Cohen, Yafit
Wheat & Herbaceous
Among mixed clusters two were composed mainly of wheat and herbaceous with similar phenology. The potential ability
to distinguish between them was checked through knowledge concerns wheat/herbaceous phenology variants and their
dependence on environmental conditions. Following that, existence of these relations in imagery data was inspected by
overlay analysis utilizing empirical samples of wheat and herbaceous from the mixed clusters. The analysis had shown
that most wheat samples were found under semi-arid conditions and most herbaceous samples were found under more
humid areas. Rainfall data was used then as a mask and combined with NDVI values constrains the mixed cluster was
splitted and the two revealed sub-clusters were then merged into wheat and herbaceous “super” classes respectively.
Orchards and Natural tree formations
Major classification confusion was found between orchards and natural tree formations, both characterized by relatively
stable photosynthetic activity with minor seasonal fluctuations. Comprehensive examination of “mixed clusters” had
shown that phenological attributes are not sufficient for distinguishing between the two classes. Threshold enhanced PCA
values for orchards and natural tree formations were determined based on empirical data and spectral-canopy connections.
Initial examination of enhanced PC values combination had shown that orchards mislabeled as natural tree formations
clusters have significantly higher values and natural tree formations mislabeled as orchards have significantly lower
values. These results had proven a consistency in the behavior of the two land cover types, which was translated to
several classification rules.
In addition, soil types map was used as splitting criteria between orchards and NTF on a limited area. Mixed clusters of
TF were efficiently splitted by soil types. Furthermore, adjacent plots of orchards on different soils, extracted relatively
distinctive phenologies. This fact shows that NDVI values are sensitive to dofferences caused by spatial variations.
3 RESULTS
In general, the integrated approach applied through knowledge-based classification rules improved recognition accuracy
of crop types.
wheat | vetch chick pea clover | cotton | sunflower Other orchards Natural other
pea crops veg
wheat 82 9 14 29 21 1 4
vetch 5 73 14 4
chickpea 9 100 7
pea 3 9 79 3
clover 43
cotton 95 5 10 2
sunflower 2 95 14 1
other crops 2 7 38 1 10
orchards 2 7 71 10
Natural veg 6 7 7 19 62
other 2 2 6 10 100
Plots # 127 17 | 10 14 14 55 22 29 126 29 10 | 75.5%]
Table 1. Confusion Matrix of unsupervised classification
chick- Other Natural
wheat | vetch pea pea clover | cotton | sunflower crops orchards veg other
wheat 94 9 7 21 1 17
vetch 1 91 22
chickpea 100 7
pea 2 93 4
clover 64
cotton 96 14 1
sunflower 2 100 14 1
other crops 1 7 38 1 10
orchards 2 83 10
Natural veg 2 7 3 7 53
other 0 6 10 100
Plots # 127 | «1 10 14 14 55 22 29 126 29 10 | 83.796]
Table 2. Confusion Matrix of classification rules
284 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.
