@InProceedings{WagnerDaStPhGlAr:2019:AnReDi,
author = "Wagner, Fabien Hubert and Dalagnol, Ricardo and Streher, Annia
Susin and Phillips, Oliver L. and Gloor, Emanuel Ulrich and
Arag{\~a}o, Luiz Eduardo Oliveira e Cruz de",
affiliation = "{Instituto Nacional de Pesquisas Espaciais (INPE)} and {Instituto
Nacional de Pesquisas Espaciais (INPE)} and {Universidade Estadual
Paulista (UNESP)} and {University of Leeds} and {University of
Leeds} and {Instituto Nacional de Pesquisas Espaciais (INPE)}",
title = "Analysing the Regional Distribution of a Key Canopy Palm Species
Using a Convolutional Network in an Amazon Forest",
year = "2019",
organization = "AGU Fall Meeting",
abstract = "Mapping plant species at landscape scale to provide information
for ecologists and forest managers is a new challenge for the
remote sensing community. Here, we use a deep learning algorithm
associated with very high-resolution multispectral images (0.5 m)
from GeoEye satellite to identify and segment a palm tree species,
Attalea speciosa, in the canopy of an Amazon forest. This study
was conducted in a region of the critically endangered Brazilian
Amazon Rainforest, between two deforestation fronts, which is a
global conservation priority due to its abundance of species of
flora and fauna and its carbon stock. The convolutional network
generated in this study for identifying palm trees was trained
with about 1024 high-resolution true colour optical images and
their labelled masks. Additionally, we analysed the spatial
distribution of the palm trees at the regional scale based on
patches locations and edaphic conditions. Our deep learning
network segmented palm trees patches with overall accuracies of
95.5 % and Dice coefficients of 0.67. Then, the segmentation of
tree species was produced over a region >2500 kmē using GeoEye
Red, Green and Blue bands pan-sharpened at 0.5 m. We found that
the palm trees covered 5 % of the natural forest canopies and were
distributed in more than one million patches. Our results based on
the palm trees distribution shown that their abundance tends to
vary primarily with local soil water content over the landscape.
Overall, their distribution over the region seems to indicate a
relatively pristine landscape. However, we observed that they are
sparsely distributed in secondary forests and could likely be used
as an indicator of large past perturbation. Our work shows how
deep learning algorithm can support applications such as mapping
plant species to understand plant distributions and landscape
features.",
conference-location = "San Francisco, CA",
conference-year = "09-13 dec.",
language = "en",
targetfile = "wagner_analysis.pdf",
urlaccessdate = "01 maio 2024"
}