@MastersThesis{Dutra:2019:MaMoCo,
author = "Dutra, Andeise Cerqueira",
title = "Mapeamento e monitoramento da cobertura vegetal do Estado da Bahia
utilizando dados multitemporais de sensores {\'o}pticos
orbitais",
school = "Instituto Nacional de Pesquisas Espaciais (INPE)",
year = "2019",
address = "S{\~a}o Jos{\'e} dos Campos",
month = "2019-03-29",
keywords = "Modelo linear de mistura espectral, moderate resolution imaging
spectroradiometer, random forest, Nordeste, Bahia, linear spectral
mixture model, moderate resolution imaging spectroradiometer,
random forest, Northeast, Bahia state.",
abstract = "Conhecer a extens{\~a}o e as taxas de degrada{\c{c}}{\~a}o da
terra para proteger os ecossistemas terrestres de um maior
esgotamento, tem sido uma das quest{\~o}es mais importantes do
nosso tempo. Entretanto, o Nordeste brasileiro tem sido
negligenciado e pobremente estudado tanto em termos de programas
de conserva{\c{c}}{\~a}o quanto de investiga{\c{c}}{\~a}o
cient{\'{\i}}fica. Os produtos de Sensoriamento Remoto
tornaram-se uma importante fonte de informa{\c{c}}{\~o}es para
monitorar as mudan{\c{c}}as de cobertura da terra, no entanto,
ainda {\'e} escasso o n{\'u}mero de estudos para detectar e
monitorar a vegeta{\c{c}}{\~a}o de clima semi{\'a}rido. Nesse
contexto, o estado da Bahia foi selecionado para a
realiza{\c{c}}{\~a}o desta pesquisa por possuir diversas
forma{\c{c}}{\~o}es vegetais e apresentar significativa
mudan{\c{c}}a no uso e cobertura da terra. Devido {\`a} alta
incid{\^e}ncia de nuvens no estado, os produtos provenientes do
sensor MODIS foram utilizados por apresentarem alta
resolu{\c{c}}{\~a}o temporal. Assim, esta pesquisa teve por
objetivo propor uma abordagem de mapeamento e monitoramento do uso
e cobertura da terra para o estado da Bahia utilizando dados
multitemporais provenientes do sensor MODIS, abrangendo o
per{\'{\i}}odo entre 2000 e 2017. Os objetivos
espec{\'{\i}}ficos foram: a) estimar endmembers (pixels puros) a
partir de imagens de melhor resolu{\c{c}}{\~a}o espacial (30 m)
do sensor OLI/Landsat 8 para posterior aplica{\c{c}}{\~a}o do
Modelo Linear de Mistura Espectral (MLME) nos produtos
provenientes do sensor MODIS (250 m) ; b) gerar uma s{\'e}rie
temporal de imagens fra{\c{c}}{\~a}o derivadas do MLME entre
2000 e 2017 utilizando os endmembers estimados na fase anterior;
c) gerar mosaicos das imagens fra{\c{c}}{\~a}o calculando o
valor m{\'a}ximo anual das propor{\c{c}}{\~o}es para os anos de
2000 e 2017; d) gerar dois mapas tem{\'a}ticos base de uso e
cobertura da terra (LULC Land Use and Land Cover) no estado da
Bahia para os anos 2000 e 2017 utilizando os mosaicos gerados na
fase anterior, aplicando o classificador Random Forest; e) avaliar
as acur{\'a}cias dos mapas de LULC obtidos; e f) analisar
qualitativamente as s{\'e}ries temporais das imagens
fra{\c{c}}{\~a}o obtidas para todo o per{\'{\i}}odo analisado.
Os resultados obtidos foram: 1) os endmembers estimados permitiram
a melhoria dos resultados no que se refere ao erro, a
variabilidade e a identifica{\c{c}}{\~a}o das
propor{\c{c}}{\~o}es dos componentes existentes nas imagens,
visto a dificuldade na determina{\c{c}}{\~a}o dos endmembers e
que a escolha indevida de pixels considerados como puros em
produtos de baixa e moderada resolu{\c{c}}{\~a}o espacial pode
afetar a qualidade das imagens fra{\c{c}}{\~a}o para uso
operacional; 2) a abordagem utilizando as imagens
fra{\c{c}}{\~a}o contendo a propor{\c{c}}{\~a}o m{\'a}xima
anual dos componentes reduziu o volume de dados, ao mesmo tempo em
que permitiu a separa{\c{c}}{\~a}o das classes de LULC em
fun{\c{c}}{\~a}o da associa{\c{c}}{\~a}o entre as
propor{\c{c}}{\~o}es de vegeta{\c{c}}{\~a}o, solo e sombra,
extraindo as caracter{\'{\i}}sticas relacionadas aos
padr{\~o}es anuais das classes de LULC; 3) Os mapas de LULC para
os anos de 2000 e 2017 obtiveram acur{\'a}cias totais de 0,77 e
0,67, respectivamente, gerando a hip{\'o}tese de que a seca
severa que atingiu o Nordeste entre 2012 e 2017 influenciou o pior
desempenho do classificador utilizado; 4) o uso de s{\'e}ries
temporais das imagens fra{\c{c}}{\~a}o permitiu o monitoramento
das mudan{\c{c}}as ocorridas na vegeta{\c{c}}{\~a}o e
tamb{\'e}m os impactos que podem estar associados aos eventos de
seca. Assim, a abordagem aqui apresentada demonstra a
potencialidade das imagens fra{\c{c}}{\~a}o para a
classifica{\c{c}}{\~a}o e monitoramento semiautom{\'a}tico da
cobertura vegetal a n{\'{\i}}vel global e regional. ABSTRACT:
Knowing the extent and rates of land degradation to protect
terrestrial ecosystems from further depletion has been one of the
most important issues of our time. However, the Brazilian
Northeast has been neglected and poorly studied both in terms of
conservation programs and scientific research. Remote Sensing
products have become an important source of information for
monitoring changes in land cover, however, there are still few
studies to detect and monitor semi-arid vegetation. In this
context, the state of Bahia was selected to carry out this
research because it has diverse vegetation formations and presents
a significant change in land use and land cover (LULC). Due to the
high incidence of clouds in the Bahia state, the products from
MODIS sensor were used because it presents a high temporal
resolution. The purpose of this research was to propose a mapping
and monitoring approach to land use and land cover for the state
of Bahia using multitemporal data from MODIS sensor, covering the
period between 2000 and 2017. The specific objectives were: a) to
estimate endmembers (pure pixels) from images of higher spatial
resolution (30 m) of OLI / Landsat 8 sensor for later application
of the Linear Spectral Mixture Model (LSMM) in products from MODIS
sensor (250 m spatial resolution) ; b) to generate a time series
of fraction images derived from the MLME between 2000 and 2017
using the estimated endmembers in the previous phase; c) to
generate mosaics of the fraction images by calculating the maximum
annual value of the proportions for the years 2000 and 2017; d) to
generate two LULC thematic maps in the state of Bahia for the
years 2000 and 2017 using the mosaics generated in the previous
phase, applying the Random Forest classifier; e) to evaluate the
accuracy of the LULC maps obtained; and f) qualitatively analyse
the time series of the fraction images obtained for the entire
analysed period. The results obtained were: 1) the estimated
endmembers allowed the improvement of the results regarding error,
variability and identification of the components proportions in
the images, due to the difficulty in determining the endmembers
and that the undue choice of considered pixels as pure in low and
moderate spatial resolution products can affect the quality of the
fraction images for operational use; 2) the approach using the
fraction images containing the maximum annual proportion of the
components reduced the data volume, while allowing the separation
of LULC classes due to the association between vegetation, soil
and shade proportions, extracting the characteristics related to
the annual LULC class patterns; 3) The LULC maps for the years
2000 and 2017 obtained a global accuracy of 0.77 and 0.67,
respectively, generating the hypothesis that the severe drought
that reached the Northeast between 2012 and 2017 influenced the
worse performance of the classifier; 4) the use of time series of
the fraction images allowed to monitoring the changes occurred in
the vegetation and also the impacts that may be associated to the
drought events. Thus, the approach presented here demonstrates the
potentiality of the fraction images for the semiautomatic
classification and monitoring of vegetation cover at global and
regional level.",
committee = "K{\"o}rting, Thales Sehn (presidente) and Shimabukuro, Yosio
Edemir (orientador) and Arai, Egidio (orientador) and Sampaio,
Cl{\'a}udia Bloisi Vaz",
englishtitle = "Mapping and monitoring of vegetation cover in the Bahia state
using multitemporal data from optical orbital sensors",
language = "pt",
pages = "139",
ibi = "8JMKD3MGP3W34R/3T298T8",
url = "http://urlib.net/ibi/8JMKD3MGP3W34R/3T298T8",
targetfile = "publicacao.pdf",
urlaccessdate = "25 abr. 2024"
}