@MastersThesis{Freitas:2007:TéAnSé,
author = "Freitas, Ramon Morais de",
title = "T{\'e}cnicas de an{\'a}lise de s{\'e}ries temporais aplicadas
{\`a} detec{\c{c}}{\~a}o de desflorestamento em tempo real",
school = "Instituto Nacional de Pesquisas Espaciais (INPE)",
year = "2007",
address = "S{\~a}o Jos{\'e} dos Campos",
month = "2007-02-28",
keywords = "Amaz{\^o}nia, detec{\c{c}}{\~a}o de mudan{\c{c}}as,
desflorestamento, MODIS, sensoriamento remoto, an{\'a}lise de
s{\'e}ries temporais, Amazon region, change detection,
deforestation, MODIS, remote sensing, time series analysis.",
abstract = "A detec{\c{c}}{\~a}o de desflorestamento em tempo real ou
pr{\'o}ximo do real {\'e} de fundamental import{\^a}ncia para
a{\c{c}}{\~o}es conjuntas de car{\'a}ter preventivo e punitivo
dos {\'o}rg{\~a}os governamentais no que tange a
pol{\'{\i}}tica de controle e preven{\c{c}}{\~a}o do
desflorestamento na Amaz{\^o}nia. Neste contexto, este trabalho
tem o objetivo de propor uma metodologia para detec{\c{c}}{\~a}o
de desflorestamento em tempo real a partir de imagens MODIS. A
metodologia consiste em caracterizar e detectar {\'a}reas
desflorestadas atrav{\'e}s de s{\'e}ries espa{\c{c}}o temporais
de imagens MODIS. A {\'a}rea de estudo proposta para
realiza{\c{c}}{\~a}o da pesquisa compreende tr{\^e}s
micro-regi{\~o}es do estado de Mato Grosso que tem sido
caracterizada pela alta taxa de desflorestamento nos {\'u}ltimos
anos. A detec{\c{c}}{\~a}o do desflorestamento em tempo real
utilizou-se das imagens MOD02 di{\'a}rias adquiridas no
per{\'{\i}}odo de 2005 a 2006. Seguindo a metodologia PRODES e
DETER a detec{\c{c}}{\~a}o das {\'a}reas desflorestadas
basearam-se nas fra{\c{c}}{\~o}es vegeta{\c{c}}{\~a}o e solo
derivadas do modelo linear de mistura espectral. Para
constru{\c{c}}{\~a}o das s{\'e}ries espa{\c{c}}o-temporais
foram utilizados os produtos de reflect{\^a}ncia e temperatura de
superf{\'{\i}}cie. A caracteriza{\c{c}}{\~a}o das s{\'e}ries
temporais foi baseada em 4 t{\'e}cnicas: wavelets, an{\'a}lise
de padr{\~o}es de gradientes, diverg{\^e}ncia de
KullBack-Leibler e expoente de Hurst. Os dados de campanhas de
campo, Projeto DETER, PRODES, imagens CBERS e TM, foram utilizados
como verdade terrestre para valida{\c{c}}{\~a}o da metodologia.
A utiliza{\c{c}}{\~a}o de imagens multitemporal do produto MOD02
apresentou uma exatid{\~a}o global de detec{\c{c}}{\~a}o dos
focos de desflorestamento (92,72%) quando comparados com os dados
de verdade terrestre. Com a utiliza{\c{c}}{\~a}o das
transformadas de wavelets foi poss{\'{\i}}vel filtrar e
caracterizar a data e o uso do solo ap{\'o}s o desflorestamento,
i.e., mudan{\c{c}}a din{\^a}mica da cobertura do solo. Com a
an{\'a}lise de padr{\~o}es de gradiente {\'e} proposta uma
metodologia para redu{\c{c}}{\~a}o da dimensionalidade de dados
que permite identificar {\'a}reas desflorestadas. Atrav{\'e}s da
Diverg{\^e}ncia de Kullback-Leibler e Expoente de Hurst foi
poss{\'{\i}}vel analisar a complexidade estat{\'{\i}}stica e
textura das imagens fra{\c{c}}{\~a}o vegeta{\c{c}}{\~a}o para
{\'a}reas desflorestadas e {\'a}reas de floresta. ABSTRACT: The
detection of deforestation in a near real time is of fundamental
importance for Government policy and surveillance of forest areas.
A near real-time detection would allow control of the increase of
new clearings and monitoring of the deforestation pattern and
dynamics in Amazonia. In this context, this work has the objective
to propose a methodology to detect deforestation in near real time
using MODIS images. The methodology consists on to characterize
and detect deforested areas using temporal spatial time series of
MODIS images. The study area is located in the Mato Grosso State,
Brazilian Amazonia, encompassing three micro regions that has been
characterized by high deforestation rates in the last years. The
detection of deforestation in a near real time used daily MODIS
images (MOD02) acquired in 2005 to 2006 time period. Following the
PRODES and DETER methodology the detection of deforested areas was
based on multitemporal soil and vegetation fraction images derived
from linear spectral mixing model. The time-series analysis was
based on the surface reflectance and surface temperature products
aquired from 2000 to 2006. For the characterization of
spatiotemporal time series was used 4 technics: wavelets
transforms, gradiente patterns analysis, Hurst exponent and
Kullback-Leibler divergency. The field campaign data, PRODES and
DETER information, and Landsat TM and CBERS CCD images were
utilized as ground truth for validation of the methodology. The
use of multitemporal images of MOD02 product presented a global
accuracy of 92.72% to detect the deforestation when compared with
ground truth. With the use of wavelets transform it was possible
to characterize the deforestation date and pos-deforestation land
use type. (croplands, pasture or regrowth), i.e., the landcover
change dynamics. The Gradient Pattern Analysis showed a new
aproach to reduce the dimensionality of data volume for in
deforestation detection. The Kullback-Leibler divergency and Hurst
exponent were used to analyze the statistical complexity and
texture of vegetation fraction images for forest and deforested
areas.",
committee = "Novo, Evlyn Marcia Le{\~a}o de Moraes (presidente) and
Shimabukuro, Yosio Edemir (orientador) and Rosa, Reinaldo Roberto
(orientador) and Valeriano, Dalton de Morisson and Haertel, Vitor
Francisco de Ara{\'u}jo",
copyholder = "SID/SCD",
englishtitle = "Times-series analysis applied to deforestation detection and
characterization in real time",
language = "pt",
pages = "197",
ibi = "6qtX3pFwXQZGivnK2Y/Q5cs9",
url = "http://urlib.net/ibi/6qtX3pFwXQZGivnK2Y/Q5cs9",
targetfile = "publicacao.pdf",
urlaccessdate = "15 jun. 2024"
}