313 
were combined into one sample for the concerning vegetation type. 
The spectral separability of vegetation types on the Landsat images was examined it. scatter diagrams 
with the ellipses of the selected training samples for each vegetation type. When the ellipses of types in the 
plot show extensive overlap, then the types cannot be distinguished in the bands that are graphed. 
A classification of each image was carried out using the maximum likelihood algorithm. Also a 
multitemporal classification was done using the bands of all three images. The accuracy of the classifications 
was assessed by comparison with the vegetation map, the aerial photographs and field knowledge. 
4.3. Results and discussion 
4.3.1. PS 11 measurements. The spectral curves of the vegetation types in May and in September (figure 5) 
showed that there were differences in reflectance between types as well its in time. The differences between 
the average values were largest in the infra red region (> 750 nm) but the variation within one type was also 
largest in this part of the spectrum. This near infra red reflectance of vegetation canopies is largely defined 
by the canopy structure and amount of vegetation which can be attributed to multiple scattering of this 
wavelength between layers of the vegetation (J.E. Colwell. 1974; Tucker ei al ., 1975). In the visible part of 
the spectrum differences were largest around 555 (green) and 675 (red) nm. The reflectance of visible light 
(400 - 700 nm) of leaves is amongst others influenced by absorption by pigments which is largest in the blue 
(ca. 450 nm) and the red (650 nm) region of the spectrum. As a consequence visible light reflectance is 
highest in the green region (ca. 540 nm) (De Boer, 1990). 
In May the reflection of the visible light (430-700 nm) was similar for the Scirpus and the Trigochlin 
vegetation but it differed for the infra red reflection. Elymus and Agrostis showed the highest and very 
similar green (550 nm) reflection but they separated in the red valley (670 nm) related to chlorophyll content 
and in the infra red region. The spectral responses in the infra red region were the same for Phragmites and 
Scirpus but they differed in red reflectance. The red dip was missing in the Phragmites spectra. This will be 
caused by the many dead stems which still remained from the last year and extended above the green 
vegetation cover. The Agrostis vegetation showed the highest green/red ratio. For this vegetation the 
influence of soil reflection will be reduced due to a greater cover. Also less dead tr • 'ial was present. 
In September Elymus was clearly distinguishable in the red and the infra red region. The curve of 
Scirpus was rising from green to red because most of the vegetation was dead (less chlorophyll) and the 
cover was less (less chlorophyll and more bare soil influence). Scirpus and Trigochlin were spectrally hardly 
separated in September. In this period Phragmites has large inflorescences which could account for the steep 
path of the infra red curve and the low reflection in the visible region (shadow). 
Reflection of the visible light had increased in September for Agrostis, Trigochlin as well as Scirpus 
whereas the infra red reflection had decreased in comparison to May. This can be described ? to the lesser 
vegetation cover in September and, consequently, an increased soil influence on reflection. Besides more 
dead material was present in this period. In September the Elymus grasses were very tall and were laying 
horizontally on the ground. Perhaps this could explain the increased reflection in September. The visible light 
reflection of Phragmites did not change but the inffa red reflection had increased in September probably due 
to the earlier mentioned inflorescences. 
4.2.3. Landsat-TM. On the scatter diagrams with ellipses of the training samples of the different vegetation 
types (figure 6) in all three images (May, July and October) the vegetation types were best separated with 
band combinations of band 3 (630-690 nm), 4 (800-1100 nm) and 5 (1.55-1.75 nm). Band 2 (520-600 nm) 
was strongly correlated with 3 and band 5 with 7 (2.08-2.35 nm). The thermal infra red band 6 (10.4-12.5 
nm) did hardly make any distinction. 
The vegetation types were spectrally best distinguishable in July. In this month the ellipses of 
Phragmites, Elymus and grazed parts were well separated. Also the Scirpus-with Atriplex was separated 
although its ellipse was close to the other ellipses of Aster, Puccinellia and Spartina. These last three types 
did all overlap. Scirpus- without Atriplex was best separated with band combination 3,4 but there was still 
some overlap with Aster, Puccinellia and Scirpus- with Atriplex. 
In the May and October images only the Phragmites and Elymus ellipses were separated although less 
distinct as in the July image. In October the pixel values were also lesser due to the lower sun angle. 
As expected from the scatter diagrams the classifications of the May and October images with band 3, 
4 and 5 did not give satisfactory results when they were compared with the vegetation map and the aerial 
photographs. The July classification on the other hand was much better. The patterns in the maximum 
likelihood classification of the July image of Elymus, Scirpus, Phragmites and grazed parts were very similar