361 
to Cropland (A). This accounted for about 
85% of the changes in the Peace River zone 
and 75% in the Central zone. Changes from 
Grazing to Cropland dominated the activity 
in the Foothills zone, accounting for 63% 
of the area of change. 
Classification Accuracy 
The classification accuracy was assessed 
at nine sites located throughout the 3 
zones of the project area. A total of 
991 land use polygons interpreted from 
Thematic Mapper imagery were compared with 
black and white aerial photography taken 
in the summer of 1986. A error matrix 
showing the degree of agreement between 
the two methods was then tabulated (Table 
4) . Individual class accuracies were over 
90%, except for Rough Grazing (K) and 
Clearcut/Unknown (U) class. 
The most serious interpretation errors 
were found in 3 areas: confusing cropland 
(A) and Grazing (K) and errors of 
omission. Confusion between Grazing (K) 
and Agriculture (A) was, in part, 
attributable to the similarity of improved 
pasture that may not have been seeded for 
a number of years and grazing lands which 
contain wild grasses of similar spectral 
appearance. The interpreter was often 
left to rely on secondary information such 
as the field shape to determine if a 
change had occurred. The confusion 
between K and A was most pronounced in the 
Foothills zone where K is the dominant 
cover type. 
Errors of omission occur when polygons of 
a given class are not recognized and 
mapped. This occurred with 8.2% of the 
test polygons, mostly in the cropland 
category. While some errors of omission 
could be attributed to human error, most 
were confined to small fields near the 
minimum size detectable using Thematic 
Mapper imagery. Therefore is it not 
surprising that the lowest omission error 
rate (3.1%) were found in the Peace River 
zone, a relatively uniform landscape 
dominated by easily visible large square 
shaped fields of cropland (Table 5). 
Errors of commission were generally low 
indicating that changes that were 
identified from Thematic Mapper imagery 
were usually genuine changes, even though 
they were not always assigned to the 
correct class. 
Throughout most of the study area, high 
accuracies were achieved where forested 
land was cleared for cropland. The use of 
winter imagery contributed to this 
accuracy since it depicted a sharp 
spectral contrast between land cleared for 
agriculture and adjacent uncleared forest. 
An additional factor leading to high 
accuracies was that there were often few 
different land use choices. This helped 
reduce or eliminate confusion among 
alternative land uses, and contributed to 
higher accuracies levels observed in some 
zones. 
Air 
Photo 
Thematic 
Mapper 
Image Interpretation 
Interp 
A 
В 
К 
T 
U 
X 
N/C 
A 
611 
1 
36 
1 
68 
В 
5 
26 
2 
2 
К 
21 
109 
1 
4 
T 
2 
49 
U 
2 
2 
2 
1 
X 
1 
13 
1 
N/C 
26 
1 
2 
2 
TOTALS 
666 
28 
153 
52 
3 
13 
76 
991 
ACC'CY 
91.7 
92.9 
71.2 
94.2 
66.7 
100.0 
Weighted Classification Accuracy: 86.7% 
Table 4. Error Matrix 
ZONE 
(sample size) 
CLASSIFICATION 
ACCURACY 
OMISSION 
ERROR 
COMMISSION 
ERRORS 
Peace River (294) 
83.2 
3.1 
2.0 
Central (556) 
93.0 
6.4 
3.1 
Foothills (141) 
88.3 
22.0 
5.7 
Table 5. Summary of Classification Accuracy by Zone.