2. Method 
2.1. Data 
Raw data are weekly composed data issued from MVC process. These data were provided by the NOAA as 
global images of radiances expressed in numerical counts. Pixel dimensions are 15 x 15 km with a "plate-carree" 
projection. NDVI resulting from this procedure are the so-called Global Vegetation Index (GVI; Kidwell, 1990). 
The provided data set is composed of the red and near infra red channels (NOAA/ AVHRR 1 and 2), the thermal 
infra red channels (NOAA/AVHRR channels 4 and 5) and finally the solar zenith angle and the scan angle 
corresponding to the weekly selected NDVI value. In this present study we only considered the African continent 
(38° North to 35° south and 18° west to 55° East). On figure 1, is presented a raw NDVI temporal profile of a 
pixel selected at random in Burkina Faso (Kouka region, 12° north, 3.52° west). 
2.2. Data preprocessing 
Existing smoothing techniques can discard high frequency perturbations and are not able to eliminate effects like 
permanent bias due to atmospheric attenuations or long term trends as sensor degradation. However, 
uncertainties on data due to these factors can be very high. According to Goward et al. (1991) changes in gain 
and offset of NOAA-11 sensors between pre-flight and post-flight can introduce accuracy errors of 20 % for low 
to moderate values of NDVI. NDVI obtained from satellite measurements can be up to 30 % lower than 
equivalent NDVI at ground level if atmospheric attenuation is not taken into account. Because we want NDVI 
data to become useful for quantitative study we decided to realize data preprocessing to correct for calibration 
drift and atmospheric effects prior to apply the INTUITIV method. 
2.2.1. Reflectance conversion. 
Radiations recorded by satellite sensors are radiances. These values vary with solar zenith angle and earth to sun 
distance. To obtain data independent of the incoming solar radiation at the observed surface, radiances were 
converted into reflectances values. 
2.2.2. Data calibration 
Satellite sensors in earth orbit degrade with time. This effect causes a drift of NDVI over time not due to changes 
at ground level. Unfortunately, there is no on-board calibration for NOAA/AVHRR red and near infrared 
channels. However, different authors attempted to estimate these sensor degradation (Holben et al., 1990 ; 
Teillet et al., 1990). The technique is based on the monitoring of satellite measurements over ground target of 
which spectral behavior is supposed very uniform in time such as Sahara desert in Egypt and Libya. We used 
calibration coefficients estimated by Kaufman and Holben (1990) and Teillet and Holben (1992) to calibrate our 
data. NDVI profile resulting from this calibration step is presented on figure 1. 
2.2.3. Atmospheric correction 
Several authors developed models to attempt to correct radiometric data from atmospheric influences on red and 
near infrared radiations (Dedieu, 1990 ; Paltridge and Mitchell, 1990). We choose the Simplified Method for 
Atmospheric Correction (SMAC) method recently proposed by Rahman and Dedieu (1993) for our data 
correction. This model take into account Rayleigh scattering and also effects due to atmospheric ozone, water 
vapor and aerosols contents. It was especially designed to allow correction of huge amounts of data with 
reasonable CPU time consumption. To run the model, different input data were used. Input reflectances data 
were calibrated values. Measurement geometry was derived from solar zenith angle and scan angle data enclosed 
in the NOAA data set. Atmospheric water vapor contents were obtained by vertical integration of 11 
atmospheric levels of relative humidity data. This data set concerning the 1958 to 1973 period has been gathered 
by Oort (1983). Final file is expressed as a 5° longitude to 2.5° latitude grid of monthly mean values. 
Atmospheric ozone contents are issued from 1957 to 1967 measurements collected by London et al. (1976). 
Resulting data are constant mean annual values by 10° longitudinal strips. Finally, for aerosol optical thickness 
values at 550 nm, due to the weak availability of measurements, a constant value was used. It corresponds to an 
equivalent horizontal visibility of 23 kilometers (clear sky conditions). This value will often under estimate 
atmospheric aerosol content. However, due to aerosol effect correction formulation, error induced on reflectance 
computation is much lower for under estimation of aerosol content than for over estimation (Dedieu, personal 
communication). Profile resulting from this atmospheric correction step is presented on figure 1. 
2.3. INTUITIV method. 
Despite of NOAA/AVHRR data processing (MVC) and efforts accomplished on data calibration and 
atmospheric correction, temporal NDVI profiles still highly fluctuate. In fact, everyday about 50 % of the earth 
is covered by clouds. Over some regions, clouds sometimes stagnate for more than a week (e.g. rain forest, 
temperate region during winter). The remaining noise is also due to directional effects depending on scan 
geometry and optical anisotropy features of observed entities at ground surface. Atmospheric scattering 
properties are also highly anisotropic. The radiation trajectory through the atmosphere increases with increase of 
off-nadir scan angle. So, the higher the atmosphere impact on reflectance values is (Holben, 1986).