Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986 
Multi-temporal Landsat for land unit mapping on project scale 
of the Sudd-floodplain, Southern Sudan 
Y.A.Yath 
Jonglei Executive Organ, Karthoem, Sudan 
H.A.M.J.van Gils 
International Institute for aerospace survey and earth sciences, Enschede, Netherlands 
ABSTRACT: The Sudd floodplain is extremely flat, covered mainly by grasslands, only locally interrupted by 
clusters of fields and seasonlly flooded by rain and/or river water. Since photo-interpretation is based on 
relief, vegetation structure and field pattern, such technique is not satisfying for the Sudd floodplain. 
Moreover, the most important environmental component - the flooding - can be assessed only on sequential series 
of photographs. 
Studies of the Sudd-floodplain on regional scale have been carried out as environmental impact study for the 
Yonglei Canal project. However, such information proved of limited value on project scale. For the latter a 
combination of Landsat, aerial photography and field survey has been tried and proofed successful. The methodo 
logy, results and limitations are outlined for application in similar floodplain areas as there are the Llanos 
in Southern America, Zambesi floodplain, Kafue flats and the Okavango swamps in Africa. 
Material 
The available remote sensing material for the area 
was at the time of the survey (1983): 
- Black and white panchromatic aerial photography 
scale 1:40 000 from december 1980. 
- Computer Computable Tapes (CCT) of four sequential 
Landsat MSS Scene 186/055 
02/05/79 
11/10/79 
31/12/79 
18/01/80 
Of these the first two and the last are used for 
multitemporal imagery. This sequential Landsat series 
is the first and the last set available for this area 
between 1972 and today with such short intervals. 
95 releves following standard ITC procedures (Gils 
et.al. 1984) were available as field reference 
materials. 
Methods 
Photo-interpretation followed the landscape-guided 
method (Gils et.al. 1984). 
Pre-processing of Landsat CCT's started with radio- 
metric corrections (sun angle, haze) and producing 
square pixels using standard IPL - ITC methods (Mul 
der 1982). The geographical correction could not be 
carried out by routine procedures due to lack of 
topographical orientation points in the survey area. 
Therefore the three Landsat images have been super 
imposed visually. The normalised vegetation index 
IR-R is calculated for each of the three sequential 
IR+R scenes. 
Each of the three temperorally different vegetation 
indices has been coded by a colour. The vegetation 
index values have been scaled from zero to hundred. 
The highest to the lowest vegetation index in January 
has been assigned a corresponding hue in red. 
Similarly the vegetation indices in May are coded in 
green and those of the October image in blue. The 
three seasons superimposed produce a coloured map on 
approximate scale 1:250 000. For details and back 
ground see Mulder (1982). 
The scene of October 1979 has been subjected - 
after standard correction (compare under sequential 
imagery) - to a supervised classification (SC) with 
the help of the field data. 141 Pixels were located 
on the scene were the land units were known. The Red 
(x-axis) and Infrared (y-axis) radiation intensities 
of the 141 samples were plotted in a feature space. 
The 141 were classified first into their land unit. 
A cluster analysis was performed and regions were 
delineated. Each delineated region in the feature 
space was colour coded. Thereafter all the pixels 
were given a colour according their place in the 
classified feature space. 
The land units are named in the legend according to 
their vegetation, since this is their main charac 
teristic to be observed both on the image and on the 
ground, due to the flatness of the area and scarcety 
of crops and artifacts. 
Result and discussion 
The supervised classification of the Landsat image 
resulted in 8 main legend units implying also 8 main 
vegetation legend units. The legend units are repre 
sented in the legend of figure 1. The land units 
might be compared with those obtained with other 
image processing techniques of Landsat data as there 
are the false colour composite used by the Mefit- 
Babtie (1983) survey and the multitemporal image of 
the present paper. 
The standard Landsat image plus aerial photography 
interpretation by the Mefit-Babtie (1983) team re 
sulted for the same area covered by the present paper 
in five vegetation legend units. There is more 
differentation in the Toich area with the supervised 
classification of the Landsat data as compared with 
the map based on standard Landsat products. However, 
the lower resolution of the standard might be an 
artifact, because the map from which this conclusion 
derives is on a scale 1:500 000 and the variety 
within the Toich could possibly not be mapped on this 
scale, but is cartographically representable on the 
250 000 scale of the supervised classification. 
The field sampling - on which the supervised 
classification was based - was designed in the hypo 
thesis of an east-west catena of land (including 
vegetation) units. This catena is well expressed 
both in the image (fig.l and fig.2) and the Mefit- 
Babtie map (1983). However, the images (fig.l and 
fig.2) show a north-south catena in the toich form 
high (dry) to low (wet) sofar not noticed. This 
north-south catena within the Toich still has to be 
confirmed by field observations. 
The multitemporal image (fig.2) has basically the 
same land unit pattern as compared with the super 
vised classification. The multitemporal image (fig.2)