, 1994a. 
cale river 
orms, 19, 
plications 
lent open 
gs of the 
hin-reach 
In: River 
). 
elling in 
cial issue 
iutomated 
hives of 
272. 
A ROBUST ALGORITHM FOR CORRECTING THE TOPOGRAPHIC EFFECT OF SATELLITE IMAGE 
OVER MOUNTAINOUS TERRAIN 
C.H. Liu 
Researcher, Energy & Resources Laboratories, Industrial Technoloy Research Institute, Chutung Hsinchu, Taiwan, R.O.C. 
A.J. Chen and G.R. Liu 
Professor, Center for Space and Remote Sensing Research, National Central Unverisity, Chung-Li, Taiwan, R.O.C. 
Commission 7, Working Group 1 
Keywords: Topographic Effect, Digital Terrain Model, Atmospheric Correction Model, Reflectance, SPOT, Classification 
ABSTRACT 
In order to correct the topographic effect of satellite image over mountainous terrain, a robust algorithm is developed in the 
atmospheric correction model. Optimized aerosol optical depth is retrieved in the robust algorithm by minimizing the variation of 
surface reflectances of a selected canopy over different terrains including altitudes. slopes and incidence angles. Digital count 
image is converted to surface reflectance image by using the atmospheric correction model and digital terrain model. A SPOT 
image in rugged terrain is used to verify the algorithm. Hardwood is selected in the robust algorithm. For both of hardwood and 
acacia canopies, difference of reflectances located in shaded and well-illuminated areas are greately reduced, especially in near-IR 
band after atmospheric correction. Five classes such as forest, high-reflective land, bare soil. urban and grass are well classified 
with topographic-corrected image by clustering algorithm, whereas uncorrected image classifies terrain related classes such as forest 
under low and high illumination. The overall accuracy is 91.7%. 
accuracy is improved by the proposed algorithm. 
1. INTRODUCTION 
Many applications of remotely sensed data, such as land cover 
monitoring (Jones et al 1988), crown closure and timber 
volume estimation (Schieh 1992), and forest damage assessment 
(Ekstrand 1996), have been hampered due to the topographic 
effect, which is caused by the differential sensor response to 
diverse slope and aspect even for a given canopy (Holben and 
Justice 1980). It is particularly important to correct the 
topographic effect of satellite image in Taiwan, as two thirds of 
the land are mountainous areas (Chen and Chen 1991). 
Statistical transformations such as band ratio (Holben and 
Justice 1981), hyperspherical directional cosine (HSDC) (Pouch 
and Campagna 1990) and principal component (Conese et al. 
1993a) would lose some informations, although they had 
yielded good results (Conese er al. 1993b). Scene dependent 
regressed coefficients by combination of DTM in correction of 
topographic effect (Civco 1989, Colby 1991, Naugle and 
Lashlee 1992) are also the main obstacle in multi-temporal 
applications (Conese et al. 1993b). 
To correct the topographic effect in a deterministic way. an 
atmospheric correction model is necessary (Conese ef al. 
1993b). Usage of surface reflectance images will not only 
improve the classification accuracy but also detect the land 
cover changes in multi-temporal analysis (Kusaka et al. 1984). 
Digital terrain model (DTM) could be combined in the model to 
delineate the topographic effect of the image. Even for well- 
illuminated areas, the diffuse irradiance cannot be neglected 
even though the topographic effect is mainly caused by the 
beam irradiance (Proy et al. 1989). That is why some 
developed models (Pons and Sole-Sugranes 1994, Senoo ef al. 
1990) are not suitable for shadow and near shadow areas. 
Large error such as overcorrection will appear in these areas. if 
the diffuse irradiance is underestimated. The reflected 
irradiance by adjacent slope should also be considered in near- 
Kappa statistics is 0.87. It is concluded that classification 
IR band if the terrain surrounding is full of vegetation (Proy et 
al. 1989). 
Aerosol optical depth should be obtained before determining 
diffuse irradiance. Since the diffuse irradiance dominates the 
global incidence irradiance in shadow, more accurate aerosol 
optical depth is needed than it does iri the flat terrain. Image- 
based algorithm is desired in operation. As the NDVI of raw 
image exhibits good correlation with topographic effect, dense 
dark vegetation (DDV) algorithm developed by Kaufman and 
Sendra (1988) is not applicable to retrieve aerosol optical depth 
in rugged terrain (Liu 1995). By the same token, the image- 
based retrieval algorithm of aerosol characteristics recently 
developed by the authors (Liu et al. 1996) is also not 
appropriate for the rugged terrain. 
The objective of this paper is to modify the atmospheric 
correction model previous developed (Liu e: al. 1996) to correct 
the topographic effect of satellite image. DTM is combined 
into the model to account for altitude dependent transmittance, 
terrain related irradiances, such as beam, diffuse and adjacent 
slope reflected. To map the surface reflectance even in the 
shadow, optimized aerosol optical depth is retrieved by a 
proposed robust algorithm such that minimizing the variance of 
spectral reflectances of pixels at distinct slopes, altitudes and 
illuminations including shadow for a given canopy. A SPOT 
image is used to testify the atmospheric correction model. 
Spectral reflectances of canopies at shadow and well- 
illuminated area are compared before and after correction. 
Classification results are also used to verify the method. 
2. ATMOSPHERIC CORRECTION MODEL AND THE 
ROBUST ALGORITHM 
Suppose the surface is non-uniform and Lambertian. The 
atmosphere is assumed to be horizontally stratified. Satellite 
105 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996