Navegando por Data de Publicação, começando com "2025-11-03"

Filtrar resultados por ano ou mês

Agora exibindo 1 - 1 de 1

Diversidade espectral e sazonalidade no mapeamento de tipologias florestais tropicais usando Sentinel-2
(Universidade Federal de São Carlos, 2025-11-03) Soares, Bianca Jacobe Martins; Haneda, Léo Eiti; https://lattes.cnpq.br/3096651501379867; Molin, Paulo Guilherme; https://lattes.cnpq.br/1529819650942373; https://orcid.org/0000-0002-4587-935X; https://lattes.cnpq.br/3429859543850984
The need for continuous, scalable, and low-cost monitoring of tropical forest typologies and biodiversity gains is crucial to meet global Forest Landscape Restoration (FLR) commitments and to understand forest dynamics in the face of climate change. Given the vast extent of tropical forests, enhancing low-cost remote sensing (RS) systems is fundamental to overcoming technical barriers in monitoring. Therefore, this study aimed to evaluate the potential of the Sentinel-2A and 2B satellites to replicate and scale the forest class discrimination methodology, originally tested with the VENμS sensor, in FLR programs. The research sought to assess how spectral diversity and the influence of seasonality contribute to the classification of tropical forest typologies. The study was conducted at the Itatinga Forest Science Experimental Station (EECFI-USP), analyzing forest classes such as: mature seasonal deciduous forests (over 80 years old), natural regeneration, cerrado, active restoration areas (mixed planting of native seedlings), and homogeneous plantations of Eucalyptus and Pinus. Using images generated from a 20x20 meter spatial resolution orbital scene for summer and one for the winter period, the Random Forest algorithm was employed to integrate various biophysical and structural attributes: advanced vegetation indices (NDMI, EVI, NBR, NDVI, NDVIRe, and CLre1), seasonal layers, summer and winter delta, texture metrics, and spectral diversity indicators (alpha and beta), derived from the biodivMapR package. The results demonstrated the robustness of the approach, with the model achieving an overall accuracy of over 80% and an average cross-class accuracy of 86.2%. The textural, seasonal, and spectral diversity layers stood out as the most important attributes for class discrimination, reinforcing the role of canopy structural heterogeneity. The greatest spectral variability, essential for differentiation, was observed in forest regeneration and active restoration areas. The study also confirmed the potential of the Red-Edge and SWIR bands of Sentinel-2A and 2B in structural and physiological characterization, establishing a replicable and low-cost protocol using free tools (R and QGIS). Thus, the proposed methodology significantly contributes to democratizing and expanding the continuous monitoring of FLR, offering essential technical support for the management of forest landscapes.