@MastersThesis{Chagas:2021:UsDaLi,
author = "Chagas, Gabriel Oliveira",
title = "Uso de dados LiDAR, SAR e Sentinel-2 para estimativa da biomassa
florestal acima do solo na Floresta Nacional do Tapaj{\'o}s",
school = "Instituto Nacional de Pesquisas Espaciais (INPE)",
year = "2021",
address = "S{\~a}o Jos{\'e} dos Campos",
month = "2021-08-17",
keywords = "sensoriamento remoto, Floresta Amaz{\^o}nica,
degrada{\c{c}}{\~a}o florestal, modelagem de AGB, random forest,
remote sensing, Amazon Rainforest, forest degradation, AGB
modeling.",
abstract = "A Floresta Amaz{\^o}nica, a maior {\'a}rea de floresta
cont{\'{\i}}nua do mundo, tem um papel importante na
regula{\c{c}}{\~a}o do clima local e regional, e na ciclagem da
{\'a}gua. Al{\'e}m disso, a Amaz{\^o}nia det{\'e}m um grande
estoque de carbono em sua {\'a}rea florestal, mas o
desflorestamento, a extra{\c{c}}{\~a}o seletiva da madeira, os
inc{\^e}ndios florestais e a fragmenta{\c{c}}{\~a}o florestal
t{\^e}m contribu{\'{\i}}do para a redu{\c{c}}{\~a}o desse
estoque na regi{\~a}o. Desta maneira, os dados de campo s{\~a}o
fundamentais para mensurar a magnitude e a extens{\~a}o desses
impactos na floresta e entender a din{\^a}mica das {\'a}reas
degradadas. Esses dados tamb{\'e}m permitem a gera{\c{c}}{\~a}o
de modelos preditivos de biomassa florestal acima do solo (AGB). A
AGB permite identificar {\'a}reas priorit{\'a}rias para a
atua{\c{c}}{\~a}o de projetos de conserva{\c{c}}{\~a}o
florestal, visto que a biomassa est{\'a} relacionada com os
estoques de carbono. Contudo, devido {\`a} extens{\~a}o da
Floresta Amaz{\^o}nica, a extrapola{\c{c}}{\~a}o dos dados de
campo {\'e} limitada pelo n{\'u}mero de amostras, pela {\'a}rea
das parcelas de campo e pelo tempo gasto na amostragem de campo.
Desse modo, o uso de dados de campo em conjunto com dados de
sensoriamento remoto {\'e} uma alternativa para melhorar as
estimativas de AGB. Todavia, devido {\`a} extens{\~a}o da
Floresta Amaz{\^o}nica, {\`a} variabilidade de sua estrutura e
aos diferentes graus de degrada{\c{c}}{\~a}o florestal, existem
v{\'a}rias incertezas nas estimativas de AGB dispon{\'{\i}}veis
na literatura. Assim, o presente trabalho teve como objetivo
estimar a AGB para a Floresta Nacional do Tapaj{\'o}s e
{\'a}reas adjacentes, considerando os diferentes graus de
interven{\c{c}}{\~a}o humana, utilizando dados dos sensores
LiDAR (GEDI e ALS), SAR (PALSAR-2) e {\'o}pticos (Sentinel-2).
Inicialmente, a AGB foi calculada com uma equa{\c{c}}{\~a}o
alom{\'e}trica, utilizando dados de campo, para compor a
vari{\'a}vel dependente do modelo. As vari{\'a}veis
independentes do modelo obtidas neste estudo foram: as
vari{\'a}veis biof{\'{\i}}sicas da vegeta{\c{c}}{\~a}o e as
m{\'e}tricas derivadas dos sensores remotos utilizados (altura do
dado GEDI e as m{\'e}tricas que descrevem a nuvem de pontos do
LiDAR ALS); as decomposi{\c{c}}{\~o}es polarim{\'e}tricas SAR;
os coeficientes de retroespalhamento, os {\'{\i}}ndices e
raz{\~o}es SAR dispon{\'{\i}}veis na literatura; os
{\'{\i}}ndices de vegeta{\c{c}}{\~a}o do Sentinel-2
dispon{\'{\i}}veis na literatura e as imagens fra{\c{c}}{\~a}o
do modelo linear de mistura espectral (MLME). A seguir o algoritmo
Random Forest foi usado para modelar a rela{\c{c}}{\~a}o entre
vari{\'a}veis preditivas e a AGB. Os melhores modelos de AGB, com
resolu{\c{c}}{\~o}es espaciais de 16,62 m, 50 m e 555,55 m,
apresentaram RMSE de 60,35 Mg ha-1 (23,07%), 75,88 Mg ha-1
(26,42%) e 89,87 Mg ha-1 (28,04%), e um Rē de 0,72, 0,66 e 0,57,
respectivamente. De acordo com a an{\'a}lise das estimativas de
AGB, os modelos {\'o}timos obtidos neste estudo foram capazes de
descrever a AGB de acordo com o grau de degrada{\c{c}}{\~a}o
florestal. Os resultados demonstram o potencial dos dados GEDI,
LiDAR ALS, SAR e {\'o}pticos para estimar a AGB em {\'a}reas de
floresta tropical de acordo com a variabilidade da estrutura
florestal. ABSTRACT: The world's largest continuous forest area,
the Amazon rainforest, plays an important role in regulating local
and regional climate and water cycling. In addition, the Amazon
rainforest has a large carbon stock in its forest area, but
deforestation, logging, forest fires, and forest fragmentation
have contributed to the reduction of its carbon stock. Thus, field
data are essential to measure the magnitude and extent of these
impacts on forests and to understand the dynamics of degraded
areas. These data also allow the generation of forest aboveground
biomass (AGB) predictive models. The AGB allows identifying
priority areas for the development of forest conservation projects
since biomass is related to carbon stocks. However, due to the
extension of the Amazon rainforest, the extrapolation of field
data is limited by the number of samples, the area of the sample
plots, and the time spent in field sampling. Thereby, the use of
field data together with remote sensing data is an alternative to
improve AGB estimates. Although, due to the extension, variability
of the Amazon rainforest structure, and different degrees of
forest degradation, there are several uncertainties in the AGB
estimates available in the literature. Thus, the present work
aimed to estimate AGB for the Tapaj{\'o}s National Forest and
adjacent areas, considering the different degrees of forest
degradation, using data from the sensors LiDAR (GEDI and ALS), SAR
(PALSAR-2), and optical (Sentinel-2). Initially, the AGB was
calculated with an allometric equation using field data to compose
the dependent variable of the model. The independent variables of
the model obtained in this study were: the vegetation biophysical
variables and the metrics derived from LiDAR sensors (height of
the GEDI data and the metrics that describe the LiDAR ALS point
cloud); polarimetric SAR decompositions; backscatter coefficients,
SAR indexes, and ratios available in the literature; Sentinel-2
vegetation indexes available in the literature and fraction images
from linear spectral mixing model. Then, the Random Forest
algorithm was used to model the relationship between predictive
variables and AGB. The best AGB estimates with spatial resolution
of 16.62m, 50m and 555.55m had RMSE of 60.35 Mg ha-1 (23.07%),
75.88 Mg ha-1 (26.42%) and 89 .87 Mg ha-1 (28.04%), and an Rē of
0.72, 0.66 and 0.57, respectively. According to the AGB estimates
analysis, the optimal models obtained in this study were able to
describe the AGB according to the degree of forest degradation.
The results demonstrate the potential of GEDI, LiDAR ALS, SAR, and
optical data to estimate AGB in tropical forest areas according to
the variability of the forest structure.",
committee = "Gama, F{\'a}bio Furlan (presidente) and Shimabukuro, Yosio Edemir
(orientador) and Sano, Edson Eyji",
englishtitle = "The use of LiDAR, SAR and Sentinel-2 data to estimate forest
aboveground biomass in Tapaj{\'o}s National Forest",
language = "pt",
pages = "103",
ibi = "8JMKD3MGP3W34T/4593M6E",
url = "http://urlib.net/ibi/8JMKD3MGP3W34T/4593M6E",
targetfile = "publicacao.pdf",
urlaccessdate = "02 maio 2024"
}