[GRASS-user] interpolating images with coastal areas

GRASS GRASS user
1

Dear List,

I am working on interpolating 30m ASTER DEMs to 10m resolution. I am using GRASS 6.4.1 svn on Mac OS 10.6.6. Since the imagery covers a large area, I crop the 30m DEM into individual tiles in order to make the project more manageable. The interpolation process (using v.surf.rst) has been working fine and the values in resultant 10m DEMs have been quite accurate, especially for inlands. However, I have some problems with interpolating tiles where coastline is present. In these tiles, the range of values in 10m DEMs go below zero. In some cases I have values such as -21 meters whereas the original –30m– tiles have only zero. Obviously, I am concerned about the accuracy of the end product.

Do I get high negative values because of the cells with zero value?

Should I zoom further in and avoid having zero-value cells? (which is a problem with coastlines where they are not straight)

Does any one have a suggestion about how to deal with this problem?

  • I sample the original imagery at 50% in r.random,

  • I use the default parameter values in v.surf.rst (tension= 40.) and use the 30m DEM as MASKMAP,

  • I align region to resolution, using g.region.

Thanks!

–
BÜLENT

2

Dear List,

I am working on interpolating 30m ASTER DEMs to 10m resolution. I am using GRASS 6.4.1 svn on Mac OS 10.6.6. Since the imagery covers a large area, I crop the 30m DEM into individual tiles in order to make the project more manageable. The interpolation process (using v.surf.rst) has been working fine and the values in resultant 10m DEMs have been quite accurate, especially for inlands. However, I have some problems with interpolating tiles where coastline is present. In these tiles, the range of values in 10m DEMs go below zero. In some cases I have values such as -21 meters whereas the original –30m– tiles have only zero. Obviously, I am concerned about the accuracy of the end product.

Interesting problem.
AFAIK the ASTER DEM does not have bathymetry information so the coverage ends at the coastline. I suppose that when there’s a slope down to the coast, the interpolation algorithm continues that slope beyond the coastline into the “nodata” cells thus creating negative values.
If you don’t need the below sea level elevations, why not just set everything beyond the coastline to NULL?

··· 
-- 
Micha Silver
Arava Development Co. +972-52-3665918
[http://surfaces.co.il](http://surfaces.co.il)
3

Hi,

The problem is probably the data for the sea of the ASTER images. In some cases I have seen negative values for sea areas and extremely high values, errors also, in mountainous areas. For interpolating I think it is better to treat the sea areas as no data values. Use the land area as mask in the interpolation. You can also check the ASTER elevation data comparing with the corrected version of 90-m SRTM from http://srtm.csi.cgiar.org/

Best,

Edgar Pimiento Chamorro
e_pimiento@yahoo.com

From: Bulent Arikan bulent.arikan@gmail.com
To: grass-user@lists.osgeo.org
Sent: Mon, March 21, 2011 3:03:35 AM
Subject: [GRASS-user] interpolating images with coastal areas

Dear List,

I am working on interpolating 30m ASTER DEMs to 10m resolution. I am using GRASS 6.4.1 svn on Mac OS 10.6.6. Since the imagery covers a large area, I crop the 30m DEM into individual tiles in order to make the project more manageable. The interpolation process (using v.surf.rst) has been working fine and the values in resultant 10m DEMs have been quite accurate, especially for inlands. However, I have some problems with interpolating tiles where coastline is present. In these tiles, the range of values in 10m DEMs go below zero. In some cases I have values such as -21 meters whereas the original –30m– tiles have only zero. Obviously, I am concerned about the accuracy of the end product.

Do I get high negative values because of the cells with zero value?

Should I zoom further in and avoid having zero-value cells? (which is a problem with coastlines where they are not straight)

Does any one have a suggestion about how to deal with this problem?

  • I sample the original imagery at 50% in r.random,

  • I use the default parameter values in v.surf.rst (tension= 40.) and use the 30m DEM as MASKMAP,

  • I align region to resolution, using g.region.

Thanks!

–
BÜLENT

4

Hi Bulent,

Yes, I think it is better to treat the sea area as NULL data. One way to interpolate the DEM is to use a mask corresponding to the land area, another way is to replace the values of the sea with null data in the DEM.

Best,

Edgar Pimiento

From: Bulent Arikan bulent.arikan@gmail.com
To: Edgar Pimiento e_pimiento@yahoo.com
Sent: Mon, March 21, 2011 9:20:11 AM
Subject: Re: [GRASS-user] interpolating images with coastal areas

Hi Edgar,
Thank you for the explanation and suggestion. If I understand you correctly, you suggest converting elevation values zero and below to NULL and use that 30m resolution imagery for interpolation. So, this map should be used as MASK.
I hope I got it right?

Best,
Bulent

On Sun, Mar 20, 2011 at 3:14 PM, Edgar Pimiento <e_pimiento@yahoo.com> wrote:

Hi,

The problem is probably the data for the sea of the ASTER images. In some cases I have seen negative values for sea areas and extremely high values, errors also, in mountainous areas. For interpolating I think it is better to treat the sea areas as no data values. Use the land area as mask in the interpolation. You can also check the ASTER elevation data comparing with the corrected version of 90-m SRTM from http://srtm.csi.cgiar.org/

Best,

Edgar Pimiento Chamorro
e_pimiento@yahoo.com

From: Bulent Arikan <bulent.arikan@gmail.com>
To: grass-user@lists.osgeo.org
Sent: Mon, March 21, 2011 3:03:35 AM
Subject: [GRASS-user] interpolating images with coastal areas

Dear List,

I am working on interpolating 30m ASTER DEMs to 10m resolution. I am using GRASS 6.4.1 svn on Mac OS 10.6.6. Since the imagery covers a large area, I crop the 30m DEM into individual tiles in order to make the project more manageable. The interpolation process (using v.surf.rst) has been working fine and the values in resultant 10m DEMs have been quite accurate, especially for inlands. However, I have some problems with interpolating tiles where coastline is present. In these tiles, the range of values in 10m DEMs go below zero. In some cases I have values such as -21 meters whereas the original –30m– tiles have only zero. Obviously, I am concerned about the accuracy of the end product.

Do I get high negative values because of the cells with zero value?

Should I zoom further in and avoid having zero-value cells? (which is a problem with coastlines where they are not straight)

Does any one have a suggestion about how to deal with this problem?

  • I sample the original imagery at 50% in r.random,

  • I use the default parameter values in v.surf.rst (tension= 40.) and use the 30m DEM as MASKMAP,

  • I align region to resolution, using g.region.

Thanks!

–
BÜLENT

–
BÜLENT ARIKAN, PhD
Postdoctoral Scholar
Center for Social Dynamics and Complexity &
School of Human Evolution and Social Change
Arizona State University
Tempe - AZ
85287-2402