1:250,000 TM-LANDSAT color composite
covering the park. In the first step map
corners and basic features such as rivers
and roads are transferred into a
transparent sheet (overlay). Next, this
overlay is superimposed over the image
and registered by means of common rivers
and roads. Then the interpreter starts
the procedure of extracting and drawing
image content. This situation requires
the image and the map to have the same
map projection. Moreover, the image must
be precisely at 1:250,000.
Unfortunately this condition is rarely
achieved. The interpreter can not perform
a precise matching between image and map.
The traditional solution for this problem
is the use of local adjustments. The more
image scale differs from 1:250,000, the
larger is the number of required local
adjustments. Although this method is a
valid procedure taking into account the
absence of GIS or Computer Cartography
facilities, its basic characteristic is
the subjectivity in defining the sequence
and the number of local adjustments. It
is impossible for a second interpreter to
perform exactly the same registration
between image and map.
1.3 Conflict with accurate GIS database
The idea of accuracy in a GIS database is
strongly related to the capability of
evaluating errors in every step of the
current processing. This means, for
instance, that the lineage of a map has
to be known in order to estimate its
accuracy. If map inconsistencies and
their sources are known, it ^ will be
feasible to use a mathematical model to
correct them.
The digitizing of a map can be used ag a
good example in this discussion. One
operation that precedes (or follows,
depending on the system) the digitizing
is the establishment of the relation
between map and digitazing table
coordinates. An affine transformation is
normally used with this purpose
(Burrough, 1996). This function also acts
as a reliable model to correct map
errors, in case these errors represent
the effects of rotations, translations
and scale factors. But who knows the
result of applying such model over a map
without lineage documentation? How does
an affine function act over that overlay
produced by successive subjective local
adjustments? It is very difficult to get
a reasonable answer. One can just believe
that everything is correct.
This is the basic conflict. An overlay
produced by the conventional
photointerpretation technique can not be
perfectly modelled by any algorithm,
because it is impossible to represent by
mathematical functions all those
heuristic adjustments made by the
interpreter. Then, it is very hard to
access the accuracy of the resulting
digitized data. If the idea is to produce
an accurate GIS database, one must avoid
to generate that kind of overlay as an
input source of data.
702
2. PROPOSED METHODOLOGIES
2.1 Amazonia deforestation database
The Amazonia deforestation database is a
project being carried out by INPE and
will contain Full information of
deforestation process for years 1975, 78,
85, 88, 89, 90 and 91. "Legal Amazonia",
as defined by the Brazilian Constitution,
1988, includes the States of Acre, Amapa,
Maranhao (only west of 44W), Amazonas,
Mato Grosso, Para, Rondonia, Roraima e
Tocantins.
LANDSAT images have been used as a basis
for monitoring each deforestation area.
The database is organized in blocks of 1
degree of latitude by 1.5 degree of
longitude, comprising 1:250,000 map
sheets. These blocks are individual
projects for purposes of data digitizing,
analysis, and plotting of maps. Besides
the 1:250,000 map sheet partition, the
Study area can be subdivided in different
non-overlapping cells which cover the
whole area (Meira Filho, 1991):
- subdivision by state;
- subdivision by municipality;
- subdivision by nominal LANDSAT 4/5
scene;
- subdivision by nominal LANDSAT 1/2/3
scene;
- Subdivision by class of vegetation.
The database was constructed using SGI, a
geographical information system developed
in the Image Processing Division of INPE,
that runs on IBM-PC compatible
microcomputers. Some new functions were
added to the basic software to improve
the ‚control. of errors. and. to provide
faster processings. An example is the
automatic definition of nominal LANDSAT
cells given a 1:250,000. . map sneet
(intersection of two different
subdivisions). Figure 1 shows a typical
intersection between nominal LANDSAT
Scenes and a 1:250,000 map sheet.
Figure 1 - Nominal LANDSAT cells
The automatic definition of the nominal
LANDSAT cells is followed by a polygon
overlay operation that takes into account
the subdivision by states. The resulting
infolayer contains the so called basic
cells. Figure 2 illustrates basic cells:
Figure 2.- Basic cells
<< LX Fw. Fro 0007 O0 O0 (A $e AU HH
(SUM u.e€E L1. re b pe ^i PPM F^ fX Lit Lis lhda £F. A. M
7 — 9 BL Ab“ o e e S FS
2 A, ^ Bed "pu, an
o. ali d dl d
