@PhDThesis{Bendini:2019:AgLaCl,
author = "Bendini, Hugo do Nascimento",
title = "Agricultural land classification based on phenological information
from dense time-series Landsat-like images in the brazilian
Cerrado",
school = "Instituto Nacional de Pesquisas Espaciais (INPE)",
year = "2019",
address = "S{\~a}o Jos{\'e} dos Campos",
month = "2018-07-13",
keywords = "Big-data, time-series analysis, agricultural land use
classification, multi-sensor, remote sensing, big data,
an{\'a}lise de s{\'e}ries temporais, classifica{\c{c}}{\~a}o
de uso agr{\'{\i}}cola, multi-sensor, sensoriamento remoto.",
abstract = "Brazil has an important role in the world in terms of food
production and the largest native forest, providing essential
environmental services for the planet and humanity. However, this
highlights the challenge of creating an economic development model
that takes into account the environmental conservation. Brazil has
already demonstrated successful experiences in Amazon
deforestation reduction, but other biomes of great environmental
importance, such as the Cerrado, has been under great pressure of
agricultural expansion. Satellite image time series can be used to
derive phenological information of vegetation, and considering the
high heterogeneity of crop types and their respective planting
calendars in Brazil, is essential for crop classification and
monitoring. Our hypothesis in this thesis is that phenological
information can be extracted from Landsatlike dense image time
series, allowing the development of a method for agriculture
mapping with more detail. We tested the integration of different
satellite, such as Landsat-8, Landsat-7 and CBERS-4, combined with
different smoothing techniques, to generate EVI (Enhanced
Vegetation Index) image time series at high frequency in order to
extract the phenological metrics. A hierarchical classification
approach using the Random Forest algorithm was developed to
produce detailed agricultural maps. The classification results are
promising (higher than 80% of overall accuracy) and showed the
feasibility of applying the method on a large scale and over a
longer period of time for the Cerrado biome. In addition, the
phenological information obtained by the method showed a potential
to be used in the understanding of different agricultural
practices adopted by farmers in property level. RESUMO: O Brasil
tem um papel importante no mundo em termos de produ{\c{c}}{\~a}o
de alimentos e a maior floresta nativa, fornecendo servi{\c{c}}os
ambientais essenciais para o planeta e para a humanidade. No
entanto, isso destaca o desafio de criar um modelo de
desenvolvimento econ{\^o}mico que leve em
considera{\c{c}}{\~a}o a conserva{\c{c}}{\~a}o ambiental. O
Brasil j{\'a} demonstrou experi{\^e}ncias bem-sucedidas na
redu{\c{c}}{\~a}o do desmatamento da Amaz{\^o}nia, mas outros
biomas de grande import{\^a}ncia ambiental, como o Cerrado,
est{\~a}o sob grande press{\~a}o de expans{\~a}o
agr{\'{\i}}cola. S{\'e}ries temporais de imagens de
sat{\'e}lite podem ser usadas para derivar
informa{\c{c}}{\~o}es fenol{\'o}gicas da vegeta{\c{c}}{\~a}o.
Considerando a diversidade de culturas agr{\'{\i}}colas e seus
respectivos calend{\'a}rios de plantio no Brasil, essas
informa{\c{c}}{\~o}es s{\~a}o essenciais para a
classifica{\c{c}}{\~a}o e monitoramento agr{\'{\i}}cola. Nossa
hip{\'o}tese {\'e} que informa{\c{c}}{\~o}es fenol{\'o}gicas
podem ser extra{\'{\i}}das de s{\'e}ries temporais de imagens
de resolu{\c{c}}{\~a}o espacial Landsat-like, permitindo o
desenvolvimento de m{\'e}todo para mapeamento detalhado da
agricultura. Testamos a integra{\c{c}}{\~a}o de diferentes
sat{\'e}lites, como Landsat-8, Landsat-7 e CBERS-4, combinados
com diferentes t{\'e}cnicas de suaviza{\c{c}}{\~a}o para gerar
s{\'e}ries temporais de imagem EVI (Enhanced Vegetation Index) em
alta frequ{\^e}ncia e extrair as m{\'e}tricas fenol{\'o}gicas.
Uma abordagem de classifica{\c{c}}{\~a}o hier{\'a}rquica usando
o algoritmo Random Forest foi aplicada para produzir os mapas. Os
resultados da classifica{\c{c}}{\~a}o s{\~a}o promissores
(acima de 80% da acur{\'a}cia) e mostraram a viabilidade de
aplicar o m{\'e}todo em larga escala e por um longo
per{\'{\i}}odo para o Bioma Cerrado. Al{\'e}m disso, as
informa{\c{c}}{\~o}es fenol{\'o}gicas mostraram potencial para
serem utilizadas na compreens{\~a}o de diferentes pr{\'a}ticas
agr{\'{\i}}colas adotadas pelos agricultores no Cerrado, em
escala de propriedade.",
committee = "Sanches, Ieda Del' Arco (presidente) and Fonseca, Leila Maria
Garcia (orientadora) and K{\"o}rting, Thales Sehn (orientador)
and Camargo Neto, Jo{\~a}o and Feitosa, Raul Queiroz and Silva,
Silvia Helena Modenese Gorla da",
englishtitle = "Classifica{\c{c}}{\~a}o de {\'a}reas agr{\'{\i}}colas baseada
em informa{\c{c}}{\~o}es fenol{\'o}gicas de s{\'e}ries
temporais de imagens Landsat-like no Cerrado brasileiro",
language = "en",
pages = "122",
ibi = "8JMKD3MGP3W34R/3RJS628",
url = "http://urlib.net/ibi/8JMKD3MGP3W34R/3RJS628",
targetfile = "publicacao.pdf",
urlaccessdate = "03 jun. 2024"
}