methods, using RRTS-1 data. Such a provisional map could then
provide the framework for a ground checking program during the
field season to follow. Meeting the objective would require that
the boundaries between the terrain units could be accurately defined
and, to a lesser extent, that the surface materials could be correct
ly identified.
Selected Method of Analysis
The first group, concerning techniques of visual inter
pretation, was examined using black and white imagery on all four
bands, pseudo-colour composites, and several methods of image
enhancement using variations in assigned-colour densities. This
group was discarded., not because it did not offer possibilities for
terrain analysis, but because it could not generate sufficient
information to construct a suitable terrain map.
The work, therefore, has been concerned with automated
techniques of multispectral analysis using the Bendix Multispectral
Analyzer Display (MAD) unit and PDP 10 computers at the Canada
Centre for Remote Sensing.
METHODS
The criteria used to select the study area were those
required by the technique to ensure reasonably valid data, that is:
1) low to moderate vegetation cover, north of the treeline; 2)
snow-free terrain; 3) high quality data in at least two bands, in
this case, bands five and six. The area around Pelly Bay, N-.W.T.
(figure 1) was selected as it most nearly fitted the foregoing
requirements and from that frame, a small area on the Arrowsmith
River (figure 2) was used for the study.
Method of Analysis
The statistical methods used are those described by
Shlien and Goodenough (1974) for automated terrain classification.
Two assumptions were made concerning the validity of the data: 1)
that any atmospheric distortion of the spectral signatures would
be constant, and; 2) that because of this, the spectral classes
assigned to the terrain units would be internally consistent, that
is, consistent within tne area bounded by the ERT3 frame.
The technique used was to display a selected portion of
the ERTS frame on the Bendix and to train on areas of terrain,
identified, when necessary, by reference to conventional high
altitude air photography, with a software cursor. It was found
that using the cursor on full screen classification permitted a
generalisation of the probably impure terrain units ( that is,
representing a combination of several spectral signatures) which
was more suited to pattern recognition than the more precise and
purer breakdown of units provided by a computer printout of
pixels.
By this method, six units were identified, and another
added for unclassified water areas. The units were: clear water
with little or no silt content; water with high silt conten ;