in the DEM. I should note that the previously mentioned RST method can do the same thing, however it is possible to account for this in the parameters passed to the algorithm. The book Terrain Analysis (available on Amazon), has some very good discussion on these topics, but little is mentioned about how to fix them. If anyone is interested, I have found that a "terrain rescaling" algorithm in GRASS (and a Java version which I have forgotten the name of) can generalize terrain in a way that filters out the noise, while preserving the overall shape. In GRASS, this command is called r.param.scale .

----Quote: Alpine Mapping Guild ----
In your opinion, given zero knowledge of GRASS how long and difficult would it be for someone to become familiar with it enought to learn to use to be able to run the RST algorithm on some vector data and export a DEM?
----Quote: Alpine Mapping Guild ----

If you have a MacOS machine, then installing and running GRASS is a simple as any other Mac application. Here is a link to some pre-packaged downloads. As mentioned before getting started is the hardest part. Once you have your data in GRASS, I would recommend reading the manual pages on the RST method, along with the linked publications in the manual. I would say that a couple hours should be enough. Posting to the mailing list can save lots of time as well. However, there is still much debate over which algorithm will consistantly produce a more hydrologically correct DEM.

Cheers,

Dylan

Attached Files

•  dem_error_example_big.png   12.59K   51 downloads

[franciscocartographer]

Ok. I just wanted to thank you all for the "knowledge sharing" regarding this topic. By the way, “Landserf” is the java application that Dylan was referring to as a comparison to the Grass command “r.param.scale” used for filtering the noise while maintaining the appearance of the DEM.

Well as I mentioned at the beginning of the discussion I was interested in creating a 10 meter DEM for an area of about 600 acres located at the center of the Caribbean island of Puerto Rico (beautiful setting by the way). The original data I was given was a 30m Ned. As you can imagine this spatial resolution is a little bit coarse for this big scale mapping project.

The way I solved this issue was as follow:

1) I was able to download a DLG of the township, so I imported it with “MapViewer” along with its elevation attributes. This saved me from having to digitize contours.
2) Then this layer was exported as a “shapefile” and imported in “IDRISI”, were I had two options. Either convert the contours to TIN to further create the DEM, or use an algorithm called “Intercon” to interpolate from the contours. I decided for the first option.
3) After the TIN was created the process to create the DEM was straight forward from “IDRISI”.

I include the Hillshade results from the 30 meter NED and from the 10 meter DEM created from the DLG.

Be aware that I don’t have access to ArcInfo (and not planning to) so “Topogrid” was out of the question. Also “Grass” is more suitable for the Mac environment and its learning curve is a little bit demanding. Nevertheless my intention in discussing this workflow is for you to read it (if you have the time and are interested in the topic) and then to offer your input on better or more effective ways to do this.

Thank you in advance!!

Attached Files

•  30mned.png   51.33K   80 downloads
•  10MDLG2.png   97.06K   92 downloads

[Martin Gamache]

Francisco,

Just to confirm,

left image is the 10m interpolated in IDRISI from TIN and right image is 30m NED?

mg

[franciscocartographer]

Francisco,

Just to confirm,

left image is the 10m interpolated in IDRISI from TIN and right image is 30m NED?

mg

<{POST_SNAPBACK}>

Yes.