3.RESULTS 
4.DISCUSSION 
Analysis of the confusion matricies shows that the 
general level of spectral confusion is highest for 
the February scenes decreasing through April and May 
to a lowest occurence in the multitemporal scenes. 
Table 3.1 shows the weighted diagonals from the 
confusion matricies from all the classifications 
expressed as percentage of the diagonal and non 
diagonal entries. 
Table 3.1 The percentage of correctly classified 
pixels for individual cover types from each 
classification 
OSR WW WB SB G SBe Pe Wb WC WD 
Feb 
10.8 
44, 
.1 
22, 
.7 
28, 
.0 
8, 
.9 
34, 
.5 
17, 
.3 
0. 
1 
46, 
.0 
27, 
.1 
Apr 
70.9 
73, 
.6 
60, 
.1 
48, 
.1 
16, 
.1 
52, 
.1 
23, 
.6 
1. 
7 
86, 
.3 
82. 
.5 
May 
100 
60, 
.9 
41, 
.2 
85, 
.8 
56, 
.6 
69, 
.6 
27, 
.1 
0. 
9 
99, 
.7 
93. 
.3 
MTP 
100 
82, 
.9 
69, 
.7 
87, 
.4 
51. 
.3 
65, 
.0 
37. 
.4 
4. 
7 
99, 
.7 
99. 
,7 
MTP 
= Api 
rii 
May mu 
ltitemporal 
combinati 
Lon 
This shows that for most informational classes the 
multitemporal classification gives the highest 
classification accuracy, with a mean % value of 
69.78%. A 6.27% improvement over the best single 
date, May, and a 45.83% improvement over the worst, 
February. Two exceptions can be identified; 
grassland, where the May classification gives a 5.3% 
increase in classification accuracy over the 
multitemporal, and sugar beet, where again the May 
classification gives a slight increase of 4.6% purity 
over the multitemporal.A clear indication of the 
improvements in crop classification purity achieved 
through the vise of multitemporal imagery can be 
gained through examination of the mean class purity 
for the cereal crops. The May scene gives an average 
percent purity of 62.63% compared with a value of 80% 
from the multitemporal data. 
Table 3.2 Shows the pixel count, percentage area and 
weighted percentage area for the multitemporal 
classification compared with the percentage areas for 
the whole county and for the parishes which 
correspond to the training areas. Sugar Beet, Winter 
Beans and Peas have been combined as a single bare 
soil class. This is a logical grouping as until mid 
June all three crops are in a pre emergent state. The 
two Woodland classes are also expressed as one 
woodland class as no distinction between the two is 
made in the June census data. 
Table 3.2 Crop area estimation 
Cover type 
Pixel 
Percentage area 
count 
ML 
MLW CO PA 
Oilseed rape 
2295 
0.88 
0.88 
1.70 
4.26 
Winter wheat 
48978 
18.68 
17.55 
30.93 
40.60 
Winter barley 
23286 
8.88 
6.93 
14.21 
11.14 
Spring barley 
59588 
22.73 
20.91 
12.44 
11.14 
Grassland 
12810 
4.89 
3.18 
3.75 
3.06 
Bare soil 
30439 
11.62 
11.62 
9.56 
7.91 
Woodland 
3688 
1.41 
1.00 
2.52 
1.25 
Unclassified 
18060 
30.91 
37.93 
24.89 
20.96 
ML = Maximum likelihood MLW = Weighted maximum 
likelihood 
CO = County census data PA = Parish census data 
The results show rather poor area estimation from the 
Landsat data. Spring barley is over estimated by more 
than 8%, winter barley under estimated by 7.28% and 
winter wheat under estimated by over 13%. 
* 
The classification from the February data shows 
confusion between all classes. Mean classification 
accuracy is only 23.95%. This is indicative of the 
strong similarities in the crop phenologies at this 
time. All the winter cereal crops, and the oilseed 
rape have emerged, though crop cover is low and the 
contrast in crop appearence, seen later, is not yet 
realised in the poorly developed young plants. 
Similar confusion is apparent between the spring 
cereals and crops such as sugar beet, peas and beans. 
These late sown crops will not have emerged in 
February and can be considered as a bare soil class. 
By April both growth and development have occured in 
the cover classes and the spectral confusion is now 
falling into three main groups; The cereals, grass 
and oilseed rape; the sugar beet, peas and beans; and 
the woodland. A mean classification accuracy of 51.5% 
is indicative of the reduction in spectral confusion. 
The growth of the winter cereal crops and the oilseed 
rape has been sufficient to remove confusion between 
these classes and the bare soil; sugar beet, peas and 
beans. The spring barley class is still confused with 
these bare soil classes. This is due to the lower 
crop cover as a result of later planting. The 
classification accuracy of 48.1% for the spring 
barley is indicative of this confusion. The 
development of the winter crops is relatively limited 
and strong phenological contrast is still not found. 
The difference in the planting dates of the winter 
wheat and the winter barley give rise to some 
spectral separation. The barley being planted a 
little before the wheat means that crop growth and 
development are further advanced. However both 
classes can be confused with the well established 
oilseed rape crop. This limits classification 
accuracy to 60.1% for the winter barley and 73.6% for 
the winter wheat. The coniferous woodland now has a 
unique spectral response (see the coincident 
spectral plots, Figure 2.1) and whilst still showing 
a little confusion with the deciduous woodland class 
is classified with 86.3% accuracy. The deciduous 
woodland classification is less well defined with a 
classification accuracy of 82.5%. But now that the 
trees are in leaf there is strong separation from 
most other classes. 
By May changes in crop development result in good 
spectral separation of the cover classes. A mean 
classification accuracy of 63.51% illustrates this. 
The flowering oilseed rape has a unique spectral 
response and shows no confusion with any other class, 
precentage classification accuracy being 100%, though 
there are some unclassified pixels. These were all 
boundary pixels. Work on oilseed rape classification 
in Scotland by Wright, 1985 reached the same 
conclusion where no classification confusion was 
reported but failiure to classify boundary pixels was 
encountered. The spring barley was identified with 
increased accuracy. Sufficient plant growth and 
development has now taken place to give good spectral 
separation from the sugar beet, peas and beans which 
are still bare soil. However, the changes in crop 
development now result in a small amount of confusion 
with the winter cereals, especially the winter 
barley. This gives the classification accuracy of 
85.8%. The spectral differences between the two 
winter cereals apparent in the April classification 
and attributed to differences in planting date have 
now dissapeared. Considerable confusion between the 
two crops exists. Winter wheat is now classified with 
only 60.9% purity, and the winter barley having the 
additional confusion with the spring barley has a 
classification purity of only 41.2%. The two woodland 
classes are now almost classified with 100% accuracy.