@MastersThesis{Fragal:2015:ReHiMu,
author = "Fragal, Everton Hafemann",
title = "Reconstru{\c{c}}{\~a}o hist{\'o}rica de mudan{\c{c}}as na
cobertura florestal em v{\'a}rzeas do Baixo Amazonas utilizando o
algoritmo LandTrendr",
school = "Instituto Nacional de Pesquisas Espaciais (INPE)",
year = "2015",
address = "S{\~a}o Jos{\'e} dos Campos",
month = "2015-04-09",
keywords = "Amaz{\^o}nia, floresta inund{\'a}vel, altera{\c{c}}{\~a}o da
floresta, monitoramento, Amazon, wetland forests, forest change,
monitoring.",
abstract = "As florestas de v{\'a}rzea s{\~a}o importantes para a
manuten{\c{c}}{\~a}o da biodiversidade e para o provimento de
servi{\c{c}}os ecossist{\^e}micos. Entretanto, atividades
antr{\'o}picas t{\^e}m levado {\`a} redu{\c{c}}{\~a}o da
cobertura de florestas de v{\'a}rzea no decorrer do tempo. A
perda florestal impacta tanto a popula{\c{c}}{\~a}o humana local
e regional, quanto o ecossistema de v{\'a}rzea, enquanto o
desenvolvimento de nova cobertura florestal promove um novo ciclo
de servi{\c{c}}os ecossist{\^e}micos fornecidos pelas florestas.
Diversos trabalhos buscaram quantificar a perda florestal e
identificar seu agente causador a partir de s{\'e}ries temporais
de imagens de sat{\'e}lite. No entanto, abordagens de mapeamento
manuais limitam o n{\'u}mero de imagens que podem ser avaliadas.
Algoritmos semi-autom{\'a}ticos apresentam-se como alternativa
{\`a} an{\'a}lise manual, maximizando a quantidade de
informa{\c{c}}{\~o}es sobre mudan{\c{c}}as na cobertura
florestal. Nesta pesquisa foi avaliada aplicabilidade do algoritmo
\emph{Landsat-based Detection of Trends in Disturbance and
Recovery} (LandTrendr) para reconstru{\c{c}}{\~a}o
hist{\'o}rica das mudan{\c{c}}as na cobertura florestal de
v{\'a}rzea em um trecho do Baixo Amazonas, no per{\'{\i}}odo de
1984 a 2009. Para tal, foram definidos os seguintes objetivos
espec{\'{\i}}ficos: 1) Avaliar qual informa{\c{c}}{\~a}o
espectral {\'e} mais eficiente para detectar mudan{\c{c}}as na
cobertura florestal; 2) Examinar o conjunto {\'o}timo de
par{\^a}metros do algoritmo LandTrendr para ajuste de
trajet{\'o}rias espectro-temporais em florestas de v{\'a}rzea;
3) Avaliar a confiabilidade dos atributos gerados pelo algoritmo
para caracterizar mudan{\c{c}}as da cobertura florestal; e 4)
Analisar a exatid{\~a}o na discrimina{\c{c}}{\~a}o entre
agentes antr{\'o}picos e naturais causadores de mudan{\c{c}}as
na cobertura florestal, a partir dos atributos providos pelo
algoritmo. Foi utilizada uma s{\'e}rie temporal de 37 imagens
Landsat TM e EMT+, adquirida entre setembro e novembro para o
per{\'{\i}}odo de 1984 a 2009. O {\'{\i}}ndice de
vegeta{\c{c}}{\~a}o NDVI mostrou-se mais eficiente para detectar
mudan{\c{c}}as na cobertura florestal, mas 37\% da perda e 31\%
do desenvolvimento da cobertura florestal na {\'a}rea estudada
n{\~a}o foi detectada pelo algoritmo. Os valores {\'o}timos dos
par{\^a}metros foram kernel size=3x3; pval=0,05; e max
segments=6, maximizando a detec{\c{c}}{\~a}o dos eventos de
mudan{\c{c}}a e minimizando falsos eventos. As trajet{\'o}rias
espectro-temporais refletiram eventos ocorridos na cobertura
florestal, e o n{\'{\i}}vel de confiabilidade dos atributos que
caracterizam a perda e desenvolvimento da cobertura florestal foi
mais alto ao longo do rio Amazonas, em rela{\c{c}}{\~a}o ao
interior da v{\'a}rzea. Estimou-se uma maior incid{\^e}ncia de
perdas de origem antr{\'o}pica (1.071 ha) do que de origem
natural (884 ha), com Exatid{\~a}o Global M{\'e}dia de 94\%.
Contudo, houve dificuldade na discrimina{\c{c}}{\~a}o entre
causas naturais e antr{\'o}picas de perda florestal para o
{\'u}ltimo ano da s{\'e}rie temporal. Conclui-se que algoritmo
LandTrendr foi eficiente na detec{\c{c}}{\~a}o e
caracteriza{\c{c}}{\~a}o dos eventos de perda e desenvolvimento
da cobertura florestal, especialmente em {\'a}reas ao longo do
canal do rio Amazonas, podendo ser {\'u}til para avaliar
espa{\c{c}}o-temporalmente a ocorr{\^e}ncia de eventos de
mudan{\c{c}}a ao longo de toda a calha do rio Amazonas. ABSTRACT:
Floodplain forests are important for maintaining biodiversity and
providing ecosystem services. However, anthropogenic activities
have brought a reduction of floodplain forest cover over time.
Forest loss impacts local and regional human populations, as well
as the floodplain ecosystem, while forest cover growth promotes a
new cycle of ecosystem services provision by forests. Several
studies have attempted to quantify forest loss and its agents of
causation, based on time series of satellite images. However,
manual mapping approaches limit the number of images that can be
assessed. Semi-automatic algorithms can be considered as an
alternative to manual analysis, maximizing the amount of
information that can be obtained on forest cover change. We
investigated the applicability of the Landsat-based Detection of
Trends in Disturbance and Recovery (LandTrendr) algorithm for
historical reconstruction of changes in floodplain forest cover,
in a portion of the Lower Amazon River floodplain, from 1984 to
2009. We defined the following specific objectives: 1) Evaluate
which spectral information is more efficient to detect changes in
forest cover; 2) Examine the optimal set of LandTrendr parameters
for fitting spectral-temporal trajectories in v{\'a}rzea forests;
3) Evaluate how reliable are the attributes generated by the
algorithm to characterize changes in forest cover; 4) Evaluate the
attainable accuracy for the discrimination between natural and
anthropogenic causes of change in forest cover, based on the
attributes provided by the algorithm. A time series of 37 Landsat
TM and ETM+ images were acquired between September and November
for the period extending from 1984 to 2009. NDVI was the most
efficient spectral information to detect changes in forest cover,
but 37\% of mapped forest loss and 31\% of mapped forest growth
in the study area were not identified by the algorithm. The
optimal set of parameters were kernel size=3x3; pval=0,05; and max
segments=6, which maximized the detection of change events and
minimized false events. The spectral-temporal trajectories
reflected actual events in forest cover, and the reliability level
of attributes characterizing the loss and growth of forest cover
was highest along the Amazon River margins, when compared to the
floodplain interior. We estimated a higher incidence of forest
loss with anthropogenic origin (1,071 ha) versus natural origins
(884 ha), with an Average Global Accuracy of 94\%. However, it
was difficult to discriminate between natural and anthropogenic
causes of forest loss for the latter years of the time series. We
conclude that the LandTrendr algorithm was efficient in detecting
and characterizing forest cover loss and growth events, especially
in areas along the Amazon River margins. The algorithm can
therefore be applied to evaluate spatial and temporal forest
change events along the entire Amazon River floodplain.",
committee = "Novo, Evlyn M{\'a}rcia Le{\~a}o de Moraes
(presidente/orientadora) and Silva, Thiago Sanna Freire
(orientador) and Santos, Jo{\~a}o Roberto dos and Arag{\~a}o,
Luiz Eduardo Oliveira e Cruz de and Lima, Andr{\'e} de and
Sch{\"o}gart, Jochen",
copyholder = "SID/SCD",
englishtitle = "Historical reconstruction of forest cover changes in v{\'a}rzeas
of Lower Amazon using the algorithm LandTrendr",
language = "pt",
pages = "124",
ibi = "8JMKD3MGP3W34P/3J83FGH",
url = "http://urlib.net/ibi/8JMKD3MGP3W34P/3J83FGH",
targetfile = "publicacao.pdf",
urlaccessdate = "27 abr. 2024"
}