@PhDThesis{Azeredo:2017:ExPaCo,
author = "Azeredo, Marcio",
title = "Minera{\c{c}}{\~a}o e an{\'a}lise de trajet{\'o}rias de
mudan{\c{c}}a de cobertura da terra: explorando padr{\~o}es
comportamentais no contexto da degrada{\c{c}}{\~a}o florestal",
school = "Instituto Nacional de Pesquisas Espaciais (INPE)",
year = "2017",
address = "S{\~a}o Jos{\'e} dos Campos",
month = "2017-03-23",
keywords = "trajet{\'o}rias de mudan{\c{c}}a, cobertura da terra,
degrada{\c{c}}{\~a}o florestal, padr{\~a}o comportamental,
minera{\c{c}}{\~a}o de trajet{\'o}ria, change trajectory, land
cover, forest degradation, behavioral pattern, trajectory data
mining.",
abstract = "Considerando os processos que envolvem coberturas florestais, o
Processo de Degrada{\c{c}}{\~a}o Florestal {\'e}
particularmente importante, uma vez que, mantidas as
condi{\c{c}}{\~o}es que lhe d{\~a}o origem, altera a estrutura
da floresta de forma lenta e progressiva. Para a cobertura
florestal, implica na redu{\c{c}}{\~a}o das fun{\c{c}}{\~o}es
ecol{\'o}gicas e do armazenamento de carbono, na
fragmenta{\c{c}}{\~a}o de ecossistemas e na perda do potencial
do uso florestal para atividades econ{\^o}micas. Dada a sua
relev{\^a}ncia, {\'e} preciso n{\~a}o s{\'o} compreender a
din{\^a}mica de como tal processo ocorre, mas tamb{\'e}m onde,
quando e como se comportam os atores e os mecanismos associados a
essas altera{\c{c}}{\~o}es. Envolvendo etapas de maior e menor
intensidade e at{\'e} mesmo a possibilidade de revers{\~a}o, a
degrada{\c{c}}{\~a}o florestal requer longos per{\'{\i}}odos
de observa{\c{c}}{\~a}o em grandes bases de dados
espa{\c{c}}o-temporais, de forma continuada e sistem{\'a}tica,
definindo as Trajet{\'o}rias de Mudan{\c{c}}a de Cobertura
Florestal. Tais trajet{\'o}rias, por sua vez, s{\~a}o
identificadas a partir das altera{\c{c}}{\~o}es recorrentes
presentes nas propriedades das unidades de an{\'a}lise observadas
e utilizadas na sua constitui{\c{c}}{\~a}o. Para a
identifica{\c{c}}{\~a}o e explora{\c{c}}{\~a}o das referidas
trajet{\'o}rias, s{\~a}o definidos e ampliados conceitos
estabelecidos na literatura de trajet{\'o}rias de objetos
m{\'o}veis. Desse modo, esta Tese prop{\~o}e, formaliza e
implementa, na forma de uma biblioteca de fun{\c{c}}{\~o}es
parametriz{\'a}veis, os elementos que permitem estabelecer uma
nova metodologia computacional para auxiliar analistas na
explora{\c{c}}{\~a}o de grandes bases de dados no
dom{\'{\i}}nio dos estudos florestais, por interm{\'e}dio da
minera{\c{c}}{\~a}o de padr{\~o}es de trajet{\'o}rias e de
seus agrupamentos. Para tal, este trabalho traz duas
contribui{\c{c}}{\~o}es: (1) define e implementa os Padr{\~o}es
Comportamentais de Converg{\^e}ncia, Encontro,
Detec{\c{c}}{\~a}o de Inconsist{\^e}ncias, Detec{\c{c}}{\~a}o
de Anomalias, Rebanho e Lideran{\c{c}}a, encontrados na
literatura de objetos m{\'o}veis, para o contexto das
Trajet{\'o}rias de Mudan{\c{c}}a de Cobertura. Tal conjunto de
defini{\c{c}}{\~o}es foi aqui denominado de Behavioral Patterns
Mining on Land Cover Change (BPML); e (2) define e implementa uma
metodologia para agrupar as trajet{\'o}rias de mudan{\c{c}}a de
cobertura florestal, aqui denominada de Grouping by Similarity of
Temporal Evolution (GSTE), baseada nas semelhan{\c{c}}as entre as
evolu{\c{c}}{\~o}es temporais dessas trajet{\'o}rias. Esta
metodologia utiliza de forma combinada os algoritmos
computacionais Dynamic Time Warping (DTW), Classical
Multidimensional Scaling (CMDS) e K-Means Clustering. Como prova
de conceito, tr{\^e}s estudos de caso foram conduzidos, nos quais
os padr{\~o}es comportamentais (BPML) e o m{\'e}todo de
agrupamento de trajet{\'o}rias por semelhan{\c{c}}a de
evolu{\c{c}}{\~a}o temporal (GSTE) foram testados em dois
conjuntos de dados de degrada{\c{c}}{\~a}o florestal referentes
{\`a}s regi{\~o}es do entorno dos munic{\'{\i}}pios de Novo
Progresso - PA e Sinop - MT. As referidas bases de dados
utilizadas s{\~a}o constitu{\'{\i}}das por 27.815 e 27.367
unidades de an{\'a}lise (c{\'e}lulas), respectivamente, com
resolu{\c{c}}{\~a}o espacial de 1x1km, resolu{\c{c}}{\~a}o
temporal de 1 ano e extens{\~a}o temporal de 28 anos (1984 a
2011). ABSTRACT: Considering the processes that involves forest
cover, the Forest Degradation Process is particularly important
because, keeping its original conditions, the structure of the
forest is changed in a slow and progressive way. Regarding the
forest cover, such process implies in a reduction of ecological
functions and carbon storage in the fragmentation of ecosystems
and the loss of the forest use potential for economic activities.
Given its relevance, it is necessary not only to understand the
dynamics of how these processes occur, but also where, when and
how the actors and mechanisms associated with those changes
behave. Concerning higher and lower intensity stages and even the
possibility of reversal, forest degradation requires long periods
of observation in large space-time databases, in a continuous and
systematic way, defining the Forest Cover Change Trajectories.
These trajectories, on the other hand, are identified from the
recurrent changes in the properties of the units of analysis
observed and used in its constitution. For the identification and
exploration of these trajectories, this work defines and expands
concepts established in the literature of moving objects
trajectories. Thus, this thesis proposes, formalizes and
implements, in the form of a library of parametrized functions,
the elements that allow the establishment of a new computational
methodology to assist analysts to deal with large databases of
forest studies, through the data mining of trajectory patterns and
their groupings. This study brings two innovative contributions:
(1) it defines and implements the Behavior Patterns of
Convergence, Encounter, Inconsistencies Detection, Anomalies
Detection, Flock and Leadership, found in the literature to deal
with moving objects, in the context of Forest Cover Change
Trajectories. This set was called Behavioral Patterns Mining on
Land Cover Change (BPML); and (2) it defines and implements a
methodology to group the Forest Cover Change Trajectories, here
called Grouping by Similarity of Temporal Evolution (GSTE),
considering the similarities between the respective temporal
evolutions and using computational algorithms of Dynamic Time
Warping (DTW), Classical Multidimensional Scaling (CMDS) and
K-Means Clustering. As proof of concept, three case studies were
generated and the behavior patterns (BPML) as well as the method
of trajectory grouping by similarity of temporal evolution (GSTE)
were tested in two sets of forest degradation data, referring to
the regions surrounding the municipalities of Novo Progresso - PA
and Sinop - MT. These databases are composed by 27,815 and 27,367
units of analysis (cells), respectively, with spatial resolution
of 1x1km, temporal resolution of 1 year and temporal extension of
28 years (1984 to 2011).",
committee = "Santos, Rafael Duarte Coelho dos (presidente) and Monteiro,
Ant{\^o}nio Miguel Vieira (orientador) and Escada, Maria Isabel
Sobral (orientadora) and Ferreira, Karine Reis and Vinhas,
L{\'u}bia and Pinheiro, Taise Farias and Davis Junior, Clodoveu
Augusto",
englishtitle = "data mining and analysis of land cover change trajectories:
exploring behavioral patterns in the context of forest
degradation",
language = "pt",
pages = "150",
ibi = "8JMKD3MGP3W34P/3NGPJ7H",
url = "http://urlib.net/ibi/8JMKD3MGP3W34P/3NGPJ7H",
targetfile = "publicacao.pdf",
urlaccessdate = "27 abr. 2024"
}