GHCN Interpolated Elevation: You take the high road …
GHCN has two elevation fields. The first is the original station elevation in meters (na string is -999). The second is the station elevation interpolated from TerrainBase gridded data set.
Like most station databases, these metadata start off with station name, latitude, longitude, and elevation. Wherever possible, these were obtained from the current WMO station listings (WMO 1996b). Some stations in GHCN do not have elevation metadata. To provide all stations with some elevation information, an elevation value was interpolated to the station location from a 5-min gridded elevation database (Row and Hastings 1994) and this elevation is provided in addition to official station elevations. In areas with significant orography, these interpolated metadata will have limited specific accuracy. But they can provide useful information about the station’s elevation.
TerrainBase Global Land Elevation and Ocean Depth
The TerrainBase Global Land Elevation and Ocean Depth is the data set described by Peterson and Vose in 1997 as they GHCN interpolated elevation data set.It is available for download here” http://dss.ucar.edu/datasets/ds759.2/matrix.html (registration required). The description is as follows:
Five minute resolution global ocean depth and land surface elevation. Developed from multiple data sources and compiled at NGDC. Compared to the predecessor data set (ETOPO5) TerrainBase is mostly improved in land elevations areas, few changes were made to ocean depths.
The tbase data is in a non-standard gridded format harking back to the magnetic tape days. Each latitude is represented by 216 lines of 20 records for a total of 4320 records per lat line and a resolution of 0.0833 degrees (5 minutes square). It has full global coverage.
I wrote a small script to convert the tbase data format to a ArcInfo ASCII Grid format and then used gdal_translate to covert the file to a GeoTIFF format.
Using the GeoTiffDataReader, I pulled the elevation data for each GHCN station and compared it with that recorded in the GHCN station inventory file. You can see the ‘stacked’ comparison file here (ghcn-tbase-nointerp-compare-no-off.inv). Looking closer, I used R to compute the standard deviation of the difference between the GHCN interpolated elevation value and the value I pulled from tbase:
Standard Deviation = 276m (simple lookup, no offset)
But the GHCN station inventory uses the interpolated elevation. This page provides an overview of grid cell interpolation: http://www.geocomputation.org/1999/082/gc_082.htm
First, get the elevation for each corner of the grid for the lat,lon pain in which we are interested:
h1 = getData(lat, lon);
h2 = getData(lat, lon+stepsize);
h3 = getData(lat+stepsize, lon);
h4 = getData(lat+stepsize, lon+stepsize);
Then find the x and y distance from the origin of this grid cell:
x = lon % stepsize;
y = lat % stepsize;
Apply the binomial interpolation:
a00 = h1
a10 = h2 – h1
a01 = h3 – h1
a11 = h1 – h2 – h3 + h4
h = a00 + a10x + a01y + a11xy
However, after applying this interpolation to the elevation data provided in GHCN inventory, there is no improvement in GHCN matching over a simple grid cell look-up:
Standard Deviation = 276m (interpolation, no offset)
ETOPO1 Global Relief Model
This data is available in “grid registered” and “cell registered” versions. The difference is one I’ve alluded to in previous posts. If you think of the geotiff data as a checkerboard table, then the “grid registered” centers the lat,lon in the center of the the cell while “cell registered” places the lat,lon on the intersections.
There is a marked improvement in using the ETOPO1 data for GHCN elevation matching:
Standard Deviation = 146m (interpolation, no offset)
Shuttle Radar Topography Mission (SRTM)
The Shuttle Radar Topography Mission (SRTM) was a space shuttle mission in February 2000. It covers 60S – 60N. I used the 1km resolution file. Version 4 of the data is available at the Consortium for Spatial Information (CGIAR-CSI). The data is grid-registered. The spread of elevation matches is little wider than that for the ETOPO1 data:
Standard Deviation = 153m (interpolated, offset)
ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer)
ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) is a joint program between NASA, the Japan’s Ministry of Economy, Trade and Industry (METI) and Japan’s Earth Remote Sensing Data Analysis Center. The first ASTER EOS sat went up in Dec 1999 and they measure surface temperatures, reflectivity, and elevation. The data is available for 83S to 83N, a marked improvement over the SRTM data. Unfortunately, the ASTER data appears to be only available in tiled format and I am not currently set up to process tiles. So this data will be left for another day.
It is a little odd that the ETOPO1 gives a better match with the GHCN inventory than the TerrainBase given that the original GHCN v2 description claims that the above TerrainBase is the actual data set used. I wonder if I missed an update.
For another look, I ran a spread of the GHCN interpolated elevation -v- the GHCN listed elevation (first elevation field). SD = 184m. By comparison, the ETOPO1 interpolated elevation -v- the listed elevation has an SD = 134m.
Jarvis A., H.I. Reuter, A. Nelson, E. Guevara, 2008, Hole-filled seamless SRTM data V4, International Centre for Tropical Agriculture (CIAT), available from http://srtm.csi.cgiar.org