![](bg4.png)
geospatial data sets. The data catalog has an
immense volume of data and a wide range of
popular datasets, such as the world's largest
collection of Landsat scenes, 25 years of high-
resolution images, and other Landsat images since
1972, has variety of data types, bands, projection, bit
depth, spatial resolution, temporal. It has data from
the Sentinel, Images Moderate Resolution Imaging
Spectroradiometer (MODIS), night-time imagery-
Defense Meteorological Satellite Program's
Operational Linescan System (DMSP-OLS), digital
elevation models, slope data, surface temperature,
climate, atmospheric data beyond of global daily
satellite feeds.
The user can add and store their own data and
collections. Upload your own image with Maps
Engine, in the original projection, with all the bands
and metadata. In addition to being able to save your
data, collected points, classifications and these can
be used, manipulated and opened in traditional
programs.
The platform can be used in two ways, in the
"Explorer" mode a programming interface that was
used in this work, also contains the "Code Editor"
mode used by users with programming knowledge.
The platform has already been successfully used
for various purposes. Between them:
The European Commission's Joint Research
Center (JRC) has used the Earth Engine to develop
high-resolution maps of global surface water
occurrence, change, seasonality, recurrence and
transitions.
Collect Earth, developed by the Food and
Agriculture Organization of the United Nations
(FAO), is a free, open-source, easy-to-use tool using
Google Earth and Google Earth Engine to view and
analyze land lots to deforestation and other forms of
land use change.
Global Forest Watch, an initiative of the World
Resources Institute, is a dynamic online forest
monitoring system designed to enable better
management and conservation. Global Forest Watch
uses Earth Engine to measure and visualize changes
in the world's forests.
A team led by Matt Hansen of the University of
Maryland used the Earth Engine to research more
than a decade of global warming extension, loss, and
gains. This area is 128.8 million square kilometers,
equivalent to 143 billion pixels of Landsat data in a
spatial resolution of thirty meters.
5 METHODOLOGY
The proposed methodology starts with the choice of
analysis scales and inputs. The CART classifier for
the mesoscale Rio de Janeiro was initially evaluated.
Simplified legends (urban and non-urban) and
greater detailing (different types of coverage of
urban areas and levels of urban intensity) were
tested. The main input was the Landsat TOA (Top of
atmosphere) mosaic. The potential, time of
classification, and results were evaluated.
Using the Google Earth Engine platform in
"Explorer" mode for computer laymen and selecting
images from different years in the platform database,
the classifier was chosen. In this case the CART-
Classification and Regression Trees (The decision
tree method is a supervised learning approach, that
is, it comprises the abstraction of a knowledge
model from the data presented in the form of ordered
pairs (desired input and output) [Goldschmidt e
Passos 2005]. n this method, the production of the
results presents simplicity and readability for its
interpretation, fact that, according to Oliveira
(2005), has become one of the main advantages of
its use. With regard to the CART algorithm, one of
its main characteristics is the research capacity and
relations between the data, involving the
construction and simplification phases of the
decision tree, choosing the best variable for dividing
the data into two nodes, where the procedure is
applied recursively to the data in each of the child-
nodes and so on [Hand et al 2001].) (a classification
algorithm that has one of its main characteristics the
research capacity and relations between the data,
involving the phases of construction and
simplification of the decision tree, choosing the best
variable for dividing the data into two nodes, where
the division procedure is applied recursively to the
data in each one of the nodes (classification is a
process that finds common properties between a set
of records belonging to a database and classifies
them into different classes according to a model)
through the indication of samples pixel-by-pixel of
the different types of subtitles that were used.
Five tests were performed. In the first test we
tried to define the potential of delimitation of urban
areas considering only two classes (urban and water)
and others. in the second test the number of classes
was increased. were selected for vegetation, soil and
sand. In this case, the results were more promising.
In the third test, the classification for the
delimitation of urban areas on a regional scale for
the southeastern region of Brazil was enhanced. In
this case the Nigth time ligth file was used as support.
GISTAM 2018 - 4th International Conference on Geographical Information Systems Theory, Applications and Management
258