139 
was only one or two available pixels, restricting the alternatives for pixel selections. Third, only one pixel was to be 
selected by these algorithms with each compositing period, the detailed variation within each period may have been lost, 
unable to pick anomalies occurred within compositing period. Consequently, all of these compositing algorithms, except 
for AVG, resulted in the similar temporal profiles of NDVI of the four study sites. 
The results from SW algorithm is compared with that from BISE algorithm and the raw NDVI temporal 
profiles in Figure 3 for the four selected study sites. The dotted lines are the NDVI temporal profile before compositing, 
and the solid lines are that of BISE, while the dashed line is that of SW algorithm. One can see some detailed 
variations of the NDVI evolutions were retained for the four study sites, even within each composite period. Apparently, 
both SW and BISE retained more valuable pixels than that of MVC, or Min Chi, or Max Ch4, or Min View, or AVG. 
The BISE resulted in a profile that capped the NDVI in general. However, it sometimes also selected pixels that were 
apparently due to the appearance of clouds such as the sharp and deep valleys at day of year (DOY) of 248 at Millet 
and Tiger Bush sites and at DOYs 254 and 256 for the Tiger Bush site (Figure 3). The general pattern of the BISE 
can be described as if the BISE ’walked’ on the top of the NDVI, skipping most valleys. In comparison, the SW did 
not result in sharp and deep valleys, and it can be said that the SW ’walked’ on the top of the NDVI profile, as did 
the BISE, but never fell into the valleys. 
Both the SW and the BISE, in general, reduced the high frequency noises as seen in the NDVI profiles caused 
by external factors but retained more valuable pixels than MVC and other algorithms. One unique feature of both the 
SW and the BISE temporal profiles was that the deep valleys almost always corresponded to a temporal discontinuity. 
Some compositing periods had a limited number of pixels, sometimes only one or two. Because the small number of 
pixels limited the choice for pixel selections, the noise (if any) after compositing probably would most likely be resulted 
from the lack of data. Because of the discontinuity of the data sets, the BISE and the SW selected lower value pixels. 
This shortcoming of the BISE and SW can be overcome by increasing the slide window (or compositing period). 
However, increasing the slide window period will result in the loss of valuable data. 
Figure 2. Composited results of the four selected sites using MVC (solid lines), Min View (dotted 
line), Min Chl(dash lines), Max Ch4 (long dashed line), and AVG (circles) compositing algorithms, 
(a) Fallow sites, (b) Degraded Fallow site, (c) Millet site, and (d) Tiger Bush site.