that 
. for 
ited 
ture 
‘ces 
-km 
5 on 
vical 
tory 
ners 
the 
ban 
ions 
93). 
jean 
ve a 
aily, 
rmal 
ts in 
stem 
ellite 
rate 
DIS) 
lobal 
al to 
land 
stem 
ough 
and 
daily 
land 
d the 
t bed 
ative 
and 
a. 
| Set 
|, 10- 
lution 
orbit 
Two 
1-km 
tation 
index composites. 
The raw data are packaged as continuous 
orbital segments of the afternoon AVHRR 
acquisitions. The orbital segments are 
produced by joining several consecutive 
observations, obtained along an orbit by the 
various ground stations, into a single 
continuous data segment for that orbit. The 
size of the orbital segments vary with the 
amount of land data covered by the orbit. 
Orbits beginning over Australia and 
continuing over Indonesia and Asia are 
nearly pole-to-pole. The data volume of a 
pole-to-pole segment is nearly 250 
megabytes. The data volume of the 
shortest segments, such as one over the 
Alaska peninsula, is approximately 50 
megabytes. There are 14 orbits per day 
with an average total daily volume of 2 
gigabytes. The raw data are distributed in 
NOAA Level-1b format (Kidwell, 1991). 
The 10-day vegetation index composites 
are multiband, georegistered, global data 
sets. The 10 bands of the composite 
product are comprised of: AVHRR channels 
1-5, NDVI, satellite zenith, solar zenith, 
relative azimuth, and date of observation. 
The five AVHRR channel bands are stored 
as 16-bit. Each of which has a data volume 
of nearly 1.4 gigabytes. The other five 
bands are stored as byte data and have a 
volume of nearly 695 megabytes each. The 
total data volume of a single 10-band 
composite is approximately 10.5 gigabytes. 
Fifty-four 10-day composites will have been 
produced by the end of 1996, and the 
combined archive of orbital segment and 
10-day composite data will exceed 4.0 
terabytes. 
Compression technique for global 10- 
day composites 
Data compression is an important 
91 
management tool for image data, especially 
lossless compression (Nelson, 1991). Data 
compression reduces the size of files to be 
transferred on the network, and allows for 
more data to be placed on physical media. 
In addition, it reduces the storage capacity 
required to maintain the data online or 
near-line at the distribution site and the 
user's site. However, decompression is 
usually necessary before the data can be 
analyzed using traditional image processing 
and analysis systems. The decompression 
step can be time consuming and require 
that the processing or analysis system store 
and manipulate the full data set. 
The data volume of a multiband, 10-day 
global composite, at 1-km resolution, in the 
interrupted Goodes homolosine map 
projection (Steinwand, 1994) is 10.5 
gigabytes. Although most users may not 
need or use all ten bands of a 10-day 
composite, the practical use of the NDVI 
data as a time series would require 
approximately 25 gigabytes for a one-year 
time period. 
A data compression technique has been 
developed specifically for use in the 
storage and distribution of the global 10- 
day composite data (Kess et al., 1994). 
The technique functions on a single band at 
atime. Each band is divided into a matrix 
of 128 by 128 pixel blocks. Blocks that are 
entirely a single value such as ocean, or 
the interrupted area created by the map 
projection, are compressed with run length 
encoding. The land data blocks are 
Huffman encoded and ordered so that each 
resolution level can be progressively 
decompressed. The geographic 
coordinates for each block and all of the 
information necessary to decompress the 
data in that block to different resolutions is 
stored in a header file. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996