CHANGE DETECTION USING RELATIVE ATMOSPHERIC CORRECTION OF SATELLITE IMAGES AT
DIFFERENT TIMES
Takashi Kusaka, Takao Kakehi and Masanori Ootuka
Kanazawa Institute of Technology
Nonoichi-machi Ishikawa 921, JAPAN
E-mail: kusaka@manage.kanazawa-it.ac.jp
Commission VII, Working Group 2
KEY WORDS: CHANGE DETECTION, LAND USE, ATMOSPHERIC CORRECTION, LANDSAT
ABSTRACT:
This study describes a method for detecting land cover change that uses the relative atmospheric correction of multi-
temporal satellite images. The radiance change arising from the difference in atmospheric conditions and sun angles between
satellite images taken at different times, can be estimated in a relative way, by taking one image as reference.First, it is
derived from the radiative transfer model that the relation between radiance values of remotely sensed data acquired at two
different dates has a linear form. Next, it is shown that the surface reflectance ratio, A(t2)/A(t1), at two different times, tl
and t2, is evaluated by pixel values, X(t2) and X(t1), the coefficients of the estimated linear equation and the additive path
radiance in the image taken as reference. The distance d of A(t2)/A(t1) from the value of 1 was introduced to detect the area
of change. The threshold values of d were automatically determined by using the 2-dimensional frequency distribution of
distance values in two different spectral bands. Finally, the present method was applied to Landsat TM images taken at Oct.
1987 and Nov. 1991. As a result, it is shown that the areas of land cover change are successfully detected by using the
threshold of the distance selected in this study.
1. INTRODUCTION radiance in the image taken as reference (Kusaka, 1994).
The additive path radiance is estimated as the minimum
Land cover change detection is one of the major count level at the reference image. Next, the change
applications of remotely sensed data and is useful in the detection using the surface reflectance ratio at two different
analysis of environmental change such as land use change, ^ times is performed. To do that, the distance d of A(t2)/A(H)
assessment of deforestation and damage assessment. from the value of 1 was introduced. The change detection
Several methods for detecting land cover change have been method in this study depends on the selection of the
proposed. They include multitemporal classification, image ^ threshold value of d and therefore the threshold of d was
differencing/rationing, vegetation index differencing, ^ automatically determined by using the 2-dimensional
principal component analysis and change vector analysis. ^ frequency distribution of distance values in two different
Various methods for change detection are in detail reviewed spectral bands. Finally, the present method was applied to
by A.Singh (Singh, 1989). Landsat TM images taken on Oct. 10, 1987 and Nov. 6,
1991. Two images are taken at the same season, and at
In this study, another approach is described that uses the —— time intervals of 4 years. This area covers the suburban
relative atmospheric correction of multi-temporal satellite ^ area near Kanazawa city in Japan. By taking the image of
images. Relative atmospheric effects between satellite = Oct. 1987 as reference, the relative atmospheric correction
images taken at different times, such as the radiance change in the paired images was performed, and the areas of land
arising from the difference in atmospheric conditions and cover change in the study site were detected by using the
sun angles, can be estimated in a relative way, by taking ^ relative atmospheric correction images.
one image as reference (Caselles, 1989).
First, it is derived from the radiative transfer model that the 2. RELATIVE ATMOSPHERIC CORRECTION
relation between radiance values of remotely sensed data
acquired at two different dates has a linear form. The Consider a radiative transfer in a plane-parallel atmosphere
coeffients of this linear equation only include atmospheric — bounded by a flat reflecting layer with non-uniform
parameters at different times, and are estimated by using ~~ Lambertian reflectance. The upward normal radiance at the
radiance values of ground objects which are assumed to top of the atmosphere, L,, is approximately expressed in à
have almost constant reflectances with time.The spectral ^ quadratic form with respect to the surface reflectance
surface reflectance ratio, A(t2)/A(t1), at two different times, —— (Kawata, et al., 1988, Kawata et al., 1990). L,is given by
t1 and t2, is obtained from count levels of the corresponding L, = (T + SA)A; + (Q - T)A + (P - S)A? +R (1)
pixel in two different images, X(t2) and X(tl) the where the coefficients P, Q, R, S and T are expressed by the
coefficients of the linear equation and the additive path transmission and reflection function in the free atmosphere,
384
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996