[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
You seem to have javascript disabled. Please note that many of the page functionalities won't work as expected without javascript enabled.
 
 
Sign in to use this feature.

Years

Between: -

Subjects

remove_circle_outline

Journals

Article Types

Countries / Regions

Search Results (1)

Search Parameters:
Keywords = Betampona Nature Reserve (BNR)

Order results
Result details
Results per page
Select all
Export citation of selected articles as:
29 pages, 9419 KiB  
Article
Forest Conservation with Deep Learning: A Deeper Understanding of Human Geography around the Betampona Nature Reserve, Madagascar
by Gizelle Cota, Vasit Sagan, Maitiniyazi Maimaitijiang and Karen Freeman
Remote Sens. 2021, 13(17), 3495; https://doi.org/10.3390/rs13173495 - 3 Sep 2021
Cited by 3 | Viewed by 4271
Abstract
Documenting the impacts of climate change and human activities on tropical rainforests is imperative for protecting tropical biodiversity and for better implementation of REDD+ and UN Sustainable Development Goals. Recent advances in very high-resolution satellite sensor systems (i.e., WorldView-3), computing power, and machine [...] Read more.
Documenting the impacts of climate change and human activities on tropical rainforests is imperative for protecting tropical biodiversity and for better implementation of REDD+ and UN Sustainable Development Goals. Recent advances in very high-resolution satellite sensor systems (i.e., WorldView-3), computing power, and machine learning (ML) have provided improved mapping of fine-scale changes in the tropics. However, approaches so far focused on feature extraction or the extensive tuning of ML parameters, hindering the potential of ML in forest conservation mapping by not using textural information, which is found to be powerful for many applications. Additionally, the contribution of shortwave infrared (SWIR) bands in forest cover mapping is unknown. The objectives were to develop end-to-end mapping of the tropical forest using fully convolution neural networks (FCNNs) with WorldView-3 (WV-3) imagery and to evaluate human impact on the environment using the Betampona Nature Reserve (BNR) in Madagascar as the test site. FCNN (U-Net) using spatial/textural information was implemented and compared with feature-fed pixel-based methods including Support Vector Machine (SVM), Random Forest (RF), and Deep Neural Network (DNN). Results show that the FCNN model outperformed other models with an accuracy of 90.9%, while SVM, RF, and DNN provided accuracies of 88.6%, 84.8%, and 86.6%, respectively. When SWIR bands were excluded from the input data, FCNN provided superior performance over other methods with a 1.87% decrease in accuracy, while the accuracies of other models—SVM, RF, and DNN—decreased by 5.42%, 3.18%, and 8.55%, respectively. Spatial–temporal analysis showed a 0.7% increase in Evergreen Forest within the BNR and a 32% increase in tree cover within residential areas likely due to forest regeneration and conservation efforts. Other effects of conservation efforts are also discussed. Full article
(This article belongs to the Special Issue National REDD+ Monitoring and Reporting)
Show Figures

Graphical abstract

Graphical abstract
Full article ">Figure 1
<p>(<b>a</b>) WorldView-3 0.3 m pansharpened RGB imagery with minimal cloud cover, over the study area—the Betampona Nature Reserve (BNR) (white boundary polygon), located on the (<b>b</b>) eastern coast of Madagascar, which is (<b>c</b>) an island to the southeast of Africa.</p>
Full article ">Figure 2
<p>In situ images of a few classes being mapped, collected during the ground reference data collection in 2018. Additional classes include invasive guava, residential, and open water.</p>
Full article ">Figure 3
<p>U-net architecture with the encoding (<b>left</b>) and decoding (<b>right</b>) sections that produce the characteristic U-shape of the architecture, producing pixel-wise classification of the input imagery. ResNet layers forms the encoder.</p>
Full article ">Figure 4
<p>Network architecture implemented for the Deep Neural Network (DNN) model along with the number of neurons that were optimized for each hidden layer. The output layer contains 11 neurons, corresponding to the number of classes to be classified.</p>
Full article ">Figure 5
<p><span class="html-italic">F</span>1 score (Equation (6)), for all classes and models, showing the superiority of the FCNN U-Net model.</p>
Full article ">Figure 6
<p>Comparison of confusion matrixes for (<b>a</b>) DNN where SWIR bands were included, (<b>b</b>) DNN where SWIR bands were excluded, (<b>c</b>) U-Net where SWIR bands were included, and (<b>d</b>) U-Net where SWIR bands were excluded. A greater number of <span class="html-italic">FP</span>s and <span class="html-italic">FN</span>s are seen in (<b>b</b>,<b>d</b>) because of the removal of the SWIR bands. The DNN and U-Net models are chosen for comparison because of the highest and lowest reduction in accuracy due to the removal of SWIR.</p>
Full article ">Figure 7
<p><span class="html-italic">F</span>1 scores (Equation (6)) for the U-Net and DNN models including (16 bands) and excluding (8 bands) SWIR bands across all classes. The U-Net and DNN models trained on 16 bands show a higher <span class="html-italic">F</span>1 score compared to the models trained on eight bands.</p>
Full article ">Figure 8
<p>Detailed and highly accurate classification map of the BNR and surrounding areas, created using WorldView-3 imagery and post-classification editing, showing the distribution of 11 classes.</p>
Full article ">Figure 9
<p>(<b>a</b>) The 2010 classification map and (<b>b</b>) the associated classification map created in 2019. The use of identical classes enabled the quantification of land cover change over time.</p>
Full article ">Figure 10
<p>Residential areas in 2019 (red boundary polygon) in the study area (extent shown in inset map) overlain over the 2010 classification map showing the increase in residential area extent and the emergence of new hamlets in 2019 for (<b>a</b>) a location east of the BNR and (<b>b</b>) northwest of the BNR.</p>
Full article ">Figure 11
<p>An increase in tree cover in residential areas in hectares from 2010 (gray diagonal bar) to 2019 (black solid bar) is attributed to maturing native trees and successful agroforestry efforts.</p>
Full article ">Figure 12
<p>Percentage of classes in 2010 that were converted to Shrubland in 2019, within the study area based on the classification map extent seen in <a href="#remotesensing-13-03495-f009" class="html-fig">Figure 9</a>. This information is useful for awareness-raising and conservation efforts.</p>
Full article ">Figure 13
<p>Evergreen Forest areas in 2010 (green boundary polygons) that were converted to Shrubland in 2019 located east of the BNR, overlaid over the 2019 classification map. The inset map shows the 2019 classification map.</p>
Full article ">Figure 14
<p>Percentage of classes in 2010 that were converted to Mixed Forest in 2019 within the BNR, shows a higher conversion of invasive plants—Molucca Raspberry, Madagascar Cardamom, and Guava—to Mixed Forest compared to Evergreen Forest.</p>
Full article ">Figure 15
<p>Invasive plant species within the BNR and the land cover types seen 100 m within the BNR boundary (gray line) and within the ZOP (100 m extending from the BNR boundary). Forest covers are observed along with a few scattered agricultural fields within the 200 m wide region on two sides of the boundary. Note that wider streams represented by the Open Water class are located further south of this map extent, which is not visible here. The Open Water and Streams classes should be treated as the same class.</p>
Full article ">Figure 16
<p>Classification map created using the SVM model, with an 88.6% overall accuracy, zoomed into a location on the western section of the BNR that shows the salt-and-pepper effect.</p>
Full article ">Figure 17
<p>Reflectance spectra of all classes, showing the overlapping reflectance spectra for vegetation classes—for example between Evergreen Forest, Grassland, and Mixed Forest.</p>
Full article ">Figure 18
<p><span class="html-italic">F</span>1 scores of all classification methods employed including (16 bands) and excluding SWIR bands (8 bands), showing the percent change due to the removal of SWIR bands for (<b>a</b>) Guava, (<b>b</b>) Madagascar Cardamom, (<b>c</b>) Evergreen Forest, (<b>d</b>) Mixed Forest, (<b>e</b>) Molucca Raspberry, (<b>f</b>) Row Crops, (<b>g</b>) Residential, (<b>h</b>) Fallow, (<b>i</b>) Shrubland, and (<b>j</b>) Grassland.</p>
Full article ">Figure 19
<p>Percent change of (<b>a</b>) Evergreen Forest and (<b>b</b>) Agriculture—including Row Crops and Fallow—over the BNR overlain over a DEM, showing areas of increased (red) and decreased (blue) extents based on a 10 m × 10 m grid cell. The BNR boundary is shown via the black line and the ZOP and 100 m region within the BNR used for change analysis is displayed through the gray polygon.</p>
Full article ">
Back to TopTop