Research Methodology - Historical Land Use/Land Cover Classification Using Remote Sensing - page 82

The masking operation enables researchers to use an image file to choose
(mask) definite areas and/or values from a matching raster file, and use those areas
and/or values to generate one or more new files. The input mask file and input file
must be the same as masking will be performed on the image area that both files
have in general through the intersection process. This operation was used too often
in the Multi Stage Classification Approach (see Sect. 5.7.1.2 ), especially in crops
classification (see Sect. 5.11 ). The masking areas were selected by generated class
values or were based on NDVI transformation (the masking operation was a
processing and not a pre-processing step). All the class values of classes to be
masked were set to zero or recoded to zero, then all unwanted zero signed features
will be ignored when masking was executed.
The sub-setting operation was used broadly in this study to cut and remove the
distorted margins of the LANDSAT-data; to subset only the study area (ERB
borders) from each image or from the whole data set mosaic scene; to reduce
processing time; and to reduce the geographical local extent that increased the
spectral differences of the existing ground surface features. The final subset of the
study area was about 50,335 km 2 .
Mosaics for the ASTER-May and August-Data in 2005 were produced, eight
paths from left to right (path-1: 4 rows, path-2: 4 rows, path-3: 3 rows, path-4: 3
rows, path-5: 3 rows, path-6: 4 rows, path-7: 5 rows, and path-8: 4 rows).
After the enhancement of the three bands of the ETM+-data (six bands for the
scene (p173r035), i.e., the bands (1, 2, 3, 4, 5 and 7) which covered a part of the
study area that the ASTER data did not cover), scenes were collected in one
mosaic-scene. Here, before mosaicing, subsets were completed for each scene to
remove margin deformations. After that, a geographic registration was applied for
the ASTER-mosaic-scene with the ETM+-mosaic-scene as master-scene, using the
image to image method. Before the last step, the three bands of ASTER data were
composited with the three bands of ETM+-data (one layer-stack). The last step
created a mosaic for the last scene which resulted from fusing ASTER-bands with
ETM+-bands, and for the p173r035-scene of ETM+-data that covered the rest of
the study area. The final result was the creation of one compound mosaic scene
from both the ASTER- and ETM+-data that was homogeneous: Radiometrically
(i.e., no or acceptable spectral appearance of the same features overall in the
mosaic-scene); spatially (15 m); and spectrally (six bands).
In order to reduce temporal and effort processing series on the remote sensing
scenes which covered more than the spatial distribution of the study area, these
scenes were subsetted to include only the spatial distribution of the ERB borders.
5.3 Design of the LULC-Classification System
A LULC-classification starts with defining a classification system. A successful
LULC-classification requires a suitable classification system and an adequate
number of training sites. Its design is related to: the needs of the user; the spatial