Research Methodology - Historical Land Use/Land Cover Classification Using Remote Sensing - page 108

160

140

120

100

80

60

40

20

0

Band 1 Band 2 Band 3

Fresh Water Saline Water Artificial Surfaces Bare Areas Fallow

Nat. Woody Nat. Herbaceous Trees- Rainfed Trees- Irrigated Wheat -Rainfed

Wheat -Irrigated Barley-Rainfed Barley- Irrigated Pastoral -Irrigated

Fig. 5.30 Spectral class signatures (band means) related to ASTER data (3 spectral bands, 15 m)

200

100

0

Band 1 Band 2 Band 3 Band 4 Band 5 Band 6

Saline Water Artificial Surfaces Nat. Woody Nat. Herbaceous

Fresh Water Bare Areas Fallow

Trees-Irrigated Wheat-Rainfed Barley-Irrigated Pastoral-Irrigated

Trees-Rainfed Wheat-Irrigated Barley-Rainfed

Fig. 5.31 Spectral class signatures (band means) related to fused ASTER data with LANDSAT-
ETM+ data (6 spectral bands, 15 m)

including: the interaction among classes of land uses and the natural coverage
distribution; and the lack of concrete borders to separate them. There were two
factors affecting and complicating this: The geographical location of the study
area; and nature and type of classes of LULC, which was affected generally by the
geographical location.
Another main reason is that the selection of the training sites is not completely
an objective process, affected by the person who selects and trains the sites. When
a researcher selects the training samples, they do so because they consider them fit,
appropriate and representative to the LULC in the study area. The training process
may not include all areas and classes in a study area (especially within the same
class); as, there are several kinds of wheat (hard and soft), some are rain-fed and
some irrigated, some are located on dark humid soils, while others are on light and
less moisture-rich soils, some have organic and chemical fertilizers added, while
others grow in different quantities; some wheat-fields may be peppered with
natural herbs and plants that grow within the wheat plants, while other fields have
homogeneous growth of only wheat plants; and finally, some wheat-fields may be
infected with disease. Where these differences are related to one class (i.e., wheat)
this will make spectral and spatial bias between wheat and other crops hard. Each
difference (or more collected differences) leads to diverse spectral appearances on
the image. So, the analyst has to gather training samples that satisfy the entire
different spectral responses of the crop especially if there are natural or agricultural
crops in the area with a similar spectral response.