Improving the resolution of 90m SRTM data to 30m for use in cartography outside USA
If you are studying an area outside North America, you may be frustrated by not being able have higher than 90 metre resolution DEMs available. This would be important if you would like a backdrop for a large scale map (i.e. a nice hillshade) or would like to do some more detailed topographic analysis. I have read with great interest the following articles of the Web.
Valeriano et al. “Modeling small watersheds in Brazilian Amazonia with shuttle radar topographic mission-90m data” http://www.mamiraua.org.br/admin/imgeditor/File/publicacoescientificas/2006/artigo_15.pdf
Keeratikasikorn & Trisiristayawong “RECONSTRUCTION OF 30M DEM FROM 90M SRTM DEM WITH BICUBIC POLYNOMIAL INTERPOLATION METHOD”
Grohmann “Resampling SRTM 03”-data with kriging”
Each of these seem excellent methods for improving the resolution of 90m SRTM to 30m through resampling and interpolation. However, my maths and understanding of interpolation are limited to be able to work out how to do this in ArcGIS Spatial Analyst. While the actual data is not improved, its representation and also its ability to be used in more refined topographical analysis is.
Has anyone accomplished this using ArcGIS Spatial Analyst? Does anyone know which is the best interpolation method? Kriging? Bicubic polynomial? Regularized Splines with Tension (RST)? And if so, what are good parameters to use within the tools?
This process would be extremely useful to thousands of ArcGIS users outside the U.S. Also, developing a script/tool which one could input a SRTM and output a higher resolution one would be amazing.
Any help on this would be much appreciated.
Mapping Center Answer:
Okay, before I describe how I would go about this, I feel I would be horribly irresponsible if I did not explain the likely pitfalls of this sort of endeavor.
First, understand the nature of elevation rasters--as resolution decreases the statistics of derived products (slope, aspect, etc.) are subject to the law of averages. That means if you computed the average slopes for four resolutions of DEMs covering an identical extent you'd see something like this:
- 2m LIDAR: Avg Slope = 28%
- 10m DEM: Avg Slope = 23%
- 30m DEM: Avg Slope = 20%
- 90m DEM: Avg Slope = 18%
The effect is diminished if the average slope is lower, and it is exascerbated if the average slope is higher (mainly due to the idea of uniform pixel sizes).
The problem is that you cannot introduce the richness that generalizing a DEM removes from your data by "reversing" the process through interpolation. If you produced a 10m DEM from a 90m DEM via interpolation, then calculated the average slope, you'd find they had the same slope. If you're in a mountainous area, I think you're just wasting your time.
So, all that said, I still like bilinear interpolation. If building a DEM from points, I like IDW. I also would use the Filter tool using the Lower option.
I also hear that folks really value smooth looking contours, and they find down-sampling their DEM and then running Focal Stats (circle, 4) to smooth the result before generating contours. The purpose is to achieve a smooth aesthetic in their contours like those in late 19th and early 20th century topographic maps. Personally, I think this is misguided. I was reading Erwin Raisz's 1948 "General Cartography" a few years back and found an interesting passage on page 109. Raisz is talking about the "Pivot Pen" used to draw contours and notes its disadvantage of making lines rounded and expressionless. I agree--the whole point of contours is to render the shape of the landscape. That said, if the more realistic contours (there's a phrase to be crucified over), don't look good on a map, consider changing the symbol so they're not so prominent.
All that said, I would rather spend energy urging the folks who control the availability of the 30m SRTM data to make it available.
If you would like to post a comment, please login.