141 
Vegetation Indi 
06 
S 0.4 
1 
I 0.35 
I 0.3 
> 
0.25 
0.2 
Figure 4. SW and BISE composited results with NDVI (solid), SAVI (dotted), MSAVI (dash ), and 
GEMI (with circles) for Fallow (a). Degraded Fallow (b). Millet (c), and Tiger Bush (d) sites. 
CONCLUDING REMARKS 
The MVC compositing algori thm showed similar results to the minimum channel 1 (Min Chi), maximum channel 4 
(Max Ch4), average (AVG), and minimum view (Min View) angle algorithms. This was due mainly to the fact that 
within each compositing period, the candidate pixels were limited, only one or two available in some periods, 
consequently restricting the choices of different algorithms when compositing. This similarity among these compositing 
algorithms may also be due to the threshold (20%) used in this study for the AVG, Min View, Min Chi. and Max Ch4. 
The increase in the threshold would enable one to see the difference between these algorithms and the MVC, but would 
also increase the risk of selecting more noisy pixels. The AVG algorithm appeared to have resulted in a lower temporal 
profiles than other algorithms, but smoother with time. These algorithms most likely omit short-term (high temporal 
frequency) anomalies, although they are able to detect long term (low temporal frequency) anomalies such drought, 
which may last months to years. 
The slide window (SW) algorithm showed a significant difference from either the BISE or MVC algorithms. 
It retained more valuable data than the MVC and contained less noisy pixels than the BISE. The advantage of this 
algorithm is that it retained all valuable data (no waste) while discarding noisy pixels, enabling scientists to monitor the 
earth surface in a fine temporal step confidently, not omitting any anomaly occurred within compositing periods. This 
algorithm is dependent both on its pixel selection criteria, such as the threshold ( 20 % in this study) and on its classifier 
used. A good classifier would certainly increase the reliability of composited products. It should be pointed out, 
however, all of compositing algorithms did not take geometric registration into account. Uncertainty in geometric 
registration may be up to several pixels and, therefore, causes problems since all existing compositing algorithms are 
on "pixel-to-pixel" bases.