To appear
Previous: 6 Compressing the DEM in Pieces
In the above section, it seems that progcode is the best lossless
compression of elevation data. Does this extend to other data? The full
test was to take a random sample of all the USGS DEMs[35] (except
Alaska because of its different size). For each letter of the alphabet
(except X and Z), we took the alphabetically first available DEM starting
with that letter. They are shown in Figure f:all24. We losslessly
compressed a 
extract from each file thrice, with
gzip (for comparison since it is so popular),
progcode, and sp_compress.
Table t:24 shows the resulting number of bpp, and the compression ratio, measured relative to the binary file size. We report the arithmetic average, not the, lower, geometric mean at that bottom since this is more conservative and also more relevant when compressing and storing many files at once.
Note that
both the gzipped and our compressed sizes vary wildly from file to
file. Each original file was 
bytes. All the
compression times were in the range from 12 to 26 seconds. In every case,
sp_compress and progcode were better than
gzip, compressing the whole set of 24 files down to 2.0
and 2.1 bpp, respectively, or half the size of gzip. Therefore, both
sp_compress and progcode seem excellent choices for
compressing gridded elevation data.
Testing progdecd at different bit rates on the files compressed by progcode showed that on the average there were no differences at an uncompression rate of 0.7 bpp. This shows us a goal for lossless compression.