You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Remote sensing for resources development and environmental management
Damen, M. C. J.

SPOT simulation 1982-06-30 Time: 12.32.42PM
IR Imagery 1982-06-30 Time: 12.32.42PM
BW Imagery 1975 Summer (July).
2.4 Laboratory work
Digital and visual classification were carried out at
Telemetry Station (TTS).
2.5 Classification
The procedures followed are illustrated in the flow
chart (figure 2). A supervised classification method
has been elaborated using MAX-LOG-LIKELIHOOD classi- .
fier with treshold of 3.00, 4.00 and 5.00, respective
ly. In addition to this, several caanael combinations
and ratios were used both for digital and visual
2.6 Accuracy assessment
Then®are two major types of accuracy assessment prose-
1.Non-site-specific (total area extent)
2.Site-specific (location).
Non-site-specific accuracy is usually expressed as the
similarity between the total numbers of hectares in
each vegetation-cover type as determined by a Landsat
or SPOT classification, compared to the corresponding
total area determined from the digital vegetation map.
The non-site-specific method compares only the total
area without regard to location. Site-specific accu^
racy, however, considers the spatial nature of the da
ta when two spatially defined data sets (one ground
truth) are registred and compared for the amount of
agreement (Reichert & Crown 1984). The chosen method
in this study is the no-site-specific method.
The results of the prosject is here in this chapter
presented and discussed. The basic tabels and figures
from the prosject is not presented here, but I refer
to my thesis-report (T0mmervik 1985a).
3.1 Floristic and Phytosociology
The area is very varioues and rich what species con
cern, and of the the 489 plant taxa at species and
subspecies levels encountered in the ground truth pro
gram, 371 were vascular plants, 82 mosses and 36 lich
ens. The result of the phytosociological study which
was carried out during this prosject, was that 39 main
vegetation-units on several levels in the phytososio-
logical hierarchy were picked out.
3.2 Digital vegetation map
A digital vegetation map has been generated on basis
of the ground truth program, and the legend of this
map consists of 31 vegetation cover types. The work
with the digital vegetation map has shown that it is
a good tool for resource studies and vegetation map
ping purposes, and that it can be updated very easily.
3.3 Digital image processing
Digital image processing was done both on SPOT-simu-
lated imagery and Landsat 5 TM imageries. Unfortuna
tely both the spring-scenes from 1982 and 1984 were
taken too early in the spring due to snowcover, and
the autumn-scene was taken to late in the autumn to
Figure 2. Methodology flow chart for detection of the
vegetation cover types.
give a sufficient basis for a good digital classifi
cation .
3.4 Digital classification
Digital classification was based on a supervised met
hod using MAX-LOG-LIKELIHOOD classifier.
The resulting themes left many pixels around vege
tation type boundaries unclassified. Boundary pixels
presented a special problem, as they represented port
ions of different vegetation cover types. Their values
were a function of the amount of the area of each
vegetation type within the pixel and the relative ref
lectance of each material as a whole. This was also
the result within the resulting themes, and this was
a result of the variation in phenology, heterogeneous
vegetation and distribution of snowcover.
The interpretation and classification were checked
by comparing the classified imagery with the digital
map, and the accuracy of the interpretation was asses
sed on a quantitative basis. Two areas within the
digital map area were checked, Dividal-Cavarre and
Saratr0a-Habafjell (Table 1 and Table 2).
3.4.1 Alpine vegetation
Hi Extremely dry shrub were rather good mapped by the
two sensorsystems (Landsat 5 TM and SPOT-HRV) with a
optimal accuracy of 89 % for the SPOT-simulation
(treshold: 3.00) and 93 % for Landsat 5 TM (treshold:
5.00). For H7 Rich shrub was the Landsat 5 TM-sensor
the best sensor for mapping of this vegetation cover
type with a accuracy of 89 % (treshold: 3.00). H2 Dry
shrub was rather bad detected, 36 % for the SPOT-simu
lation and 66 % for Landsat 5 TM (treshold: 5.00),
3.4.2 Mire vegetation
Q5 Rich mire and P2' Wet shrub were bad detected and
mapped, due to the high amount of watercontent and the
distribution of snowcover at these types of vegetation.
But Q4 Poor mire (intermediate) showed a very good
accuracy of 83 % for the SPOT-simulation (treshold: