Correction of non-uniform aerosol effect of Landsat TM image by using blockwise approached atmospheric correction model (BACM)
Chien-hui Liu
Department of Management Information System, Transworld Junior college of Commerce,
Tel: (886)-5-5570-866 Fax: (996)-5-5570-869
E-mail: chliu@ws231.tjcc.edu.tw
E.F. Vermote
Department of Geography, University of Maryland, College park, USA
Tel: (1)-301-286-6232 Fax: (1)-301286-1775
Abstract.
An blockwise approached atmospheric correction model (BACM) for correting non-uniform aerosol effect is developed in this paper. By using the dense dark vegetation, non-uniform aerosol optical depth can be retrieved in every specified block of image. Three Landsat TM images in LTER sites including two Hog Island images of uniform aerosol effect and one Madison of non-uniform aerosol effect are used to testify BACM. The results show that the mean errors of retrieved aerosol optical depths in these three images are 0.14 and 0.05 in TM1 and TM3 bands by comparing the field measurements from AERONET. Higher aerosol optical depth (
t (TMI) =0.84) in the hazier region is derived compared with lower value (
t(TM1)=0.62) in the clearer region of Madison scene. Non-uniform aerosol effect is then well corrected. Although in lack of field measurement, typical bare soil and vegetation reflectance can be retrieved even I the very hazy region (visibility~5km) of the scene after atmospheric correction.
Introduction
Quantitative analysis of remotely sensed data relies on the accurate correction of atmospheric effect, since modulation of the atmosphere impedes many applications of multi-temporal satellite images, such as agriculture monitoring (hill and Sturm 1991), land-cover change, remote sensing of surface albedo and classification (Liu and Chen 1995). Therefore, it is indispensable to convert the digital count of satellite image to surface reflectance which is reasonably independent of atmosphere and better related to the surface characteristics (Moran et al. 1992).
Dense dark vegetation (DDV) algorithm for land developed by Kaufman and Sendra (1988) is now the most widely used method (Richter 1996, Liu et al. 1996) to retrieve aerosol optical depth, because of the stable and low reflectance in the visible bands. This algorithm Is not very satisfactory, since the normalized difference vegetation index (NDVI) used to select DDV is affected by the aerosoland thus tends to select pixels with low aerosol concentration given sililar surface reflectance (Holben et al. 1992, Kaufiman and Tanre 1996). Based on the less sensitivity to aerosol scattering in longer wavelength (e.g. mid-IR 2.2 mm or 3.7 mm) and still sensitive to surface characteristics, global remote sensing of aerosol and subsequent atmospheric correction procedure for AVHRR and future ESO-MODIS has been developed (for details, se Kaufman and Tanre 1996, Kaufman et al. 1997). Alternative way to correct the non-uniform aerosol effect by DDV algorithm from NDVI is to partition the remote sensing image into blocks (or sectors, Richter 1996), which are assumed with uniform aerosol.
In this paper, a blockwise approach Atmospheric Correction Model (BACM) is developed to correct the non-uniform aerosol effect of Landsat TM images. Evaluation of retrieved aerosol optical depths is made possible by comparison with the field measurements from AERONET (Holben et al. 1997).
Atmospheric correction model for non-uniform effect
To correct the non-uniform aerosol effect, Blockwise approached Atmospheric Correction Model (BACM) is developed. Those include geometry (sun and sensor), date, image (e.g., blocksize to correct non-uniform aerosol effect, window size to correct adjacency effect), sensor (absolute radiometric calibration coefficients to convet the DC to radiance), molecule (scattering and absorption optical depths), aerosol (refractive index, Junge parameter and size range) and DDV assumption (f1,f2, f3 and
rddv in blue and red for landsat TM, or green and red for SPOT HRV). At first, the total image is divided to Nx*Ny blocks where uniform aerosol effect is assumed in every block. The fraction f1* and f2* in every block are then equal to f1/(Nx*Ny) and f2(Nx*Ny) respectively. DC
ddv for retrieving he aerosol optical depths in TM1 and TM3 bands in very block are then determined by using the three parameters f1, f2 and f3 (e.g. 0.2, 0.5 and 0.01). Look-up table (LUT) between digital count and aerosol optical depth at assumed reflectances
rddv (0.01 and 0.02 in blue and red bands, respectively) Kaufman and Tanre 1996) is used to derive aerosol optical depth by determined DC
ddv for every lock. Angstrom formula (Angstrom10064) is used to interpolate the aerosol optical depths in the other TM bands. Surface reflectance image with uniform target assumption can be obtained through LUT between digital count and surface reflectance at derived aerosol optical depths. In practice, spherical albedo
rs, total upward transmittance T(
mv), upward diffuse transmittance td(
mv)and upward beam transmittance exp(
-t/mv) are computed and stored for every block to correct the adjacency effect (equation 15, Liu et al. 1996). Window size is specified about 1 km, which is approximately the same as Richter (1990)
