The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
3. DATA AND METHOD
3.1 Available datasets
The dataset we used in this study is MODIS 16-day composite
NDVI time series data provided by the Data Center, United
States Earth Resources Observation System (Earth Observation
System, EROS). The spatial resolution is 250 meters. The data
were acquired in its vegetation growing season, from April to
October extending from 2000 to 2006, which can better reflect
the vegetation growth and changes. Geometric correction and
the band image calibration were performed as early processing
step in this research.
The land use/land cover map of Kai County for 2000 was
provided by the Chinese Academy of Surveying and Mapping,
which was classified from SPOT5 images with the resolution of
2.5m by manual interpretation (Fig.2b). This classification map
was used as reference data to assess the accuracy of the fraction
of vegetation cover (FVC) retrieval in our study.
3.2 Data preprocessing
The image datasets were preprocessed including the maximum
value composite (MVC) and cloud removed. To reduce the
image noises from the atmospheric clouds, particles, shadows,
etc, we synthesized two 16-day composite NDVI images to one
32-day composite NDVI image in sequence by using the MVC
method (Ding, 2006 ) . At the same time, monthly NDVI
images were obtained.
Although the MVC method reduce the impacts form the
atmospheric clouds, particles, etc., there are still clouds
pollution. Therefore, to solve the problem, the composite
images are processed using the best index slope extraction
(BISE) methods (Wang, 2005).
dNDVI =
dNDVI l l+i =
(1)
ND VI t _y
(2)
NDVI,.,
Non-dense Vegetation Model are the methods in common use.
According to the sub-pixel structure characteristic, Dense
Vegetation Model based on pixel linear decomposing was
adopted in this study (Gutman, 1998). Based on the resultant
monthly NDVI images, the vegetation fractions from April to
October of each year are calculated. And then, annual
vegetation coverage percent was achieved by averaging the
seven monthly vegetation coverage data in each year, which
can be used to analyze the annual change of vegetation cover.
3.4 Accuracy assessment
Field measurement of pixel points is the conventional method to
estimate the accuracy of vegetation cover fraction. But in this
study it was hardly to survey the vegetation cover fraction in a
unit as large as 250 meter multiplied by 250 meter, so a new
solution using SPOT5 classification images was adopted instead
of field survey. We choose kai County as test area. Firstly, all
the land use/land cover types of SPOT5 classification images
with the resolution of 2.5m were divided into “vegetation” and
“non-vegetation” types which were expressed as “1” and “0”,
on the hypotheses of vegetation types being full vegetation
coverage, and vice versa. Secondly, the vegetation and non
vegetation image was resampled with the spatial resolution of
250 meter, and each new pixel value was replaced by averaging
of corresponding 10,000 sub-pixel values. The resampled image
was used as real vegetation cover data (i.e. reference image) to
estimate the accuracy of FVC retrieval.
(a)
where NDVIt-1 and NDVIt+1 denote the NDVI values of time
t-1 and t respectively; dNDVIt-l,t and dNDVIt,t+l show the
variation rate from t-1 to t and from t+1 to t respectively. It is
assumed that the pixel at time t is affected by clouds if dNDVIt-
l,t and dNDVIt,t+l are both surpass 20%, then the t time pixel
value is corrected by the average of time t-1 and time t+1. We
applied the algorithm to detect the contaminated position point
and smooth the NDVI time series data for all the pixels in our
study periods excluding the starting and ending points. For the
first and end pixels, the improved BISE method is adopted: if
dNDVIl,2 is more than 20%, the first pixel is substituted by the
average of time 1 and time 2. The last pixel value is processed
in the same way.
3.3 Retrieval of FVC
The fraction of vegetation cover (FVC) is the vertical projection
of the crown or shoots area of vegetation to the ground surface
in a unit area, expressed as the fraction or percentage (Guo,
2007; Purevdor, 1998). In general, Dense Vegetation Model and
Fig.2 sampling points of accuracy assessment of FVC retrieval:
(a) sampling points in FVC retrieval image in 2004; (b)
sampling points in classification image in 2004
1398