3. PRODUCTION OF THE MODEL INPUT DATABASE 
The production of data for use as input to the 
erosion risk model can be separated into three 
main sources of data : thematic maps, 
topographic maps and remote sensing. All these 
data were converted into a common form using 
the PC based ICONOCLAST Image Processing and 
Spatial Information System, developed by Image 
Technology Systems Ltd. 
Prior to the digitisation process the effective 
scale of the required data was chosen. As the 
input to the model and manipulation of these 
data is essentially a grid cell (raster based), the 
size of the cell is the major factor. This was 
chosen to be 150m, giving a thematic database of 
500 x 700 cells (pixels). 
3.1. Thematic Maps 
The production of the thematic map database 
produced inputs related to the resitance of soil 
to erosion, namely the Geology and Soils of the 
region. 
Digitisation of soil and geology class boundaries 
was achieved using both a digitising tablet, and 
a process of video digitisation and boundary 
extraction. Both resulted in a class boundary 
vector database extracted in the pre-defined 
co-ordinate system. 
The boundary data from various mapped sheets 
was then aggregated and any map sheet edge 
problems resolved. The boundaries were then 
used in a graphical seed fill process, thus 
converting each pixel within a boundary to a 
pixel value indicative of class. The boundaries 
are then removed by a process of line 
replacement from adjacent pixels. 
In this type of raster based Spatial Information 
Systems, pixel brightness is related to a table 
describing class attributes. Thus to obtain the 
class of any pixel is a simple case of looking up 
the class table. The location and topology of each 
pixel is only known from its position in the grid. 
A number of other thematic databases where also 
created at this stage to assist in future work. 
These included Land Management Units, (official) 
Land Use and Stream Sub-catchments. 
3.2. Topographic Maps 
Given that slope and slope length, aspect and 
elevation are important considerations in the 
model a digital elevation model (DEM) had to be 
constructed. In may parts of the developed world 
DEM are commercially available, for developing 
countries this information is often restricted. It 
is therefore necessary to have tools for 
generating a DEM from any available data. For 
the study area large scale topographic maps 
were not available, due to government 
restriction, and an alternative source of data was 
needed. Tactical Pilotage Maps are generally 
available worldwide and this was used as a 
source of all topographic information for the 
study. 
Individual section of the pilotage map were 
raster digitised at high resolution to facilitate 
the extraction of contour data. All relevant 
contour information was extracted by a manual 
process of on-screen digitisation - an extremely 
laborious task. Other topographic features such 
as roads, built areas and the drainage network 
were also extracted to assist in the geometric 
rectification of the remote sensing input. 
The elevation data consisting of contours and 
spot heights was then converted to a raster 
representation of height by an interpolation 
process. A number of methods were tested 
including various Least Squares (Jancaitis and 
Junkins, 1973) and Multiquadric (Hardy, 1971) 
procedures. Later work suggested the more 
efficient use of triangulation routines. 
Given a satisfactory raster representation of 
elevation, where pixel brightness is directly 
proportional to height, slope, aspect and other 
related measures can be derived. The algorithm 
used for generation of slope and aspect for the 
model was a modified version of the algorithm 
developed by Ritter (1987). 
3.3. Remote Sensing 
The difference between land use detailed on 
available maps and actual land use was 
considered to be significant; particularly in 
upland areas where considerable de-forestation 
had occurred. Satellite remote sensing was seen 
as a means of providing up to date land use 
information at sufficient scale for use in the 
model. 
A Landsat MSS scene was acquired and 
subsequently used to produce a land use map 
using multi-spectral classification techniques. A 
1536 x 1536 extract was taken from the full MSS 
scene. This was first used to define a number of 
ground control points using on-site experience 
and the topographic map. 
The extract was then corrected to the database 
co-ordinate system, primarily for use as a 
frontispiece in a report overlayed with 
topographic features and catchment boundaries. 
What is also extremely evident on this image is 
the siltation of the reservoir compared to the 
downstream irrigation channel. 
The extract was then used to define training 
areas for land use. Training statistics were 
produced and used in a Maximum Likelihood 
classification of the extract. The accuracy of the 
classification was visibly checked for a further 
set of areas for which ground truth was known. 
In light of the available alternative, and 
subsequent empirical processes involved in the 
model, the classification was considered to be 
sufficiently accurate and the classification was 
then corrected to the database co-ordinate 
system. 
131