Identificação de áreas de clareira na floresta amazônica por meio de dados lidar, imagens de média resolução espacial e inteligência artificial

• Autor(a) principal: Oliveira, Bruna Andriéli Simões de
• Data de Publicação: 2020
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Manancial - Repositório Digital da UFSM
• Texto Completo: http://repositorio.ufsm.br/handle/1/23365
• topic: Clareiras no dossel florestal; Amazônia; Inteligência artificial; Sensoriamento remoto; Gaps in the forest canopy; Amazon; Artificial intelligence; Remote sensing; CNPQ::CIENCIAS AGRARIAS::RECURSOS FLORESTAIS E ENGENHARIA FLORESTAL

Tropical forests are naturally subject to disturbances, in addition to occurring naturally, they can be caused by global warming, forest fires and deforestation. Thus, it is important to conduct this study, since the practices of selective extraction in tropical forests provide the opening of gaps in the forest canopy, which need studies in order to assess the dynamics and regeneration of these spaces. This study aimed to identify the presence of gaps through medium spatial resolution images and LiDAR sensor, using artificial intelligence in areas of selective forest exploration in the Amazon biome. The study area is located at Fazenda Cauaxi, municipality of Paragominas-PA, in which the forest management activity is performed. To identify the clearings, an orbital image from the Sentinel-2A satellite was used, acquired by the Multi-Spectral Instrument (MSI) sensor, in 2017. For classification of clearings by means of the satellite image, the algorithms were used Random Forest (RF), Support Vectors Machine (SVM) e Artificial Neural Network (ANN), the supervised classification of orbital images was developed in Language R. In addition to the Sentinel-2A images, the study also included the use of LiDAR data to detect and analyze the dynamics of the clearings for the years 2014 and 2017, in order to allow the comparison of the results obtained by both methods. For the detection of gaps with LiDAR data, version 0.0.2 of the ForestGapR package was used. The results indicated that for the classification with Sentinel-2A image, all the algorithms presented values above 0.90 of global accuracy and values above 0.88 of Kappa, when verifying the machine learning algorithms and their respective adjustments, the RF was the one with the highest values, demonstrating an accuracy in the classification process of 0.9938, presenting as the best classifier for the identification of clearings in areas of selective exploration in the Amazon. For the dynamics of gaps in the forest canopy, using LiDAR data, for the years 2014 and 2017, the number of clearings increased by 10,098.00 in 3 years and an increase of 127,521.00 m² of total area. Regarding the comparison of methods, for Sentinel-2A image 229,402.97 m² of gaps were identified, while with LiDAR data 301,090.00 m² of gaps were detected, a difference of 71,687.03 m² of gaps between the methods. When relating the percentage of gaps identified with the use of MSI/Sentinel-2A images to those detected with LiDAR data under similar conditions, the potential of medium resolution images, when associated with artificial intelligence techniques, in the identification of disorders in the forest. Thus, the use of Sentinel products associated with complex processing techniques allows obtaining parameters of forest cover, including for the Amazon region. Finally, this study demonstrates the importance of identifying and analyzing clearings, through remote sensing applications, for monitoring deforestation and illegal logging in the forest, enabling sustainable management in the Brazilian Amazon.