@MastersThesis{Soares:2006:ÁrAgSe,
author = "Soares, D{\^e}nis de Moura",
title = "{\'A}reas agr{\'{\i}}colas em sensores com
resolu{\c{c}}{\~a}o espacial de 30m estimadas a partir de dados
originais e simulados MODIS e m{\'e}tricas de paisagem",
school = "Instituto Nacional de Pesquisas Espaciais (INPE)",
year = "2006",
address = "S{\~a}o Jos{\'e} dos Campos",
month = "2006-05-30",
keywords = "sensoriamento remoto, resolu{\c{c}}{\~a}o moderna, padr{\~a}o
espacial, m{\'e}trica da paisagem, an{\'a}lise de
regress{\~a}o, filtragem de textura, filtragem de maioria,
escala, MODIS/TERRA, agricultura, milho,
cana-de-a{\c{c}}{\'u}car, soja, Ipu{\~a} (SP), Guar{\'a} (SP),
S{\~a}o Joaquim da Barra (SP), remote sensing, coarse resolution,
spatial pattern, Landscape metric, regression analysis, texture
filtering, majority filtering, scale, MODIS/TERRA, agriculture,
corn, sugarcane, soybean, Ipu{\~a} (S{\~a}o Paulo - Brazil),
Guar{\'a} (S{\~a}o Paulo - Brazil), S{\~a}o Joaquim da Barra
(S{\~a}o Paulo - Brazil).",
abstract = "O agroneg{\'o}cio tem papel de destaque na economia brasileira.
Dessa forma, a cria{\c{c}}{\~a}o de metodologias para o
monitoramento agr{\'{\i}}cola {\'e} fundamental. Nesta linha de
racioc{\'{\i}}nio, a estimativa de {\'a}rea agr{\'{\i}}cola
{\'e} uma atividade importante para a previs{\~a}o de safras e
avalia{\c{c}}{\~a}o da disponibilidade de produtos para
abastecimento interno e exporta{\c{c}}{\~o}es. A
utiliza{\c{c}}{\~a}o de sensores remotos tem se mostrado
eficiente na medi{\c{c}}{\~a}o de {\'a}reas. No entanto, a
abund{\^a}ncia de nuvens representa um fator cr{\'{\i}}tico
para o sucesso de sua aplica{\c{c}}{\~a}o. Sensores com
repetitividade quase di{\'a}ria podem ser uma solu{\c{c}}{\~a}o
para essa limita{\c{c}}{\~a}o, apesar de sua
resolu{\c{c}}{\~a}o espacial moderada. Assim, este trabalho teve
por objetivo avaliar as diferen{\c{c}}as obtidas nas estimativas
de {\'a}reas de culturas agr{\'{\i}}colas quando s{\~a}o
utilizados sensores de resolu{\c{c}}{\~a}o espacial moderada
(p.ex, MODIS/Terra, com 250m), ao inv{\'e}s de
resolu{\c{c}}{\~a}o espacial fina (ETM+/Landsat-7, com 30m),
considerando dados originais e simulados, diferentes culturas
agr{\'{\i}}colas e seu padr{\~a}o de distribui{\c{c}}{\~a}o
espacial (m{\'e}tricas da paisagem). As culturas
agr{\'{\i}}colas avaliadas foram o milho, a
cana-de-a{\c{c}}{\'u}car e a soja, na regi{\~a}o de Ipu{\~a},
Guar{\'a} e S{\~a}o Joaquim da Barra, no norte paulista.
Tamb{\'e}m foram inclu{\'{\i}}das na an{\'a}lise as classes
tem{\'a}ticas mata, pastagem e solo exposto. Para atingir tal
objetivo foi estudada a evolu{\c{c}}{\~a}o dos valores das
m{\'e}tricas de paisagem em fun{\c{c}}{\~a}o de
degrada{\c{c}}{\~o}es sucessivas da imagem ETM+ para obter
imagens de 90m, 150m, 210m e 270m, utilizando filtragem espacial
de textura e de maioria. Modelos de regress{\~a}o simples
({\'a}rea) e m{\'u}ltipla ({\'a}rea e m{\'e}tricas) foram
elaborados com base em dados originais dos sensores ETM+ e MODIS,
considerando todas as classes em conjunto (abordagem geral) e cada
classe individualmente (abordagem espec{\'{\i}}fica). Os
resultados obtidos mostraram que: a) as duas t{\'e}cnicas de
simula{\c{c}}{\~a}o afetaram de maneira semelhante o padr{\~a}o
espacial das classes tem{\'a}ticas, sendo a filtragem de textura
mais real{\'{\i}}stica na tarefa de representar o sensor
MODIS/Terra; b) na abordagem geral para estimativa de {\'a}rea, a
regress{\~a}o simples entre as {\'a}reas de classes
tem{\'a}ticas obtidas das imagens originais apresentou
coeficiente de determina{\c{c}}{\~a}o (Rē) de 0,46 e a
inclus{\~a}o dos {\'{\i}}ndices de padr{\~a}o espacial no
modelo de regress{\~a}o m{\'u}ltipla elevou tal grandeza para
0,49, sendo as m{\'e}tricas {\'A}rea, LPI, LSI, TCA PLADJ e IJI
as constituintes do modelo; c) na abordagem espec{\'{\i}}fica, a
cria{\c{c}}{\~a}o de modelos estat{\'{\i}}sticos para cada
cultura agr{\'{\i}}cola elevou bastante o Rē, atingindo valores
de 0,52, 0,67 e 0,87, para o milho, a cana e a soja,
respectivamente. O modelo para o milho foi composto pelos
{\'{\i}}ndices LSI, CLUMPY, IJI, MESH e NP, para a cana pelos
{\'A}rea, LSI, CLUMPY, IJI e NLSI. Por fim, para a soja as
m{\'e}tricas relevantes foram: {\'A}rea, NP, NDCA, DCAD, COHE.
Portanto, os resultados demonstram que sensores de
resolu{\c{c}}{\~a}o espacial moderada podem ser utilizados para
predizer {\'a}reas vistas por sensores de resolu{\c{c}}{\~a}o
mais fina, especialmente para as culturas agr{\'{\i}}colas menos
fragmentadas (soja e cana) e com a ado{\c{c}}{\~a}o de modelos
estat{\'{\i}}sticos espec{\'{\i}}ficos que incorporem
m{\'e}tricas de paisagem. ABSTRACT: The national agribusiness is
very important for the Brazilian economy. Thus, the creation of
methodologies to monitor the activity is fundamental. In this way,
the estimation of crop area is an important tool for prediction of
the availability of products for national consumption and
exportation. Remote sensors have frequently been used to measure
areas, but clouds abundance, mainly in agricultural seasons,
represents a huge difficulty. Sensors with almost daily revisit
time can be a solution for that limitation, but their spatial
resolution is usually poor. Therefore, the aim of this work was to
evaluate the differences between crop area estimation from coarse
resolution data (e.g. MODIS/Terra, with 250m) and fine resolution
data (ETM+/Landsat-7, with 30m), using real and simulated data,
different crop types and their spatial pattern (landscape
metrics). The analysis was applied for three different crops:
corn, sugarcane and soybean. The study area was located close to
the Ipu{\~a}, Guar{\'a} and S{\~a}o Joaquim da Barra, cities in
the north of the S{\~a}o Paulo state, Brazil. The thematic
classes woodland, pasture and exposed soil were also included in
the analysis. To reach the goal, the behavior of the the landscape
metrics was studied as a function of the simulation of different
levels of spatial resolution (90m, 150m, 210m and 270m) from ETM+
data using texture and majority filtering. Simple (area) and
multiple (area plus landscape metrics) regression models were
constructed using original data of the sensors ETM+ and MODIS,
considering all classes together (general approach) and each class
individually (specific approach). The results showed that: a) both
simulation techniques affected similarly the spatial pattern of
the thematic classes, but the texture filtering was more realistic
to represent the MODIS/Terra sensor; b) in the general approach to
estimate crop area, the simple regression between class areas from
real data presented low coefficient of determination (Rē of 0.46).
By adding the landscape metrics, this coefficient increased to
0.49. The selected indices for this procedure were {\'A}rea, LPI,
LSI, TCA PLADJ and IJI; c) in the specific approach, the creation
of statistic models for each agricultural class increased the Rē
to 0.52, 0.67 and 0.87 for corn, sugarcane and soybean,
respectively. The model for corn was composed of the metrics LSI,
CLUMPY, IJI, MESH and NP. For sugarcane, {\'A}rea, LSI, CLUMPY,
IJI and NLSI were selected as the most important metrics. Finally,
the relevant landscape metrics for soybean were {\'A}rea, NP,
NDCA, DCAD and COHE. Thus, the results demonstrated that coarse
spatial resolution data can be utilized to predict crop area
measured from fine spatial resolution data, especially for less
fragmented crops (soybean and sugarcane) and with the use of
specific crop regression models incorporating landscape metrics.",
committee = "Epiphanio, Jos{\'e} Carlos Neves (presidente) and Formaggio,
Antonio Roberto (orientador) and Galv{\~a}o, L{\^e}nio Soares
(orientador) and Shimabukuro, Yosio Edemir and Couto, Eduardo
Guimar{\~a}es",
copyholder = "SID/SCD",
englishtitle = "Crop area in spatial resolution of 30m estimated with original and
simulated MODIS data and landscape metrics",
language = "pt",
pages = "153",
ibi = "6qtX3pFwXQZGivnJSY/M3jzT",
url = "http://urlib.net/ibi/6qtX3pFwXQZGivnJSY/M3jzT",
targetfile = "publicacao.pdf",
urlaccessdate = "11 maio 2024"
}