The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008
37
These regions were visited during our first field study on
September, 2007 (Figure 3).
Figure 3: A view from a homogeneous region on the lake.
Current Status of the Site:
Xj = digital number attributed to the pixel location
indexj
i = target pixel location index
Let’s define a set A, which contains the location indices of
pixels which are within the chessboard distance d to the target
pixel location i. For example if we choose all eight neighboring
pixels (d = /), this means A has nine elements including
location index i. So;
If jeA;W,j = 1
Otherwise, Wy = 0
Therefore; Equation (1) gives the ratio of the sum of the
weighted DNs within the set of A to the sum of the DNs for the
entire image (Wulder and Boots, 1998).
In the context of application to remote sensing digital images, a
standard version of G, (d) is used by calculation of Z score
standardized form * .
TUBITAK UZAY will be organizing field campaigns at Tuz
Gold during summer period each year. Site will be temporarily
instrumented during campaigns. The maintenance of the site is
funded by TUBITAK and the first campaign is planned to be
held during August 2008.
(Bannari et al., 2005) used the following standardized formula
obtained from Z score formula, to evaluate the homogeneity in
Lunar Lake Playa, Nevada calibration test site.
4. HOMOGENEITY ANALYSIS
As mentioned previously, spatial homogeneity is one of the
important criteria, having influence on the accuracy of the
absolute calibration results. Homogeneity of the area, which can
change over time, affects the site selection and also usability of
the area. Therefore, a special focus on homogeneity will be
given in this section.
For homogeneity analysis, the spatial autocorrelation, the
degree of dependence between the pixels digital numbers
associated with the pixel coordinate is used (Bannari et al.,
2005). The clustering of similar digital numbers denotes
positive autocorrelation while neighborhood of dissimilar
values denotes negative autocorrelation. Global or local
statistics are used to measure spatial autocorrelation. Getis
statistics (Getis and Ord, 1992) is used as a local indicator
which is an example of the local indicators of spatial
association (LISA) (Bannari et al., 2005).
Gi (d) =
j
s[W i *(n-W i *)/(«-!)]'
(2)
where w* = '£w ij (c1)
(n-(x) 2 )
The Getis Statistics, explained above, is conducted on Tuz Golii
calibration test site for homogeneity analysis using the NIR
band of MODIS (LPDAAC, 2007) image taken on 20.07.2007
and d is taken as unity.
The statistics G*(d) for some chessboard distance (Gonzalez,
2002) d is defined as (Wulder and Boots, 1998):
'L w vW x j
G/(</) = ^=-
L x j
(i)
The resultant image showing Tuz Golti homogeneity, calculated
using Getis Statistics, is given in Figure 4. According to this
statistics, if the target pixel and its neighborhood pixels have
similar high values, Getis statistics gives a high value. If the
target pixel and its neighborhood pixels have similar low values
Getis statistics gives a low value (Wulder and Boots, 1998).
Therefore; from the results, it is clear that there is a large
where w.. (¿/) = a spatial weight matrix
’ Z score standardization is:
X-jU
T
of x and T is standard deviation of x.
where JU is the mean