no more 
is then 
data is 
Found in 
; had to 
variably 
ich both 
seed is 
roughout 
‚ growth 
another. 
| enough 
id. The 
maps of 
re any 
tors to 
(ions of 
used to 
rpically 
ange of 
> should 
pear to 
rom the 
logy in 
at this 
fied in 
lo avoid 
raining 
set is 
ng, the 
Ve The 
sult is 
xels or 
, other 
cation. 
sed for 
ed from 
mission 
ultural 
must be 
gnored. 
For the 
ould be 
raining 
nd, the 
f such 
an also 
  
  
manually remove the classified pixels from the non-agricultural areas with the 
cursor. Manual removal should be done with care: pixels should either be 
totally removed, or left in both segments and the land represented by the 
segments. Pura 
Once the classification has been accepted, it is filtered with a 3x3 
spatial filter to remove single pixels classified as Canola/rapeseed and fill 
in single pixels missing in fields. The number of pixels of crop in each 
segment and in the whole subscene are recorded, and the next subscenes can then 
be analyzed using either the signature already developed or a new one based on 
data in these subscenes. 
After classification, one must also draw the Crop District boundaries on 
the video display where necessary, using maps for reference. An easier 
procedure is to simply delete all of the crop of interest which falls outside 
of the Crop District of interest. A similar procedure was adopted for removing 
overlap between adjacent scenes of the system corrected data. 
SOLUTIONS TO ANALYSIS PROBLEMS 
This section outlines the responses to the more difficult problems 
encountered requiring more complex or perhaps more controversial solutions. 
Missing or Unacceptable Segment Data: Segment data may be missing for various 
reasons. Ground data may be rendered obsolete by weather. For example, in 
1981 a number of Canola/rapeseed fields were ploughed under after Statistics 
Canada enumeration because of poor germination. A segment may also be obscured 
by cloud, cloud shadow, haze, or smoke from forest fires. Where data were 
missing, segments were ignored. Where there was a question about the veracity 
of data given, it was assessed by the project staff. Depending upon the result 
of this assessment segments were deleted; changes were made to correct the 
segments; or they were further refined by local co-operators. 
  
No Training Data: Many of the images contained no segments or segments in 
these regions did not include rapeseed fields. 
If there were training fields within the same satellite path on the same 
day, these were used for signature extension. More serious problems occurred 
where there were no training fields for the imagery acquired on one date. The 
images acquired for National Topographic Survey Map sheets 83P4 and 83P2 on 
August 10 did not contain reliable training data but they did include rapeseed. 
To solve this problem imagery was interpreted visually to identify training 
sites using information on the crop phenology. If imagery containing training 
fields had been available for either the day before or the day after, one may 
have used that signature as a point of departure for the visual analysis. This 
approach has many potential pitfalls and should be used only with greatest 
care. 
Isolated Haze or Thin Smoke: A variety of methods are available to correct for 
thin haze and other factors which affect the signature over an image. Most of 
these require time consuming recalculations of spectral values, the presence of 
clear lakes as standard targets (Ahern et al., 1977), or other relatively 
complex manipulations. Since the primary requirement in this study was to 
obtain a true representation of crop area which implies an accurate 
classification, the method developed here is based upon subjective assessment 
of the effect of haze on the displayed image. Areas of haze (often a jet 
vapour trail) appear as somewhat brighter, frequently elongated features. 
Usually one does not see haze, but rather its effects on the ground beneath it. 
The same effect is also seen at the edges of clouds. Where clouds are 
associated with haze, the haze is treated with the clouds, as in the next 
subsection below. 
391 
Tr ; 
m. a a ER Cy MPA A UC SN SE SR SU I A DE Re 
A A em tll