@MastersThesis{Oliveira:2024:MaBuAr,
author = "Oliveira, Alisson Cleiton de",
title = "Mapping burned areas in the Cerrado using time series from the
CBERS and Amazonia satellites",
school = "Instituto Nacional de Pesquisas Espaciais (INPE)",
year = "2024",
address = "S{\~a}o Jos{\'e} dos Campos",
month = "2024-02-09",
keywords = "fire, WFI, Chapada dos Veadeiros National Park, supervised
classification, random forest, fogo, WFI, Parque Nacional da
Chapada dos Veadeiros, classifica{\c{c}}{\~a}o supervisionada,
random forest.",
abstract = "The Brazilian Cerrado, a hotspot for global biodiversity
conservation, evolved under the presence of natural wildfires.
Fire has become frequent and widespread, and the Cerrado, where
natural fires have occurred for at least four million years, is
threatened by human-induced wildfires. The Chapada dos Veadeiros
National Park (CVNP), located in the state of Goi{\'a}s, Brazil,
was established in 1961 and currently covers 240,611 ha. In 2017,
approximately 66,000 ha were burned in the CVNP, and the
Integrated Fire Management (IFM) was implemented still in that
year to reduce the negative impacts of future criminal/accidental
events. Remote Sensing (RS) data show that there were fire-foci in
the CVNP during the dry months of 2020, 2021, and 2022. There are
two RS-based products for wildfires detection: products of
released heat and products of biophysical changes in vegetation.
As an example of provider, there is the Queimadas Program of
Brazils National Institute for Space Research (INPE), which
provides products on daily fire hotspots and a monthly product of
burned areas for the Cerrado. As of the current date, there are no
products that employ Brazilian satellite images for the systematic
mapping of burned areas. The objective of this research is to
explore methods for supervised classification of time series
images captured by the Wide Field Imager (WFI) sensor on board the
CBERS-4, CBERS-4A, and AMAZONIA-1 satellites, using the Random
Forest (RF) algorithm. The study area is the CVNP and its buffer
zone of 10 km, and the time window covers the years 2020, 2021 and
2022. A total of 382 images were acquired from INPE archive and
after filtering for cloud cover it was decided to keep 235 images:
50 from 2020, 72 from 2021 and 113 from 2022. The WFI sensor has
four spectral bands (BGR NIR), which is a limiting factor.
Consequently, we estimated and integrated the BAI (Burned Area
Index), EVI (Enhanced Vegetation Index), GEMI (Global
Environmental Monitoring Index), NDVI (Normalized Difference
Vegetation Index), and NDWI (Normalized Difference Water Index)
spectral indices into a regular grid with 500 m x 500 m cells,
totalling 38,957 cells. For each one of the previous spectral
indices more the NIR band, datasets containing annual and
semi-annual observations were structured and the models were
trained using samples of burned areas and unburned areas
previously collected through visual image analysis. The annual
models achieved at least 90% accuracy and the best generalization
results were observed using multi-temporal datasets. The results
of this research indicate that, given a representative sample set,
it is possible to detect burned areas in the CVNP using WFI
imagery. RESUMO: O Cerrado, hotspot de conserva{\c{c}}{\~a}o da
biodiversidade, evoluiu sob a presen{\c{c}}a de inc{\^e}ndios
naturais. O fogo se tornou frequente e difuso na atualidade, e o
Cerrado, onde o fogo natural ocorre h{\'a} pelo menos quatro
milh{\~o}es de anos, {\'e} amea{\c{c}}ado por inc{\^e}ndios de
origens antr{\'o}picas. O Parque Nacional da Chapada dos
Veadeiros (PNCV), localizado no estado de Goi{\'a}s, foi
institu{\'{\i}}do em 1961 e a sua {\'a}rea atual {\'e} de
240.611 ha. Em 2017 o PNCV teve cerca de 66 mil ha atingidos por
fogo e, ainda nesse ano, foi institu{\'{\i}}do o Manejo
Integrado do Fogo (MIF) a fim de se reduzir os impactos negativos
desses eventos criminosos/acidentais. Dados de Sensoriamento
Remoto (SR) evidenciam que houve focos de calor nos meses secos de
2020, 2021 e 2022 no PNCV. Existem, principalmente, dois
subprodutos de fogo em aplica{\c{c}}{\~o}es de SR: subprodutos
de libera{\c{c}}{\~a}o de calor e subprodutos de
modifica{\c{c}}{\~o}es biof{\'{\i}}sicas da
vegeta{\c{c}}{\~a}o. Como exemplo de provedores, tem-se o
Programa Queimadas do Instituto Nacional de Pesquisas Espaciais
(INPE), que disponibiliza produtos sobre focos de calor
di{\'a}rios e um produto mensal de {\'a}reas queimadas para o
Cerrado. Entretanto n{\~a}o h{\'a}, at{\'e} o momento, produtos
que utilizem imagens de sat{\'e}lites brasileiros no mapeamento
sistem{\'a}tico de {\'a}reas queimadas. Assim, essa pesquisa tem
como objetivo explorar abordagens de classifica{\c{c}}{\~a}o
supervisionada de s{\'e}ries temporais de imagens do sensor Wide
Field Imager (WFI), a bordo dos sat{\'e}lites brasileiros
CBERS-4, CBERS-4A e AMAZONIA-1, com o algoritmo Random Forest
(RF). A {\'a}rea de estudo {\'e} o PNCV e seu buffer envolvente
de 10 km, e a janela temporal engloba os anos de 2020, 2021 e
2022. Ao todo, 382 imagens foram adquiridas do arquivo do INPE e,
ap{\'o}s a triagem por cobertura de nuvens, optou-se por manter
235 imagens, sendo 50 de 2020, 72 de 2021 e 113 de 2022. O sensor
WFI possui quatro bandas (BGR NIR), o que {\'e} um limitante.
Procedeu-se, portanto, a estima{\c{c}}{\~a}o e a
integra{\c{c}}{\~a}o dos {\'{\i}}ndices BAI ({\'{\I}}ndice
de {\'A}rea Queimada), EVI ({\'{\I}}ndice de
Vegeta{\c{c}}{\~a}o Melhorado), GEMI ({\'{\I}}ndice Global de
Monitoramento Ambiental), NDVI ({\'{\I}}ndice de
Vegeta{\c{c}}{\~a}o por Diferen{\c{c}}a Normalizada) e NDWI
({\'{\I}}ndice de {\'A}gua por Diferen{\c{c}}a Normalizada) em
uma grade regular com c{\'e}lulas de 500 m x 500 m, o que
totaliza 38.957 c{\'e}lulas. Para cada {\'{\i}}ndice espectral
e para a banda do NIR foram estruturados datasets contendo
observa{\c{c}}{\~o}es anuais e semestrais e os treinamentos dos
modelos foram conduzidos com amostras de {\'a}reas queimadas e
{\'a}reas n{\~a}o queimadas, coletadas previamente por
an{\'a}lise visual de imagens. Como resultado, os modelos anuais
atingiram, no m{\'{\i}}nimo, 90% de acur{\'a}cia e os melhores
resultados de generaliza{\c{c}}{\~a}o foram observados
utilizando datasets multitemporais. Os resultados desta pesquisa
indicam que {\'e} poss{\'{\i}}vel, dado um conjunto de amostras
representativo, classificar {\'a}reas queimadas do Cerrado do
PNCV utilizando imagens do sensor WFI.",
committee = "Galv{\~a}o, L{\^e}nio Soares (presidente) and K{\"o}rting,
Thales Sehn (orientador) and Mataveli, Guilherme Augusto Verola
and Kuck, Tahisa Neitzel",
englishtitle = "Mapeamento de {\'a}reas queimadas no Cerrado utilizando
s{\'e}ries temporais dos sat{\'e}lites CBERS e Amazonia",
language = "en",
pages = "80",
ibi = "8JMKD3MGP3W34T/4APPK92",
url = "http://urlib.net/ibi/8JMKD3MGP3W34T/4APPK92",
targetfile = "publicacao.pdf",
urlaccessdate = "11 maio 2024"
}