hul 2004
factors
uch as
cycles,
nmental
ses the
ver the
solution
mental
prrected
Is or an
teristics
eration.
:al state
he data
Irologic
dule are
ormed.
Yetwork
the soil
ve have
ality is
nce we
special
ipplied,
niques,
nent.
ations.
ghtness
? scene
image
| can be
ference
tion are
as little
ble, the
adiance
ns the
dditive
stance,
— dark
ignores
etween
'elative
ethods
nd the
spatial
ixel is
processed in order to represent the amount of illumination it
should.
The spatial registration of a remote sensed image to a map
projection is necessary in order to locate with precision the
changes occurred in the studied zone. Geocoded images are
directly available from data distributors but additional
registration must be necessary for zones with complex terrain
configuration and when DEM used by data suppliers are not of
confidence. A good way to perform geocoding was to measure
ground control point in the field with GPS and use this data set
to calculate the correct rectification. The ground control points
are permanent, static features in the field and easily identifiable
on the image.
The most used algorithms for change detection are (Jensen):
- Change Detection Using Write Function Memory Insertion.
This is an analog method for qualitatively assessing the changes
in a region and do not provide quantitative information of the
changes occurred.
- Multi Date Composite Image Change Detection. Multiple data
sets are inregistred to a single database. This composite data set
can be used to extract information by unsupervised
classification techniques with a result of a class with change and
a class with no change. Another method is the principal
component analysis.
- Image Algebra Change Detection (Band Rationing and Band
Differencing). Subtracting an image from the other one
performs image differencing. The result is formed by positive
and negative values in areas of radiance change:
Dj = BVijg (1)- BV (2) € (4)
Where: Dj, change value pixel, BV;(1)= brightens value at
time 1, BV;;(2)= brightens value at time 2, C= constant used to
transform the negative or positive results in positive results
(normally the results are ranging in the interval —250 to 250)
i= line number, j=column number, k= band number.
The essential aspect of this process is the threshold selection of
boundaries between change/no change zones.
5.1. Spectral Change Vector Analysis
Areas with changes have a different spectral response. The
vector describing the direction and the amplitude of the change
from image 1 to image 2 is the spectral change vector. The total
change/pixel (CMpixe1) in n-dimension spectral space is :
a 2
CM pixel = D le Vik(date2) - 8 Vik(dater)| (5)
where: BV;ik(date2). ïjk(date1) = Pixel values for date 1 and date 2 in
band k .
5.2. Change detection error matrix
In order to assess the accuracy of the change detection
procedures is recommended to generate an error matrix. The
columns of an error matrix contain the reference data and the
rows represent the results of the remote sensed classified data.
This is an effective way to represent accuracy of each classified
category :
The error matrix is a multidimensional table, its cells contain
change data from a category to another. The statistical approach
of the accuracy assessment consists of different multivariate
stastitical analysis. À used measure is KAPPA (Cohen, 1960).
KAPPA is designed to compare results from different regions or
different classifications.
The KHAT statistic is:
783
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
n n
N EX X(X;*X,;)
K = {=1 [zl (6)
NZ T(x; x ;)
[+ +/
i=]
where n is the number of rows in the matrix, X;=number of
observations in the row i and column I, X;.= the marginal totals
of row i and column I, N=total number of observations.
CONCLUSIONS
Change detection is an important tool for environmental studies,
assessing the accuracy of change detection products is an
important step for the integration of remote sensed data to
environmental management system as a decision support tool.
In assessing environmental changes based on remote sensed
data, the major impediment is that the estimate values are
difficult to compute due to the complexity of the processes
involved and more often the reference data is not available for
computing accuracy. A specific attention must be given to
different methodologies to detect changes and error matrix
construction, as a function of change susceptibility of the
studied area.
In order to improve results in change detection several aspects
must be considered:
A budget of the sedimentary regime is needed; The
shoreline topometry at several time interval, with the same
precision of the measurements in order to obtain a dynamic 2D-
3D model of the region to determine substrate variability and
change detection at spatial and temporal scales of high
resolution. It is also important to know parameters and
boundary conditions controlling coastal evolution and geologic
framework such as tectonics, sea-level movements, storm and
other changes in sediment source and paleogeography; Constant
in situ observations, correlated with remote sense data
acquisition. In the space of two decades, application of this
methodology of investigation allowed a better understanding of
the coastal line evolution trends greatly improved understanding
of the coastal zone dynamics.
References
Eckhardt,A. 1990, Automated Update of an Irrigated Lands
GIS Using SPOT HRV Imagery- Photogrammetric Ingineering
& Remote Sensing 56(11) pp.1515-1522.
Fisher, P.H. et al. 1962, The Ordering of multitemporal Fuzzy
Land Cover Information Derived from Landsat MSS Data.
Freund E.J.,1962, Mathematical Statistics, Englewood Cliffs,
N.J. Prentice Hall .
Jensen, J.R. et al. 1987,Improved Remote Sensing and GIS
Reliability Diagrams, Image Genealogy Diagrams and Thematic
Map Legends to Enhance Communication- International
Archives of Photogrammetry and Remote Sensing
6(B6): pp.125-132.
Kohrram S.,1999, Accuracy assessment of remote sensing-
Derived Change Detection- ASPRS.
Richards A. John,1986, Remote Sensing Digital Image
analysis- An Introduction- ed. Springer- Verlag .
Ripple W.J.,1994, The GIS Application Book- Examples in
natural resources, a compendium- ASPRS.
Rosenfield G.H.,1982, Sampling for the thematic map accuracy
testing. P&RS vol 48, pp.131-137.
Van Gendren J.L.,1978, Remote Sensing: Statistical Testing of
thematic Map accuracy, Remote Sensing of Environment, 7,pp.
3-14.
Acknowledgements Authors thanks the CRUTA team
for help and support
