4. NOAA-AVHRR DATA PROCESSING AND
ANALYSIS
In this paper AVHRR data undergoes projection transformation,
geometry rectification, maximum value compose of NDVI,
NDVI calculation and image classification so that AVHRR
data transferred from Raw Data to the final data used in
analysis.
4.1 Projection transformation and geometry rectification
NOAA satellite does not measure parameter of the real satellite
orbit every day. Within the measure spacing the satellite orbit
parameters are obtained by forecast so that there are
accumulated errors. Since NOAA satellite scanning range is
wide so that at the margin of the image, the degree of pels
aberration is very heavy. AVHRR data offered by CLASS have
not geographic coordinates projection, but head file provided
enough Ground Control Point that could be the conversion of
geography coordinate system and the conversion of projection
system.
We pick up the data that is in the middle of scan strip,then
Transform original data Into Universal Transverse Mercator
( UTM ) Zone N47 projection,Datum-WGS-84.But this
initial projective data commonly have several or tens pixel
errors. NOAA/AVHRR images must be rectified if the data is
put into the practical application(Wei Ya-xing et al,2005;
Huang Jing-feng et al,2000). However, In this paper, the
polynomials correction techniques are used to deal with the
data by referring to 1:1,000,000 digital map of headstream
region of Yellow River, the corrected result is within the error
of one pixel.
4.2 Data Synthesis
Most AVHRR data have much cloudiness (or cloud amount),
which produces difficulties to the application of the data(Wang
Run et al,2005; Huang Yong-jie et al,2003). There are many
kinds of cloudiness processing techniques of NOAA-AVHRR
data. The research adopts the maximum value combination
(MVC) method, which carries out the synthesis processing of
removing cloudiness to deal with a lot of data by computing the
Fig.3 shows the variation tendency of the four kinds of land
cover type that NDVI data reflects.
As shown in Fig. 3, the overall NDVI variation tendency of the
study district is that the areas of the naked, water-body,
desertified land and the low vegetation cover land increase year
by year, and the one of the high vegetation cover land drops
suddenly, but the area of the middle vegetation cover land does
not change much. The reason of this kind of change can be
explained through the way of the transform sequences of the
land cover type. The high vegetation covers land degenerates
continuously and becomes the middle vegetation cover land,
some middle vegetation cover land becomes the low vegetation
cover land. The degradation speed of the low vegetation cover
land is slower than the one of the former two, but some low
vegetation cover land also degenerates to the desertified and
naked land. Therefore, in statistics, the area of the high
vegetation cover land reduces most soon, and the area of the
low vegetation cover land takes the second place, and the one
of the naked, water-body, desertified land increase minimally,
and the area of the middle vegetation cover land does not
change much in general or increase slowly.
NDVI maximum value.( Holben B N,1986) The specific
method is to firstly select data of low cloudiness more than 3
scenes month by month, and then combine these data to a
month maximum value by using the MVC method, and then
combine each three month maximum values to a quarter
maximum value, and finally combine four quarter maximum
values to an annual maximum value. The actual calculated
result proves that the MVC method shows good effects of
removing cloudiness besides the water-body and snow-ice
areas. The annual maximum value after the synthesis can
represent the best condition of the vegetation growing in this
area for this year.
4.3 NDVI calculation and statistical properties
Vegetation index can reflect the state of vegetation growing
and regional vegetation distribution. At present we can employ
the ratio vegetation index (RVI), the difference vegetation
index (DVI), the perpendicular vegetation index (PVI), the
normalized difference vegetation index (NDVI) etc. Among
them NDVI is the most extensively employed vegetation index
at present. The computing method of NDVI is (NIR-R) / (NIR-
R). Calculating specifically in NOAA-AVHRR data is (CH2-
CH1) / (CH2+CH1), and the value range is between -1 and 1.
The research area of the Yellow River Source lies in Qinghai-
Tibet Plateau. With sparse population, the land cover type
changes mainly from the changing vegetation, being reflected
by NDVI is that NDVI varies with time. Regional NDVI lattice
and point numbers in the different value range can represent the
general variation tendency for various land cover types in the
region. We have counted the lattice and point numbers of six
NDVI raster data of 1990, 1992, 1994, 1996, 1998 and 2000 in
the research area of the Yellow River Source by interval of
NDVI fetching value of 0.01, and carried out the clustering
analysis. Through analyzing the practical meaning of
classifications and the contrast and amalgamation among
classifications, the change of NDVI is divided to four big
classes, which represent roughly 4 kinds of land cover type, i.e.
the water-body, naked or desertified land, the low vegetation
cover land, the middle vegetation cover land, and the high
vegetation cover land
Area Change of Major NDVI Class
♦ Higher Vegetation Cover Land A Middle Vegetation Cover Land
▼ Lower Vegetation Cover Land ■ Naked or Desertized Land
Fig.3 Area Change of
Cluster Analyse Generated NDVI Class
4.4 Images classification and results analysis
Statistics and classifications can not represent the detailed
change of each specific land cover type, it is necessary to
utilize the categorized method of the remote sensing image to
divide the land cover type and analyze the changing state and
space distribution of the land cover type every year by
comparison. In addition, because the space resolution ratio of
