953 
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
each land cover type in one year. The MOD 15 LAI and fAPAR 
are 1-km products provided on a daily and eight-day basis. 
Where the globe is tiled into 36 tiles along the east-west axis, 
and 18 tiles along the north-south axis, are each approximately 
1200x1200 km. A brief summary of the LAI algorithm is 
provided by Myneni(R. B. Myneni 2002). The algorithm is 
based on rigorous three dimensional radiative transfer (RT) 
theory(R. B. Myneni 1990). A lookup table (LUT) method is 
used to achieve inversion of the three-dimensional RT problem. 
The 250- and/or 500-m resolution bands are aggregated into 
normalized 1-km resolution grid cells prior to ingestion(R. E. 
Wolfe 1998). 
In this study, based on different types of vegetation and the 
measured data and the simulated data, we can get the statistic 
relationship between LAI and NDVI, and based on the 
relationship, we can estimate the LAI of every pixel. And in 
this study, the quality of MODIS LAI has been analyzed: 
selected the pixels inversed LAI by the brief algorithm, and 
used the filter algorithm for pixels to get the available LAI 
inversion results(Xiao Zhiqiang 2008). 
From the fig.2 and fig.3, we can see that two classification 
maps (MOD12Q1 and Beijing-1 image) match very well to a 
certain precision. Based on the two classification maps, the 
Beijing-1 classification map as the sub-pixels, we can get the 
percents of land types (forest, grass, cereal crop and broadleaf 
crop) in every MODIS pixels. Considering the percentage of 
for one land type more the 95%, we can think this pixel is the 
pure pixel for one land type, and we can get the pixel counts of 
one land type (shown in fig.6). However, because of the coarse 
resolution of MODIS LAI products,lkm, and the pixels are 
most mixed, so in our research, we use pixel unmixing methods 
to get the LAI variation of every land cover types in time series 
at MODIS sub-pixels. The formula is shown as Equation 1. 
L = 'Lf, L J +e ’ 
M 
i = 1,2,3 • • •, w, 
(1) 
f, 
is percent of each land cover 
Where, L is LAI of mixed pixels, 
type in one pixel, ^ is the LAI of J 1 percent land cover type, 
is error. And selecting the mixed pixels, using the linear model 
decompose, then we can calculate LAI at sub-pixels by the 
Least Square method(Wan Huawei 2007). So extracting LAIs 
from the pure pixels of MOD15A2 product, we can get the 
statistic MODIS LAI in time serials of the whole year, shown 
in fig.7. 
a» olfiat 
(a) Broadleaf crop 
Fig. 6 The counts of pure pixels in 
MODIS product matched with Beijing-1 
classification map 
(c) Grass 
Fig.7 MODIS Collection 4 2006 LAI trajectories for the reprojection MODIS LAI 
product: Means and one standard deviation values are shown. 
2.2.3 The generation of time serials LAI of the whole year at 
Beijing-1 spatial resolution 
And using the LAI estimation of BJ-1 image to modify the 
MODIS LAI Variation curve of every land cover types in one 
year, and in the course of modifying, we consider the error of 
BJ-1 LAI estimation and the error of MODIS LAI. There are 
three scenes of estimated LAI of Beijing-1 images, which can 
be used to modify the LAI variation curve for every land types 
of MODIS. And then we can generate the time serials LAI 
(eight-day) of the whole year at Beijing-1 spatial resolution. 
In this paper, the estimated LAI maps of Beijing-1 image in 75, 
134 and 168 day of year are used, which are corresponding to 
the MODIS LAI in 73, 131, and 169 day of year. And the 
MODIS LAIs are considering as the background values, and 
the three scenes of estimated LAI of Beijing-1 as the observed 
values to generate the time serial LAI of each land type. The 
method can be expressed by the Equation.2. In the formula, n is 
the number of observed data; f is the pixel, and X h is the 
background value,. In our study, the background value is the 
time serial MODIS LAI , and the X 0 is the estimated LAI of 
Beijing-1 image. w(r., r. ) is the weight function, which can 
be valued as needed. Thus, we generate the time serials LAI 
(eight-day) of the whole year at Beijing-1 spatial resolution.