Keywords: AVHRR, Atmospheric Correction, 6S code, Look-up Tables

Abstract: Based on 6S code, an atmospheric correction algorithm for AVHRR data in an operational mode have been developed. The algorithm works with look-up tables (LUTS), which cover a broad possible range of atmospheric conditions, Sun-Target-Sensor geometry and elevation of target. The correction algorithm was validated by using the ground truth data of Inner Mongolia grasslands and demonstrated its effectiveness.

1. Introduction
The Normalized Difference Vegetation Index (NDVI) calculated from the National Oceanographic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR) has been commonly used around the world for monitoring vegetation conditions and mapping land cover over large areas (Ehrlich et al. 1994). AVHRR measures of surface radiance, however, usually distributed by atmospheric effects depending upon atmospheric contents and depth, Sun-Target-Sensor Geometry (STSG) (Tanre et al. 1992).

Most studies of vegetation monitoring using AVHRR data have focused on the use of composite images to overcome some of problems associated with cloud attenuation, atmospheric effects caused by day-by-day variations in atmospheric and STSG conditions. These composite images are produced using the maximum value of composting technique based on NDVI from a series of images in a 7-day or 10-day period (Holben 1986). However, the problems associated with atmospheric effects on composite images are not resolved quantitatively. It is necessary to prompt the research and development of atmospheric quantitative correction. For this, an atmospheric Radiative transfer code can be considered to use (Markham et al., 1992, Moran et al., 1992). Although there is presently no agreement as to which atmospheric Radiative transfer code should be used (Vermote et al., 1996), it should be noted that for application to regional and global data sets atmospheric corrections must be fast and relatively straightforward. In this article, we propose an atmospheric correction algorithm adaptive the global AVHRR imagery data by using 6S code (Vermote, et al., 1997) which is one of developed atmospheric Radiative transfer codes.

2.Atmospheric Correction Algorithm for AVHRR Data by 6S Code

2.1 Radiative transfer equation in 6S code
6S code is a semianalytical but reasonably accurate atmospheric code that lends itself well to surface reflectance retrieval. It was modified to facilitate reverse mode computations and include elevation of target dependence. In the 6S formulation, one can write the following expression for surface reflectance retrieval:
r_i^t=Y_i/(1+Y_iS_i)         (1)
Where
Y_i=A_id_S²L_i*-B_i
A_i=p/[E_oiT_gi(q_v,q_s,z)T_si(q_s,z)T_vi(q_v,z)m_s]
B_i=r_i^a(q_v,q_s,Df,z)/[T_si(q_s,z)Tvi(q_v,z)]
Where rⁱ = surface reflectance, S=atmospheric spherical albedo, T_g= gas transmittance, T_s =scattering transmittance in solar direction, T_v = scattering transmittance in sensor direction, r^a=atmospheric reflectance, E₀=exo-atmospheric solar irradiance, L*=radiance arriving at sensor, m_s==cos(q_s), d_s= solar distance in astronomical units(a,u), q_s= solar zenith angle, q_v=sensor zenith angle, Df =relative azimuth angle between solar azimuth and sensor azimuth, z= elevation of target, and the subscript i refers to channel number (1or 2).

The values of A_i(q_v,q_s,z), B_i(q_v,q_s ,Df,z) and S_i(Z) for each pixel can be obtained by running 6S code, but it takes a long time to calculated them. In order to reduce the calculation time, a look-up tables (LUTs) with parameters of q_c,q_s ,Df,z and visibility (V) of aerosol concentration are introduced to our processing system, where q_c is sensor scale angle directly referring to AVHRR image. A_i, B_i and S_i for a pixel can be obtained by linear interpolating of LUTs by the specified parameters.

2.2 LUTs for atmospheric correction
LUTs are generated for 15 combinations of five standard atmospheric models, i.e., tropical (TROP), Midlatitude summer (MLS), subarctic summer (SUS), Midaltitude winter (MLW) and Subarctic winter (SAW), and three NOAA satellites, i.e., NOAA-9, -11 and -14. the standard atmospheric models are from a standard climatology with latitude and seasonal dependence introduced by McClatcheyl et al., (1972). The aerosol model was assumed as the continental type. In LUTs, the parameters and their quantification spacing i are shown in Table 1. The parameters cover a range of possible q_c,q_s ,Df , z appearing in the whole global and seasons. The quantification spacing were determined by simulation so as for the errors in the surface reflectance retrieval on the basis of linear interpolation of the LUTs to be within ± 5% of their, real values.

Figure 1 shows one of those simulating results of A_i, B_i and S_i variations as a function of q_c,q_s ,Df, z and V for TROP, MLS and SUS atmospheric models.

Aerosol concentration V:	10,15,23,40,60,80(km)
Solar zenith angle qs:	0,10,20,30,40,45,050,55,60,65,(deg.)
Sensor scan angle qc:	0,10,20,30,35,40,45,50,(deg.)
Relative azimuth angle Df:	0,10,30,40,50,60,70,80,90,100,110,120,130,140,150,160, 170, 175, (deg.)
Elevation of target z:	0,0.5,1,2,3,6,(km)

Table 1: the quanitazation spacing of variables for Ai, Bi and Si in look-up tables