718 
were taken using a matrix camera linked to the 
image processor, which recorded images on 35mm 
colour slide film. 
The colour slides were projected onto transparent 
overlays using a 'Lamprey box' ( see Figure 2) . 
From these projections, details of drainage 
systems were drawn onto the transparent overlays 
and then input to a Hewlett Packard digitiser, 
coupled to a BBC microcomputer. A series of 
software programs were applied to calculate stream 
lengths and lake areas. Quantitative comparisons 
were made between various Landsat MSS and Landsat 
TM enhancements,the Black & White aerial 
photography and drainage details extracted from a 
1:50000 Ordnance Survey map of the area. 
transparent overlay (on glass) 
Z' 
The black & white aerial photographs were 
examined using a Wild ST4 with 8x magnification. 
Drainage details were recorded on acetate sheets 
and a mosaic of sheets created to cover the 
research area. Details from the mosaic were 
transferred to a single transparent overlay and 
input to the digitiser. 
An initial field survey of the area was carried 
out from 9-11 March 1985. This survey 
concentrated on recording stream width 
measurements and a first look at the general 
landcover types in the area. The measurements 
were taken as "ground truth" for comparison with 
interpretations of each imagery type, for 
quantitative assessment of image capabilities. A 
further field survey was carried out from 24-26 
June 1985, to coincide with the season of image of 
acquisition. 
3 RESULTS AND DISCUSSION 
In comparing Landsat MSS and Landsat TM imagery 
the most striking difference is the effect of the 
increased spatial resolution and spectral 
possibilities of Landsat TM, permitting near 'true 
colour' representations, not unlike small scale 
colour aerial photography. The increased spatial 
resolution results in substantially easier 
recognition of terrain features, this being aided 
by shadow effects which provide some idea of 
topography. Image interpretation is further aided 
by band combination possibilities which 
facilitate the creation of images with similar 
colour renditions to those experienced in the 
field. These effects are subjective ones affecting 
the visual processes and as such cannot be 
quantified, however they result in much easier 
image interpretation. 
When examining individual Landsat TM bands, 
bands 4 & 5 revealed the most detail, though a 
three-band combination is more useful in 
discriminating streams and rivers from their 
surroundings by providing a 'true colur' 
contextual background. Band combinations 1,4,5 ; 
2,4,5 and 3,4,5 assigned respectively to blue, 
green and red colour guns on the image processor, 
proved to be the best band combinations. There is 
little to choose between them though the author 
prefers the 1,4,5 combination. Two-part linear or 
manual stretches were applied to these band 
combinations. The use of principal components 
analysis was found to be less informative than 
simple band combinations. Edge enhancement was 
found to increase drainage detail, the Laplacian 
filter, Sobel operator and directional fitering ( 
Seidel, Ade and Lichtenegger, 1983) using 3x3 
kernels proving the most successful. 
When enhancing the Landsat MSS imagery the best 
results were obtained using a standard 4,5,7 band 
combination assigned respectively to blue, green 
and red, with a gaussian stretch and Laplacian 
filtering of band 7, using a 3x3 kernel. 
The drainage network was most easily discerned 
over moorland areas, less easily so in 
agricultural areas where there were small field 
sizes and with some difficulty in thickly wooded 
or forested areas, where the spectral contrast is 
small. Discrimination between roads and rivers 
which frequently proves problemmatic using Landsat 
MSS imagery is greatly reduced in the case of 
Landsat TM due to topographic and contextual 
effects. 
A study of 32 small lakes in the study area and 
surrounding region revealed that the Landsat TM 
imagery could be used to identify water bodies as 
small as 0.3 hectares in areas of open moorland, 
however an overall figure of 0.6 hectares is more 
realistic due to difficulty in identifying small 
water bodies within forested areas. This compares 
with figures of 2.4 hectares for Landsat MSS and a 
suggested value of <0.5 hectares for SPOT in 
multispectral mode i.e. with a 20 metre- spatial 
resolution ( Chidley and Drayton, 1986). Using 
aerial photographs the spectral contrast between 
water bodies and other terrain features was 
sufficient to enable lakes of 0.2 hectares and 
smaller to be identified. 
Table 2a Water body identification. 
Imagery Resolution Minimum area of lake 
detected 
Landsat TM 
30 m 
0.6 
ha 
Landsat MSS 
80 m 
2.4 
ha 
SPOT XS 
20 m 
<0.5 
ha 
B&W Photos 
1:50000 scale 
<0.2 
ha 
Landsat TM 
imagery permits 
much easier 
landcover 
identification than using Landsat MSS imagery. In 
addition to areas of open water, areas of forest, 
moorland, lowland agriculture and urban areas can 
be easily and accurately identified.Using aerial 
photography the various textures enable 
discrimination between these major landcover 
types. Relation of these landcover types to their 
hydrological characteristics is important for 
hydrological modelling. 
Use of a 2x zoom on the Landsat TM imagery was 
found to provide the optimal imagery for 
interpretation. The time required to analyse this 
four-fold increase in quantity was more than 
compensated for by the much greater ease of 
interpretation. However, over large areas this 
method of analysis may prove prohibitively 
expensive. Little geometric distortion was found 
in the TM imagery and the minor necessary 
corrections were 'manually' applied. 
Table 2.Comparison of channel lengths and drainage 
densities. 
MSS TM B&W Photos Map 
Total Channel 25.21 88.75 121.50 272.24 
Length (km) 
Drainage 0.11 0.38 0.52 1.15 
Density 
(km/km 2 ) 
Table 2 gives details of total channel length 
and drainage density identified on all three types 
of imagery as compared with details extracted from 
thel:50000 
borne in mi 
1: 50000 ma 
the area. ( 
very poor, 
Landsat TM 
readings th 
Table 3) . 1 
to over hai 
streams les 
Table 3. 1 
Landsat TM 
Streams ove 
metres wide 
Streams ove 
metres wide 
The main d 
on the enha 
The detail 
accuracy in 
than is pos 
smaller sti 
also be i 
inconsiste 
identifiabl 
with minor 
Landsat MSS 
delineating 
(of which 
streams as 
sporadicall 
S 
S 
Figure 3. C 
When str€ 
scheme of 
following r