26 replies to this topic

[merft]

I've been following this thread and decided to take a closer look at the SRTM data now that I just upgraded to 9.2. Perhaps someone has some suggestions on how to create a hillshade that looks like the .gif that is delivered with the data. The default hillshade I created looks pretty poor.

You can use the offset method but it is sketchy.

The process I used mainly because I was also calculating aspect and slope was to reproject the DEMs. However, ESRI's raster reprojection algorithm leaves a bit to be desired. It typically results in artifacting and yes I've tried several different methods. I use either of these two different methods of reprojecting depending on the project.

1) Purchase a tool that reprojects better. I use Global Mapper quite a bit for data conversion and it seems to reproject well. Its well worth the $250 versus the next process. I use Global Mapper a lot to downsample 30-meter DEMs to 90-meter and reproject to custom projections all the time. Great tool for working with large areas.

2) This requires Spatial Analyst and Arc/Info and I tend to use this process on smaller site specific projects.
- ArcGIS: Open your DEM and contour your DEM
- ArcGIS: Reproject the shapefile contours to the desired projection.
- ArcInfo: Convert shapefile contours to coverage
- ArcInfo: Create new DEM using TOPOGRID

There is a new tool in Spatial Analyst that replaces the TOPOGRID command but it can get flaky. I usually get a "Too many vertices" error and it bombs. There are a lot of little extras I might do depending on the project; smooth the contours, obtain point data for peaks and sinkholes, etc. I rarely use command-line ArcInfo anymore but keep it around just for the TOPOGRID command and a variety of AMLs I still use.

-Tom

[MaryLynne]

Thank you Tom & Matt. I ran a hillshade with the Spatial Analyst changes and it looks much better.
I appreciate the ArcInfo and Global Mapper suggestions, I will spend some more time with AI to see if that will work for me or if I need to go to Global Mapper.

Mary

[Charlie Frye]

The data is in decimal degrees - so in Spatial Analyst-->Surface Analysis-->Hillshade... the Z factor should be set to .0001 (or something close) instead of the default "1."

Actually, the reason you need to do this is that the linear units of the coordinate system are often not defined for DEMs. The hillshade tool in ArcGIS expects to know the ratio of vertical units to horizontal units. In your case its meters to decimal degrees.

If your DEM is using decimal degrees and your vertical unit is meters, use 0.000008103
If your DEM is using decimal degrees and your vertical unit is feet, use 0.00000274

Note that since these values work best 30-50 degrees north or south latitude. You may need to adjust them.

If you want nicer looking hillshades try using the Hillshading models we've developed over the past few years, including the swiss and MDOW methods. They're on the Basemap Data Model hompage. Use the find tool and type in Hillshade to find the link amongst the others.

Also, Tom--when's the last time you tried raster projections in ArcGIS--we worked on them extensively in 9.1 and 9.2--They're definitely improved, I'd be interested in your opinion.

Also, I've found that for the contour lines, once I get them in shape file format (though with 9.2 using the new File Geodatabase should get around the file size and memory limits that the personal Geodatabase had), I can run the Simplify line tool with a very small tolerance, using the Point Remove option (which is actually the Douglass-Peucker algorithm). I can remove between 40 and 80% of the vertexes without significantly changing the shape of the line. I actually do a little bit more aggressive simplification to produce a contour line dataset for labeling efficiency as well.

Anyway, we're also working on a new contour line algorithm for 9.3 that will provide more control and better results in general.

[merft]

If you want nicer looking hillshades try using the Hillshading models we've developed over the past few years, including the swiss and MDOW methods. They're on the Basemap Data Model hompage. Use the find tool and type in Hillshade to find the link amongst the others.

Also, Tom--when's the last time you tried raster projections in ArcGIS--we worked on them extensively in 9.1 and 9.2--They're definitely improved, I'd be interested in your opinion.

Also, I've found that for the contour lines, once I get them in shape file format (though with 9.2 using the new File Geodatabase should get around the file size and memory limits that the personal Geodatabase had), I can run the Simplify line tool with a very small tolerance, using the Point Remove option (which is actually the Douglass-Peucker algorithm). I can remove between 40 and 80% of the vertexes without significantly changing the shape of the line. I actually do a little bit more aggressive simplification to produce a contour line dataset for labeling efficiency as well.

Anyway, we're also working on a new contour line algorithm for 9.3 that will provide more control and better results in general.

You got me on the reprojection. To be honest haven't tried reprojecting rasters in ArcGIS probably since early 8.x versions. I'll have to take a look at it again.

Since I have ArcGIS ArcView, I use ET Geowizards for the majority of my smoothing and point reduction. However, the recent post to theMapShaper website looks really promising. Especially the ability to choose simplication methods and play with the simplification on the fly.

Its good to hear that ESRI is working on cleaning their alogrithms. I beating a dead horse here but ESRI really needs to implement bezier curves especially when displaying organics such as contours, streams, vegetation boundaries, etc. I do realize that this is a daunting task as it would require rewriting nearly every algorithm. But I can dream.

-Tom

[Matthew Hampton]

Thanks for the clarification and explanation Charlie. I knew the "what" but not the "why." I am intrigued to play with the hilllshade tools, and might take a little recess from work to jump in the sandbox and build some mountiains...

FWIW: Tom, I just put text on a bezier curve in 9.1. It made me feel like a cartographer.

[MaryLynne]

If your DEM is using decimal degrees and your vertical unit is meters, use 0.000008103
If your DEM is using decimal degrees and your vertical unit is feet, use 0.00000274

I must be missing a step. I brought in the SRTM .dem data directly into ArcGIS 9.2. Do you find the best result is to to create the hillshade directly from the .dem or to convert the .dem to a grid first, then create a hillshade?

Charlie, when you say use 0.000008103, is that for the Output cell size in the Hillshade dialog box?

I can't get any number smaller than 0.0001 into that box without an error message.

Mary

[Charlie Frye]

I must be missing a step. I brought in the SRTM .dem data directly into ArcGIS 9.2. Do you find the best result is to to create the hillshade directly from the .dem or to convert the .dem to a grid first, then create a hillshade?

Charlie, when you say use 0.000008103, is that for the Output cell size in the Hillshade dialog box?

I can't get any number smaller than 0.0001 into that box without an error message.

I've usually created a GRID from the DEM, then in the hillshade tool these values are for the Z-Factor parameter; which by default is set to 1. The cellsize is usually set to be the same as the input raster.

Do that only when the Linear Unit of the Spatial Reference is undefined (this is one of the properties of the raster dataset).

I beating a dead horse here but ESRI really needs to implement bezier curves especially when displaying organics such as contours, streams, vegetation boundaries, etc. I do realize that this is a daunting task as it would require rewriting nearly every algorithm. But I can dream.

We've had bezier curves since early in 8.x days. The main issue is that there aren't many tools that automatically generate parametric geometry (curves); in fact, I can't think of any off hand. There are also a few tools that still densify curves into vertexes.

In the editor, there are a number of sketch tools, one is called End Point Arc, and that does a curve--it can be unwieldy, especially for... well, all but hard core curve editors, so in response we added the Mid Point Arc tool, which is much easier to use.

[Matthew Hampton]

there aren't many tools that automatically generate parametric geometry

My favorite is Illustrator's Pencil tool. It's downright magic and also doubles as an interactive smooth tool.

[merft]

My favorite is Illustrator's Pencil tool. It's downright magic and also doubles as an interactive smooth tool.

I have to second that and add Illustrator's Pen tool. Between the Pen and Pencil tool, I can enter/edit data 10x faster than using the vertex-oriented tools in ArcGIS. It is not unusual for me to export shapefiles to Illustrator and I will just redraw them on a new layer in Illustrator. Fewer vertices, cleaner more organic features. Go Go Wacom Tablet!

[MapMedia]

Where can you get 30m SRTM version 3 data easily (with an interface)?

I have found a few good tools to D/L 90m v3 SRTM (www.ambiotek.com/srtm has GE kml tool - very nice!),
and XY Download SRTM tool for Arcmap is super easy (d/ls right selected 90m tiles right into your data view - but only version 2).

[Martin Gamache]

It is only available for the US. Best place IMO is from the NED website' Seamless Data Distribution System:SDDS

Also worth mentioning is that the Landcover 2001 for the US is now done along with the impervious and canopy cover datasets, all available from SDDS as well. This is quite a significant dataset.

[Matthew Hampton]

I think it is also worth highlighting that the 2001 NLCD Landcover (based on 30m Landsat) has both the Impervious and Canopy layers classified at the sub-pixel level. Both of these classes are derived as the "percentage of coverage" in a particular 30m raster cell - so there is greater detail.