Improving Ndvi Time-series Using Imposed Threshold on Irt, Ir and Visible Values (INTUITIV) : A 
method for reducing cloud contamination and noise in NDVI time-series over tropical and sub-tropical 
regions. 
P. LOUDJANI*, F. CABOT, V. GOND, and N. VIOVY** 
Laboratoire d’Etudes et de Recherches en Télédétection Spatiale (LERTS), 
CNES/CNRS UMC00010, B.P.I. 2801, 18 Avenue Edouard Belin, 31055 Toulouse 
cedex, FRANCE. 
* Now at Institut National de la Recherche Agronomique (INRA), Station de 
Bioclimatologie, Domaine St Paul, B.P. 91, 84143 Montfavet cedex, FRANCE. 
** Now at Laboratoire de Modélisation du Climat et de l’Environnement (LMCE), 
Centre d’Etudes de Saclay, Bâtiment 709, l’Orme des Merisiers, 91191 Gif sur Yvette 
cedex, FRANCE. 
Abstract : The INTUITIV method is proposed as an alternative to eliminate cloud 
contaminated NOAA/AVHRR data and to filter temporal NDVI profiles. Threshold on 
visible, near infrared and thermal infrared NOAA/AVHRR data is the basic concept of 
the procedure. Preliminary to this method, preprocessings are applied on satellite data 
(calibration, atmospheric correction). Results of the entire process allow substantial 
reduction of noise characterizing raw NDVI time-series. Obtained temporal NDVI 
profiles show good agreement with present ground-based knowledge and provide a 
tool for new applications such as interannual comparisons of satellite-derived 
vegetation attributes or characterization of different stages of vegetation cycle. 
However, due to properties of the used threshold, INTUITIV method can only be 
applied over regions where temperature at ground level is never lower than 10° C 
(tropical and sub-tropical regions). 
1.Introduction 
The NOAA Advanced Very High Resolution Radiometer (AVHRR) provides daily earth observations. Spectral 
measurements in the red and near infrared wavelengths were demonstrated to be closely linked to plant-canopy 
attributes. These two channels were combined to produce the so-called and so-used Normalized Difference 
Vegetation Index (NDVI) (Rouse et al., 1974). Numerous studies showed potentials of NDVI for the estimation 
of Photosynthetically Active Radiation Intercepted (IPAR) by a canopy (Hipps et al., 1983 ; Asrar et al., 1984) 
or for above-ground phytomass estimation (Diallo et al., 1991 ; Wylie et al., 1991). Likewise, temporal NDVI 
profiles have been used to classify or to study the dynamics of major vegetation types (Justice et al., 1985 ; 
Goward et al., 1987). In spite of these qualitative encouraging results, high uncertainties appear when 
considering quantitative results. Different factors act on satellite instruments and solar radiation which can 
hinder or transform satellite measurements, drastically decreasing results accuracy. Among main factors we can 
mention : instrument calibration, temporal drift, cloud occurrence, radiation scattering and absorption by 
atmospheric components, off-nadir viewing. Goward et al. (1991) have made a complete review of these 
influencing factors. According to them, deviations in excess of 50 % between satellite measurement and 
equivalent ground observation can occur if correction efforts are not undertaken. 
Some causes of deviation are relatively well-known, such as data calibration or atmospheric attenuation and 
different authors have proposed methods to suppress or at least to reduce these error sources. For example, 
calibration coefficients are derived from the temporal evolution of satellite measurements over selected ground 
targets (Holben et al., 1990 ; Teillet et al., 1990). Other effects such as cloud contamination and variations of 
view angles are presently not well-understood or well-controlled. However, methods to attempt to eliminate 
supposed erroneous data along NDVI temporal profiles are proposed such as : the Maximum Value Composite 
(MVC) technique (Holben, 1986) and the Best Index Slope Extraction (BISE, Viovy et al., 1992). In this last 
method, when considering temporal NDVI profile of one pixel, a decrease of the signal that lasts no more than a 
pre-defined period is considered as noise and eliminated. One can see that if the chosen period is too long, values 
corresponding to effective plant response can be eliminated. On the contrary, if the chosen period is too small, 
noise due to a long cloudy period will not be excluded. 
Here, we propose a method which intends to take into account these problems. Furthermore, a radiative 
temperature threshold is used to filter-cloud contaminated satellite records. Preliminarily to these operations, 
data pre-processing are realized, considering present knowledge, in order to provide as clean as possible NDVI 
time-series. 
93