Research Methodology - Historical Land Use/Land Cover Classification Using Remote Sensing - page 65

contrast (''more of the available range of digital values is used, and the contrast
between targets and their backgrounds is increased'' (Jensen 2005 )); (2) spatial
enhancement (spatial filtering, edge enhancement, and Fourier analysis etc.); and
(3) spectral transformation (generating more valuable data or products—e.g.,
NDVI—based on manipulation—e.g., division of several spectral bands of data).
Despite the LANDSAT-images being level 1G corrected (Level-1G was cor-
rected from USGS, and this modification consists of the basic corrections of
radiometric and geometric distortions. But these corrections are not suitable for
each application and thus user have to make additional corrections if the from
USGS corrections are not sufficient), they are not good enough accurately regis-
tered in form pixel-to-pixel. USGS had pointed to a possible error of up to 250 m,
and had not atmospherically corrected the data, thus all findings were subsequently
re-corrected geometrically for this work (see Sect. 5.2.1 ). Atmospheric effects on
the spectral signal were also minimized with a correction method (see Sect. 5.2.2 ).
In addition to radiometric normalization (see Sect. 5.2.3 ), the ASTER data were
delivered in Level 1A without any corrections.
ETM+-bands 6 and 8, plus TM-band 6, were eliminated from the entire pro-
cessing and classification. The panchromatic information of band 8 was only used
for pan sharpening. The sixth thermal spectral band—with its thermal informa-
tion—was not used because the two reasons: it has a coarse spatial resolution; and
it can recording only the transmitted radiation from objects in contrast to other
spectral bands that measure the reflected radiation.
All image processing, classification and preparing the final results were carried
out using two programs: ENVI, Version 4.6 and ArcGIS, Version 9.3.
5.2.1 Geometric Data Processing
''The more time steps involved for a change analysis, the more effort should be
spent on image registration and radiometric adjustment'' (Wulder and Franklin
2003 cited in Schultz 2011 ). So, the goodness or badness of method used in the
registration of remote sensing data (image/s) will determine the quality/accuracy
of the resulting change detection product (Schultz 2011 ). Townshend et al. ( 1992 )
assumes that the ''problems created by misregistration are likely to be greater in
the sensing of land surfaces compared with the atmosphere or many ocean
properties''.
There are several common expressions used to explain geometric correction
process (registration, rectification, geo-coding and ortho-rectification) (Scho-
wengerdt 2007 ). This process corrects the two different errors types (systemic and
nonsystematic) resulting from the two different sources (within the remote sensing
system itself, and during the recording of images) (Lo and Yeung 2002 ). The
various applications of geometric correction on remotely sensed data are: co-
registration of images that cover the same area on the Earth but they were obtained
from two or more different sensors, or they were obtained at two or more different