788 
minimum area rules for amalgamating groups of 
pixels. This kind of filtering results in a 
cartographically acceptable map and is also required to 
reduce the number of homogeneous areas (polygons) 
as many GIS’s are limited in the number of polygons 
that they can handle. A record is kept of the 
percentages of the original classes that were 
amalgamated to form the single class output polygons. 
Improvements in digital classification through the use 
of ancillary data (including digital elevation data) can 
be achieved in three ways: incorporating those data 
either before, during or after classification, through 
stratification, classifier operations or post classification 
sorting (Hutchinson, 1982). Ancillary data is used to 
construct masks within the image analysis system then 
Boolean logic is employed to separate classes that are 
spectrally similar. Equivalent results can be achieved 
through GIS processing, with potentially a wider 
selection of attributes available for post-classification 
accuracy improvement. 
Topography plays a dominant role in many biological 
and geomorphic processes. Terrain segmentation as 
implemented is based on: surface morphology, 
hierarchical watershed boundaries, and potential solar 
radiation received at the surface. Segmentation based 
on morphology begins with irregularly spaced points, 
from which a triangulated irregular network (TIN) is 
created. Units are defined through the amalgamation 
of neighbouring triangles, based on similarity with 
respect to slope and aspect. Watersheds are formed by 
the examination of hydrologic flow across the triangles. 
Radiation based segments are created through merging 
of contiguous triangles based on criteria examining the 
results of shading and shadowing values sampled at set 
intervals from sunrise to sunset over specific days of 
the year. 
At some point the raster products require conversion 
to a vector format to be compatible with most current 
GIS formats. Currently our capability is limited in the 
size of raster supplied as input and the number of 
resultant polygons output. Presently a single value 
attribute can be attached to the vector database 
created, there is a requirement for the transfer of 
multiple attributes, for example the record of the 
original composition of a context filtered polygon. 
4 RESULTS 
The following discussion of results is restricted to the 
1:250 000 version. At the time of writing (June 1990) 
work on the ground cover part of the 1:20 000 version 
is proceeding under contract. 
4.1 Present Land Use 
The following land uses were identified and delineated 
from the satellite image transparency (codes based on 
"Land Use Classification in British Columbia", (Sawicki 
et al., 1986)); 
A000 
rural activities 
A100 
mixture of tillage crops, 
tree and vine fruits 
forage 
crops and 
A130 
orchards and vine fruits 
cooo 
urban built-up areas 
and 
residential 
concentrations 
E100 surface extraction 
FI 10 recent logging (i.e. visible clear cuts) 
Fill clear cuts or groups of clear cuts with up to 
30% standing trees) 
R100 ski hills, parks 
N100 no perceived activity with grass cover 
N200 no perceived activity with forest cover 
N320 bedrock 
N400 lakes 
Because the logging clear cuts were a highly visible 
land use, smaller units than usual for the final map 
scale of 1:250 000 were delineated. This resulted in 
1300 land use polygons in total for the 1:250 000 map 
sheet. 
4.2 Ground Cover Classification 
For the 1:250 000 ground cover classification the 
imagery was decimated to 100 m pixels. This 
represents a reduction in data volume by sixteen. The 
effect of this decimation on the accuracy of the ground 
cover classification was tested for the clear cut areas. 
The overall accuracy for this particular ground cover 
classification (12 vegetation types) was 89% for 25m 
pixels and 93% for 100m pixels. 
Ground truth was derived from fieldwork for ground 
cover within logging clear cuts. Ministry of Forest 
1:20 000 forest cover maps provided the ground truth 
for forested areas for approximately 10% of the area 
under consideration. Intensive agricultural land use 
mapping for three 1:20 000 map sheets provided the 
ground truth for the ground cover within the 
agricultural areas. 
Classification accuracy is presently being assessed. 
Preliminary results for the ground cover within 
clearcuts indicate overall thematic mapping accuracy in 
the 70% to 90% range. 
Context filtering was applied to reduce the complexity 
of the raw classification in a cartographically 
acceptable manner. This process reduced the number 
of polygons by a factor of 50 to 100. Usually, but not 
always, this process would result in increased map 
accuracy. 
4.3 Topographic segmentation 
The TRIM program 1:20 000 topographic data 
provides a dense grid of elevation points from which 
it is possible to construct a high quality digital 
elevation model (DEM). For the 1:250 000 topographic 
data the elevation data is conveyed by digitized 
(scanned) contour lines. Although experienced map 
users find it easy to infer surface information from 
contours typical DEM interpolation algorithms often 
give incorrect results because of the lack of elevation 
data between contour intervals. 
In an attempt to improve the quality of the DEM 
derived from the 1:250 000 data, the density of 
elevation points was increased in two ways. A "z" value 
was interpolated for the hydrographic network based 
on intersections with contour lines. Areas with sparse 
contour lines were manually assigned elevation points 
from 1:50 000 topographic maps. These processes 
resulted in a greatly improved DEM. 
Currently the software employed for topographic