joints.
Data Analysis
The data analysis, in the remote
sensing data flow and Geographic
Information Systems, involves the
relationship between the data variables
and the subsequent interferences that
can be developed. This error
acumulation stage focus on the statistic
techniques validation. Difficulties on
the statistic analysis of space data
source involves the typical adoption of
the common linear model, composed by the
Space autocorrelation effects. The data
analysis is also subject to errors
proceeding from the variability of the
analysis specialists. This: variability
involves the choice of pre-defined
relevant variables or the synthesis of
new multiple parameter variabies,
correlated or not.
The classification systems can be a
Source of error in the remote sensing
data integration in Geographic
Information System. Some of the
potential error sources induced by the
classification systems are related to
the inability of the systems to
categorize mixed classes, transition
Zones or ambigous classes definitions,
the human subjectivity and the lack of
compatibility between different
classification systems used with remote
sensing data and traditional data types.
Thematic data levels created using
general remote sensing data require the
use of some kind of classification
system to facilitate the data
categorization for subsequent space data
analysis by GIS. When we work with
mixed pixels, or polygons and transition
zones, or dynamic systems, inconsistent
classification occur with the whole
classification system. This inserts an
error element which is particularly
difficult to be quantified.
The data generalization is usually
made during the remote sensing analysis
by two reasons: space resolution and/or
thematics or spectral data resolution.
The space generalization involves the
analysis to produce the minimum unit of
the map. The sampling for a space
resolution better than the original
results in a substantial error.
Spectral generalization can be executed
by filters that enhance certain
characteristic like the frame or
homegenize similar pixels. Some filters
preserve the frames, while reduce the
noise. Although, because some filters
change the original value of the pixels,
some errors, like the precise frame
location or the loss of spectral
similar, still the only resource, may
happen.
The data generalization after
classification takes two patterns,
spacial and thematics. The thematics
generalization 1s the classes grouping
to create significant homogeneus
categories. Because of this, it is made
with the analyst interference and the
tendency errors can be introduced and
information can be lost if the analyst
352
does not recognize a unique resource,
Data Conversion
By the growth of the Use of ‘the
Geographic Information Systems and the
necessity of incorporating digital
remote sensing data as a quick and
reliable information source, it was
inevitable the necessity of converting
matrix data into vectorial format data.
Matrix format data are data arranged
in a symmetrical spacing and same grid
size. Satellite data are common
examples of matrix data. These data are
easily loaded into a computer, as a
number matrix.
Vectorial data keep the actual shape
of a polygon, using series of vertexes
connected by lines. Vectorial data are
preferred for the representation of the
most thematic maps in the GIS, due to
the uniform lines for the frame viewing.
Most part of the mapped products,
including the result of photo-
interpretation is usually represented in
vectorial format.
Unfortunatelly, there is sign-
ificative error in the conversion from
matrix to vectorial format and : from
vectorial to the matrix format. The
magnitude of these errors depends on the
algorithm used in the conversion
process,on the cell grid used and on the
orientation used for the matrix
representation. By no considering this
error it is possible to insert
considerable problems in the analysis.
Error Avaliation
The quantitative error analysis can
be done during any stage of the data
processing, including data acquisition.
Theoretically, an error value must be
determined after each stage of the
analysis, but in remote sensing
projects, the error value ratios are
determined only after the complete data
analysis is made and usually directed to
the thematic and location accuracy.
In determining the accuracy ratios an
important point to be considered is the
sample sizes. The high cost of each
sample point carts the reduction of the
sample size to its minimum, but keeping
enough size to validate any statistic
analysis. There are several
recommendations and different equations
to define the appropriate sample size.
The sample scheme used is an
important factor in the accuracy
evaluation. The error matrix must be
representative for the whole classified
image, and the choice of an
unappropriated scheme (poor) may result
in a tendentious error matrix, causing
over or underestimation of the real
accuracy.
It is said that space autocorrelation
occurs when the presence or the absence,
or the degree of a certain
characteristic afects the presence,
absence, or degree of the same
characteristic in the neighbor units.
This condition is particularly important
in the evaluation of the accuracy for
positive or negative error influences in
ne:
sel
Me:
fra
re
The
av
re)
mii
po:
Al'
th:
the
the
exi
tol
ha:
the
gei
er]
sai
cul
the
the
in
the
ma
the
ea
the
pre
ery
der
des
mea
agr
prc
per
inc
cor
dif
agr
mad
cal
of
the
cal
in
rep
map
inc
by
err
per
map
Fin
of
Inf
pro
and
adın
sev
the
sta
sta
ana