Research Methodology - Historical Land Use/Land Cover Classification Using Remote Sensing - page 76

independent scale of radiance that serves as a more direct link between image and
biophysical phenomena, then addressing the errors in pixel values. It is then
possible to manipulate these values to maximize their information for studies that
are based on the digital processing of remote sensing data (Wulder and Franklin
2003 ; Liang 2004 ; Lillesand et al. 2008 ).
Schowengerdt ( 2007 ) has listed three levels of radiometric calibration. The first
converts the sensor DNs to at-sensor radiances. The second transforms the at-
sensor radiances to radiances at the Earth's surface. The third transforms it to
surface reflectance.
The radiometric correction/adjustment set includes the three mechanisms: (1)
calibration of the sensor: it is the process of converting the DNs to at-sensor
radiance for inter-sensor data comparison. Gains and offsets are well-known for
each remote sensing sensor, and these used to the recorded signals to generate the
DNs. This first mechanism is frequently calculated at the satellite ground stations;
(2) atmospheric correction (see Sect. 5.2.2 ); and (3) radiometric normalization
(absolute and relative). (A) Absolute radiometric normalization: ''for a linear
sensor, is performed by ratioing the digital numbers (DNs) from the sensor, with
the value of an accurately known, uniform radiance field at its entrance pupil''
(Liang 2004 ). In this case, user has to carry out atmospheric corrections, which
require atmospheric information at the time of the image acquisition (see
Sect. 5.2.2 ). However, when it is difficult to obtain these atmospheric parameters
and/or the absolute surface radiance is not necessary, one can change to (B)
relative radiometric normalization: it is an in-image technique which uses the
information contained within the image itself, and used when the full radiometric
calibration for remote sensing data is complex. The concept is based on the sup-
position that it is possible, by application of linear functions, to estimate the at-
sensor radiances recorded at two different times and for the same area but under
different conditions (Yang and Lo 2000 ). This technique has the disadvantages of
difficulty and time-consuming, where it has to determine the suitable time-
invariant features upon which the normalization is based (Teillet and Fedosejevs
1995 ; Schowengerdt 2007 ). This method is applied especially in applications
based on LULC-classification and post classification change detection (Song et al.
2001 ).
Several methods (Schott et al. 1988 ; Moran et al. 1992 ; Du et al. 2002 ) were
developed and proposed to be applied as techniques for the relative radiometric
normalization in remote sensing applications. Canty et al. ( 2004 ) proposed a
method based on MAD, which use the advantage of the invariance properties of
MADs. Canty and Nielsen ( 2008 ) further improved this approach by introducing
an iteratively re-weighting method of the MADs, which executed superior in
isolating no-change pixels fit to use for the relative radiometric normalization. The
MAD method, after the modifications by Canty et al. ( 2004 ) and Schroeder et al.
( 2006 ) provides better radiometric normalization than those achieved with manual
selected invariant features.