quadratically with the number of bands to classify. A 
classification algorithm based on nonparametric 
techniques, the piecewise linear classifier, was 
described by Lee & Richards (1984). The 
classification time increases only linearly with the 
number of bands included. To classify 8 bands into 6 
classes, the average classification time was 
approximately 10 times as high for the ML 
classification compared to the piecewise linear 
classifier while the accuracy was much the same (Lee 
& Richards 1984). 
5.2 Vegetation indices 
The spectral properties of green vegetation in the 
visible and the near infra-red part of the spectrum 
follow a very distinct pattern, see Table 2 and 
Figure 1. The relationship between visible red and 
near infrared have commonly been used for studies of 
vegetation amount (e.g. Kauth & Thomas 1976 Tucker & 
Maxwell 1976, Richardson & Wiegand 1977, Tucker 1979, 
Pollock & Kanemasu 1979, Curran 1980, Wiegand St 
Richardson 1982). The aim is to reduce the number of 
wavelength bands to a single band, highly correlated 
with amount of vegetation or accumulated green bio 
mass on the ground. 
We can see two different approaches to the 
generation of vegetation indices, ratios and linear 
combinations of spectral bands. A wellknown example 
of the first type of vegetation indices is the 
normalized difference vegetation index NDVI (Curran 
1980, Holben et.al. 1980, Justice et.al. 1985). The 
Tasseled Cap transformation by Kauth St Thomas (1976) 
and Crist Sr Kauth (1986) is one of the most commonly 
applied transformations of the latter kind. The 
correlation coefficients for relationships between 
different vegetation indices and measured vegetation 
amount on the ground were very high, 0.86 - 0.90, 
using hand-held radiometer (Tucker 1979) . Correlation 
between Landsat MSS derived vegetation indices and 
vegetation amount is not as high, 0.6 - 0.7, but 
still 95% significant (Richardson & Wiegand 1977) . 
In the literaure we can find a large number of 
different vegetation indices, based on either linear 
combinations or ratios. Perry St Lautenschlager (1984) 
found and tested "some four dozen of these formulae". 
Their conclusion were that for all practical purposes 
several widely used indices are equivalent. What is 
needed now is empirical research aimed at answering 
some of the following questions: 
- In what biomass intervalls (e.g. desert, grassland, 
woodland, forest) are vegetation indices most 
effective. 
- How does the type of vegetation (field cover vs. 
canopy cover) affect the relationship. 
- How to compensate for the influence from varying 
soil colour. 
All these questions are closely connected to the type 
of satellite data that is used, especially when using 
NOAA data, covering vast areas and recorded off-nadir 
at high angles, it is important to know more about 
these problems. 
5.3 Accuracy of remote sensing 
A study area that have attracted comparatively 
little attention is methodology for estimation of 
accuracy of remote sensing results. Nevertheless is 
proper estimation of accuracy of ultimate importance 
for all development of new techniques. Estimation of 
accuracy involves sampling of reference data and 
the application of statistical measure of 
correspondence between the remote sensing product and 
the reference data. These procedures have been 
described by e.g. Hord & Brooner (1976), Hay (1979), 
Turk (1979) and Hellden (1980). Rosenfield & 
Fitzpatrick-Lins (1986) described a "coefficient of 
agreement" (based on work by Cohen 1960 & 1968) as a 
measure of classification accuracy as a whole and for 
the individual categories. It desirable that the 
remote sensing community adopts one methodology for 
estimating accuracy, and that this also becomes 
widely used for different remote sensing applica 
tions . 
6. GEOGRAPHICAL INFORMATION SYSTEM, GIS 
Environmental information presented in map form is a 
necessary instrument for planning and management of 
natural resources, as well as for research on the 
distribution and allocation of resources. Maps can be 
seen as a means for communication between 
researchers, planners and decision makers. The amount 
of information that can be presented in map form is 
tremendous. Both status, trends and projections can 
be presented in a conceptually simple way. 
A geographical information system (GIS) is a 
computer based tool for storing analysing and 
presenting spatial data. All kinds of data that are 
spatially determined can be stored, updated, analysed 
in a multivariate fashion and presented, either in 
map form or as tabular data. Data at different scales 
can be linked to each other and that provides a base 
for generating spatially full-covering information 
from case studies over limited regions. 
Traditional computer assisted cartography and GIS 
was entirely vector based, i.e. the data was 
represented by x, y and z coordinates. The technical 
development of image processing has lead to an 
alternative to vector storage and analysis, the 
raster form, where every point on the map, in the 
form of a rectangular grid system, is stored. Both 
systems have their advantages and drawbacks, see 
Table 5. 
Table 5. Positive versus negative features of raster 
and vector representation of cartographic data. 
vector data raster data 
+ storage efficient + areas are conveniently 
represented and area 
operations (e.g. over 
laying) are efficient 
+ object oriented, + easily integrated with 
searching and retrieval 
of objects is easy 
+ geometrical and scale + 
operations are efficient 
- access by location is 
inefficient 
- thematic overlaying is 
inefficient 
- input (raster scanner) 
and output (screens and 
plotters) devices are 
inefficiently used. 
A landscape is composed of a mosaic of different 
features, physical, cultural and economical. Some of 
these features can be regarded as more or less 
static, they do not change rapidly over time (e.g. 
geology, climatic zonation and settlement pattern). 
Other must be regarded as extremely dynamic, they 
change rapidly over time. Dynamic factors may be 
e.g. image data in 
raster format 
output devices (screens 
and plotters) are 
efficiently used 
space consuming 
geometrical and scale 
operations are time- 
consuming 
searching for and ret 
rieval of objects is 
inefficient 
sensitive t 
or more oi 
static fact 
know, they 
landscape, 
topography, 
vulnerabilit 
vegetation, 
agents caus 
degree of cl 
dynamic and 
of different 
information 
ideal tool, 
are describe 
a) Improve 
remote se 
data intc 
(type anc 
lary date 
data, the 
Therefore 
sensing v 
b) Generatic 
•nation, 
informati 
proper ue 
for difj 
most cor 
maps. n 
backgroui 
grazing i 
informatj 
facility 
result c 
form. 
c) Integrate 
climate, 
drought, 
features 
last dec 
sequences 
these cc 
disciplij 
scales, 
several c 
to carr 
relation; 
will be ; 
7. geograph: 
Generally 
frequently 
describing 
biomass or ; 
resource mt 
further, t< 
ACCESSIBILI r 
Data whi< 
statistics 
remote sens: 
lation can 
data (in ; 
meteorologii 
that are i 
populations 
in this way 
or valid oi 
influence o: 
1985) can : 
of data, an 
raster form, 
monitoring : 
ibility of 
integrated