928
2. REMOTELY SENSED DATA
2.1 Data Acquisition
Techniques have been developed for position-based
mapping of surface cover using Landsat digital data
and implemented for a study area in the upper
Kananaskis Valley of southwestern Alberta, located
in part of the Rocky Mountains (see Figure 1)
(Lodwick, 1981; Paine, 1984). The research involved
approximately one-twentieth of an image of the
Calgary scene, taken on September 20, 1975, and
comprised natural areas of woodland and rock. Input
is in the form of digital computer-compatible tapes
of Landsat multispectral scanner data supplied by
the Canada Centre for Remote Sensing in Ottawa. The
output products are maps at scales and accuracies
suitable for a wide range of environmental
applications.
Scale in km
Figure 1. The Kananaskis Valley in Alberta
The analysis procedure comprised four major stages
to preprocess, enhance, position and classify the
data. Firstly, data from the computer-compatible
tapes were reformatted and radiometrically corrected
to remove errors in video response values introduced
during image acquisition. Secondly, the data were
enhanced for investigation and interpretation using
principal components analysis. The advantage of
this technique is that the original information can
be defined in one, or at most two variables, which
are readily interpretable in terms of natural
surface phenomena. Thirdly, the image was
geometrically adjusted, using a second order
polynomial model and ground control points to
spatially resect it to the earth's surface.
Individual pixel values were then resampled to a UTM
grid of 50 m spacing using the nearest-neighbour
technique. Finally, interpretation of the data was
carried out with a supervised classification scheme
using a parallellepiped technique.
The results indicated that the first and second
principal components contained over 99 percent of
the information in the four original data sets.
Classification using these data resulted in accurate
definition of the surface cover of the test area.
The rectification established the positioning
accuracy of these data with an RMS error of less
than 50 m, which is suitable for mapping at a scale
of 1:50,000. The resulting map contained ten
distinct surface cover classes, which were used as
one input data set for the LRIS. The significant
advantage of this approach is that sequential
Landsat imagery enables updating on a regular basis.
2.2 Format Conversion (Raster to Vector)
Remotely sensed data are usually collected in a
raster format and the decision about storage,
manipulation and retrieval of these data often
hinges on the question of raster versus vector
format. Broadly, the methods of describing the
positional extent (or coding) of spatial entities
are either vector-based, using coordinates, or
raster-based, using scan lines. Arguments have been
made in favour of each of the raster and vector
formats. Generally, particular formats and format
conversions are not of concern to the system user
but of the information system itself (Haralick.
1980).
Using vector format, polygons are described by a
series of lines or points, lines are also described
by a series of lines or points (Peucker and
Chrisman, 1975; Edson, 1975) and points are
described either by absolute coordinates (relative
to the coordinate system origin) or by relative
coordinates (relative to a previous point) (Baxter,
1976; Burton, 1979).
Using raster format, polygons, lines and points
are all represented by those parts of an overlain
grid matrix, or system of scan lines, which cover
them. A raster matrix, or grid map may be full,
where all cells or pixels are stored, or sparse,
where only significant cells of each scan line are
stored (Miller, 1980; Barber, 1982). The usual
comment made, when discussing the relative merits of
vector and raster formats, is that vectors are more
efficient in terms of storage space, and rasters are
more efficient in terms of computation time (Barber,
1982).
With Landsat data, storage in raster format is an
appropriate approach. However, when dealing with an
LRIS, where survey data as points and lines are to
be stored, positional accuracy is an important
consideration. Also efficiency considerations
favour storage by polygons (vectors) (Mepham and
Paine, 1986). To polygonize the remote sensing data
set required that the data be smoothed to a level of
generalization which would provide a more efficient
storage format. This smoothing/filtering process
requires careful selection of a spatial filter which
will not degrade the data (Paine and Mepham, 1986).
This resulted in a data set represented by polygons
formed from line segments based on unique points.
3. CONVENTIONAL MAP DATA
3.1 Selection of Data Types
One of the requirements of this prototype LRIS was
for test data that represented a broad range of
spatial land information. Thus, three basic data
types were selected: 1) primary data (positional),
2) secondary data (thematic), and 3) digital
elevation model (DEM) data.
The first two types are fairly standard and have
accepted definitions. Primary land-related informa
tion comprises three main components: 1) geographic
positioning systems established with a range of
survey techniques, 2) mapping systems established
from the basic survey data, and 3) land registries
containing crown land and land titles information
(LRIS, 1981).
Secondary land-related information can loosely be
defined as thematic data (Kozak, 1980). It is
usually represented graphically as a map with the
appropi
These c
land c
use.
also ii
and mai
usually
collect
that se
than tl
changes
require
The 1
data bt
utilize
hand1in
—-Ml
’’•’Me 1
Classes
Survey
Contri
(Primal
Hydrogi
(Second
Geolog}
''Second
Transpc
ation
(Second
Surface
Cover
(Second
Digital
Elevati
Models
(DEM)
With
classes
process
have ma
informa
some d
some di
and son
3.2 Pr
Survey
transpo
maps.
six p,
positic
is a
mountin
utilize
residua
adjuste
non-per
follow
descrip
describ
display
collect
blunder
data ca
remove
