135 
ON CURRENT COMPOSITING ALGORITHMS 
J. Qi 
USDA-ARS Water Conservation Laboratory, 4331 E. Broadway, Phoenix, Arizona 85040, USA 
Y. Kerr 
LERTS-CNES-CNRS, 18 Avenue Edouard-Belin, 31055 Toulouse Cedex, France 
ABSTRACT: 
Several techniques exist for compositing the multitemporal Advanced Very High Resolution Radiometer (AVHRR) data 
for vegetation studies. The major pixel selection criteria of these techniques rely cm the characteristics of the NDVI: 
appearance of clouds, poor atmospheric conditions, and off-nadir viewing geometries would depress the NDVI values. 
Consequently, selecting the pixels with the maximum value of NDVI would presumably eliminate these external 
pertubating effects. However, the maximum NDVI does not always correspond to these ideal conditions. In fact, the 
NDVI varies with these external factors in an unpredictable way. There was an indication that the maximum NDVI 
tpnrtpti to favor the off-nadir view in the forward direction. The resultant composite product would be consequently 
affected. To improve the multitemporal data via compositing, therefore, both the pixel selection criteria and the classifier 
NDVI need to be modified or corrected for external factors. In this study, the current compositing algorithms were 
reviewed, and alternatives were proposed to use the combinations of the red and near-infrared channels and biological 
characteristics of vegetation as second criteria in pixel selections. In addition, the traditional classifier NDVI was 
replaced with different vegetation indices. The new approach was applied to an AVHRR data set over Hapex study site 
in Niger in 1992. The results showed that the new approach improved the AVHRR time series quality and was 
promising towards the development of an efficient compositing algorithm. The new approaches will be presented and 
limitations will be further discussed. 
KEY WORDS: Composite, AVHRR.Vegetation IndexRemote Sensing, Satellite 
INTRODUCTION 
The advanced very high resolution radiometer (AVHRR) on the National Oceanographic and Atmospheric 
Administration (NOAA) satellite series has been the major sensor that provides scientists with continuous remote sensing 
data at regional and global scales over much of the Earth’s surfaces for global change studies. The major constraints 
of the AVHRR data have been both generic and external problems. The generic problems include the radiometric 
calibrations of the sensors, especially the first two spectral channels for which no on-board calibration is available. The 
external problems are caused by cloud masking, atmospheric contamination, geometric registration, and sensor viewing 
angle variations. Consequently, the observed AVHRR data contain substantial uncertainties, preventing scientists from 
making a quantitative analysis of the Earth’s vegetation dynamics (Gutman, 1991). 
Compositing techniques have been used to reduce uncertainties due to external factors, especially due to clouds 
masking. These techniques involve choosing a subset of data that are cloud-free and have least atmospheric 
contaminations from a large data set. Several techniques exist for compositing. The most popular one is the maximum 
value compositing (MVC) algorithm proposed by Holben (1986). This algorithm first defines a compositing period 
within which the normalized difference vegetation index (NDVI) classifier is assumed to change little. Within each 
compositing period, the pixel with the maximum value of NDVI is selected. The rationale behind this is that clouds or 
poor atmospheric conditions or larger off-nadir view angles depresses NDVI value. Selecting the maximum value of 
NDVI presumably reduces the chances of obtaining cloud-masked, atmosphere-contaminated, and off-nadir viewing 
angle data. Because this technique employs a single pixel selection criterion (maximum NDVI), the quality of the 
composited data relies on the characteristics of the NDVI. 
Although major noises, especially the cloud-masking, can be reduced substantially, problems remain because 
of the nature of the NDVI classifier and became of the lack of precise pixel selection criteria. The NDVI is vulnerable 
to soil background variations (Huete, 1989) and atmospheric conditions (Kaufman, 1989; Goward et al. 1990). The 
maximum NDVI favors off-nadir viewing angles in the forward directions (D’lorio et al., 1990 and 1991). Subsequent 
selection of highest NDVI may, therefore, be biased toward larger off-nadir view angles. In addition, the atmosphere 
and the view angle effects are coupled (Qi et al., 1994a). Reduction in one noise may be offset by an increase in another 
type of noise.