Since the GIF format and many image processing implementations require 8-bit data, we might split our 16-bit data into two separate 8-bit planes, containing the hi-order and lo-order bytes, respectively, of each elevation, and encode each plane separately. Of three different compression algorithms applied to the hi file, GIF was the best, somewhat better than gzip on the binary file. Gzip was slightly the best on the low file, but all the method were close enough that the difference is probably not significant.
Combining GIF on the hi with gzip on the binary lo gave 191,670 bytes, or 5.9 bpp, or less efficient than some following methods.
Splitting also illustrates a weakness in gzip. It compressed the 1-byte-per-point lo file down to 182,038 bytes. However, we tried expanding the lo file to 2 bytes per point, with the hi byte always zero. Gzip compressed this only to 217,335 bytes, 20% worse. This also occurred when the hi file was expanded to 2 bytes per point. Gzip's compressed size grew from 9.711 to 11,433 bytes. Also gzip on the two halves of the file was more efficient than gzip on the whole file. This was true for the file in either ASCII or binary. In the latter case, the difference was 17%.