787
The method being prototyped involves the production
of three themes: ground cover, present land use and
topographic features. The area chosen for this initial
work is in the southern interior of the Province,
specifically NTS map sheets 82 E (1:250 000) and 82
L/3 (a 1:50 000 map sheet, but mapped at 1:20 000).
The ground cover database provides the generalized
type, extent and distribution of present vegetation
cover. The main criteria for classification is vegetation
structure and composition (i.e. physiognomy). The
initial classification from the satellite imagery is refined
using the relationship between specific vegetation
distribution and topography. The ground cover theme
is extended to include some detail on water bodies,
wetlands, rock outcrops and the like. A ground cover
classification schema is given in appendix "A".
The present land use theme is based on a modified
form of the "Land Use Classification in British
Columbia" (Sawicki et al., 1986). The level of
classification is restricted to area-based uses versus site
uses. Site uses of the land are generally not extensive
enough to be identified on satellite imagery. It should
be noted that a combination of spectral classification
of satellite imagery and interpretation is used to
produce the present land use classification.
The topographic features database is derived by
segmenting the digital elevation model (DEM) into
significant landscape units. These units are described
in both raster and vector format. Watershed
boundaries are delineated. Attributes that are attached
to these units are the maximum, minimum, and mean
values for: elevation, slope and aspect. Also included
are: area, surface irregularity, and surface shape.
Potential solar radiation received is also calculated. It
is anticipated that these attributes will significantly add
value to the final Baseline Thematic Map product for
many users.
The satellite imagery utilized is presently Landsat TM,
which has good spectral information for the
identification of ground cover and for interpreting
present land use. Two versions of the product are
being prototyped, corresponding to the two available
topographic data sets. The digital elevation model and
base map for the 1:250 000 version will be supplied
from scanned 1:250 000 NTS map sheets. Energy
Mines and Resources Canada has scanned all of the
84 sheets covering B.C. and these digital map files are
available from them. The Terrain Resources
Information Management (TRIM) program (the
creation of 1:20 000 digital base mapping for B.C.) will
supply the digital elevation model and the base map
for the 1:20 000 version. Presently there are about
1200 of the 7000 map sheets completed with a current
production rate of approximately 500 map sheets per
year.
3 METHODOLOGY
For the 1:20 000 version the new digital mapping from
the TRIM program forms the base map. For the
1:250 000 version product, scanned 1:250 000 NTS map
sheets will provide the base map. The datum for the
prototype work is NAD 83 (North American Datum
1983). It is envisioned that all provincial mapping will
migrate to NAD 83 over the next five years.
Transformations to other datums (NAD 27) and
projections are well defined for those users desiring
them.
The first manipulation of the input data is to co
register the different data sets (imagery, DEM’s, and
other ancillary map data). To avoid any systematic
subpixel mis-registration all rasters follow the same
convention about the origin with respect to the
reference UTM grid. UTM coordinates ending in 00
(i.e. every 100 metres) align with the boundary
between pixels. Pixel sizes are chosen in a nested
hierarchy as follows: 6.25, 12.5, 25, 50, and 100 metres.
The actual size is dependant upon the data utilized
and the map scale produced.
For areas of high relief Landsat TM imagery has
geometric distortions due to the panoramic view of the
sensor. At the edge of the 185 kilometre wide image
swath the look angle is 7.47 degrees, causing a
horizontal displacement of 30 m for every 230 m
change in ground elevation (Wong et al. 1981). Many
parts of B.C. have relief in the order of thousands of
metres. This is an unacceptable amount of distortion
for 1:20 000 mapping.
Correcting for relief displacement is accomplished
using a DEM. We have demonstrated that the
1:20 000 TRIM data produces a DEM that is
sufficiently precise to allow correction of Landsat TM
to plus or minus one pixel (25m) accuracy. Currently
we are investigating the precision obtainable utilizing
1:250 000 topographic data.
The present land use classification was produced
through interpretation of a transparency of the Landsat
TM imagery using the PROCOM projection device.
The transparency was the same date as the digital data
with bands 3, 4 and 5 portrayed as blue, green and
red. The interpretation depends upon colour, tone and
pattern recognition and adjacency considerations. The
initial interpretation work was done at 1:100 000 scale
with the results then being digitized into the existing
digital base maps.
Image classification for ground cover identification
proceeds from field work to selection and refinement
of training sites on the satellite imagery. To reduce the
complexity of the task the classifications where
stratified according to the land use classification. Thus
many separate classifications (eg. agricultural,
rangeland, logging clear cuts, etc.) with a restricted
number of classes where performed. Within these
strata single ground cover classes are described by
more than one set of training sites so illumination and
ecological differences due to topographic position are
accounted for. After classification with these separate
training sites, the results are merged into a single class
type.
Post-classification processing is of two main types:
context filtering to transform the raw image
classification into a more acceptable map like product
and modification of the resulting classification through
the use of ancillary (primarily topographic) data.
Context filtering makes use of the similarity between
classes as well as length of common boundary and
