Publicação

Assessment of carbon sequestration in forest areas using deep learning

Ver documento

Detalhes bibliográficos
Resumo:Growing awareness of environmental impacts is making it more important than ever to explore regions with dense vegetation. Remote monitoring is a viable solution for the surveillance of large areas, such as forests. Based in intelligent systems, this work aims to develop a methodology for assessing carbon sequestration in forest areas. Deep learning (DL) structures were used to predict the heights and stand densities in tree colonies. Light Detection and Ranging (LiDAR) sensor scans obtained by Unmanned Aerial Vehicle (UAV) overflight were processed to extract elevation values and images. Point clouds were processed using QGIS software. The LAStools extension was employed to manipulate Digital Elevation Model (DEM) and rasters, obtaining relevant information. This data was then used to create a dataset for implementation in Convolutional Neural Network (CNN) models. Specific biometric relationships were implemented to estimate additional data such as Above Ground Biomass (AGB) and phytovolume. After evaluating different architectures, the VGG19 CNN model was highlighted as the most promising. An area of 46.6 hectares was covered, with an estimated total value of 4225.81 tons of carbon. This value provided an accuracy of 91%, based on forest inventories carried out in the same region. The study was conducted in the northern region of mainland Portugal, encompassing two distinct Pinus pinaster Ait. forests.
Autores principais:Britto, Raphael Duarte
Assunto:Carbon sequestration Forest areas Convolutional neural network RGB images QGIS software
Ano:2025
País:Portugal
Tipo de documento:dissertação de mestrado
Tipo de acesso:acesso aberto
Instituição associada:Instituto Politécnico de Bragança
Idioma:inglês
Origem:Biblioteca Digital do IPB
Descrição
Resumo:Growing awareness of environmental impacts is making it more important than ever to explore regions with dense vegetation. Remote monitoring is a viable solution for the surveillance of large areas, such as forests. Based in intelligent systems, this work aims to develop a methodology for assessing carbon sequestration in forest areas. Deep learning (DL) structures were used to predict the heights and stand densities in tree colonies. Light Detection and Ranging (LiDAR) sensor scans obtained by Unmanned Aerial Vehicle (UAV) overflight were processed to extract elevation values and images. Point clouds were processed using QGIS software. The LAStools extension was employed to manipulate Digital Elevation Model (DEM) and rasters, obtaining relevant information. This data was then used to create a dataset for implementation in Convolutional Neural Network (CNN) models. Specific biometric relationships were implemented to estimate additional data such as Above Ground Biomass (AGB) and phytovolume. After evaluating different architectures, the VGG19 CNN model was highlighted as the most promising. An area of 46.6 hectares was covered, with an estimated total value of 4225.81 tons of carbon. This value provided an accuracy of 91%, based on forest inventories carried out in the same region. The study was conducted in the northern region of mainland Portugal, encompassing two distinct Pinus pinaster Ait. forests.