Sensoriamento remoto na estimativa do estoque de carbono em fragmentos florestais da Mata Atlântica
Arquivos
Data
2021-07-26
Autores
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Federal de Viçosa
Resumo
O aumento da concentração atmosférica de gases poluentes preocupa a comunidade internacional, que estabeleceu metas de controle das mudanças climáticas. As florestas são um componente importante neste processo, pois capturam parte destes gases. Métodos de inventários tradicionais são onerosos e infactíveis em áreas extensas, ao passo que estimativas baseadas em sensoriamento remoto podem fornecer uma contabilização precisa dos estoques de carbono. Objetivou-se avaliar o potencial de variáveis de sensoriamento remoto de alta resolução espacial na estimação do estoque de carbono em fragmentos da Mata Atlântica. As áreas nativas da Universidade Federal de Viçosa, Campus Viçosa, foram classificadas em três estágios sucessionais, onde foram estabelecidas 10 parcelas (20x50 m). Foram obtidos a altura total (Ht) e o diâmetro a 1,3 m de altura (dap) dos indivíduos arbóreo-arbustivos com dap > 5,0 cm. O estoque de carbono foi estimado com o modelo de Schumacher e Hall ajustado para a vegetação secundária da região. Foram utilizadas imagens do satélite Sentinel-2, com resolução de 10 m, dos períodos seco e chuvoso. Foram obtidas sete variáveis espectrais: as reflectâncias das bandas Azul, Verde, Vermelho e Infravermelho Próximo, além dos Índices de Vegetação (IVs) Enhanced Vegetation Index, Normalized Difference Vegetation Index e Soil Adjusted Vegetation Index. A correlação entre as variáveis e o estoque de carbono foi verificada pela Correlação de Pearson. Foram treinadas Redes Neurais Artificiais (RNA) do tipo Multilayer Perceptron (MLP) com diferentes combinações de variáveis. A validação das RNA foi feita com 1.000 valores simulados para cada variável de entrada. O carbono foi modelado com a RNA de melhor desempenho no tremmamento e validação. Foram inventariados 4.114 indivíduos e 4.579 fustes. O estrato intermediário apresentou maior número de indivíduos (36,83%) e de fustes (36,34%). A Ht média foi de 9,74,9,55 e 11,98 me o diâmetro quadrático (Dq) de 11,33, 10,68 e 12,91 cm, nos estratos inicial, intermediário e avançado, respectivamente. Nesta ordem, o estoque de carbono médio amostrado foi 24,99, 35,79 e 82,28 Mg ha-¹, e média geral 47,68 Mg ha!. A distribuição diamétrica apresentou o padrão J-invertido em todos os estratos sucessionais. As classes diamétricas < 25 cm contribuiram com 62,92% do estoque de carbono no estrato inicial e 52,20% no intermediário. No estrato avançado indivíduos > 25 cm representaram 79,73% do estoque. As variáveis do período chuvoso apresentaram melhores correlações com o carbono estocado. A RNA treinada com as reflectâncias do período chuvoso apresentou bom desempenho, e fo1 selecionada para validação e estimação do carbono da área. O desempenho das RNA no treinamento não foi melhorado com a adição dos IVs. Os dados espectrais simulados foram consistentes e adequados para validação da RNA selecionada. O estoque total de carbono obtido pela modelagem da RNA foi de 41.962,15 Mg. Os valores variaram de 6,68 a 108,29 Mg ha-¹, e média de 48,70 Mg ha-¹. A metodologia utilizada éeficiente, e pode direcionar a contabilização acurada do carbono e o estabelecimento de práticas e políticas voltadas à mitigação das mudanças climáticas. Palavras-chave: Estrutura florestal. Inventário florestal. Modelagem não-paramétrica. Sentinel- 2. Sucessão florestal.
The increase in the atmospheric concentration of polluting gases worries the international community, which has set targets for controlling climate change. Forests are an important component 1n this process, as they capture part of these gases. Traditional inventory methods are costly and infeasible over large areas, while estimates based on remote sensing can provide an accurate accounting of carbon stocks. This study aimed to evaluate the potential of high spatial resolution remote sensing variables in the estimation of carbon stock in Atlantic Forest fragments. The native areas of the Universiade Federal de Vicosa, Campus Vicosa, were classified into three successive stages, where 10 plots (20x50 m) were established. Total height (Ht) and diameter at 1.3 m in height (dgh) of arboreal-shrubby individuals with dgh > 5.0 cmwere obtained. The carbon stock was estimated using the Schumacher and Hall model adjusted for secondary vegetation 1n the region. Sentinel-2 satellite images, with a resolution of 10 m, of the dry and rainy periods were used. Seven spectral variables were obtained: the reflectance of the Blue, Green, Red and Near Infrared bands, in addition to the Vegetation Indices (VIs) Enhanced Vegetation Index, Normalized Difference Vegetation Index and Soil Adjusted Vegetation Index. The correlation between variables and carbon stock was verified by Pearson's correlation. Multilayer Perceptron (MLP) Artificial Neural Networks (ANN) were trained with different combinations of variables. ANN validation was performed with 1,000 simulated values for each input variable. Carbon was modeled with the ANN with the best performance in training and validation. 4,114 individuals and 4,579 stems were inventoried. The intermediate stratum had a greater number of individuals (36.83%) and stems (36.34%). The average Ht was 9.74, 9.55 and 11.98 m and the quadratic diameter (Dg) of 11.33, 10.68 and 12.91 cm, in the initial, intermediate, and advanced strata, respectively. In this order, the average carbon stock sampled was 24.99, 35.79 and 82.28 Mg ha-¹ , and overall average 47.68 Mg ha™!. Diametric distribution presented the inverted-J pattern in all successional strata. Diametric classes < 25 cm contributed 62.92% of the carbon stock in the initial stratum and 52.20% in the intermediate. In the advanced stratum, individuals > 25 cm represented 79.73% of the stock. The rainy period variables showed better correlations with the stored carbon. The ANN trained with the rainy season reflectances performed well and was selected for validation and estimation of the area's carbon. The performance of ANNs in training was not improved with the addition of IVs. The simulated spectral data were consistent and adequate to validate the selected ANN. The total carbon stock obtained by ANN modeling was 41,962.15 Meg. Values ranged from 6.68 to 108.29 Mg ha-¹ , with an average of 48.70 Mg ha-¹ . The methodology used 1s efficient and can guide accurate carbon accounting and the establishment of practices and policies aimed at mitigating climate change. Keywords: Forest inventory. Forest structure. Forest succession. Non-parametric modeling. Sentinel-2.
The increase in the atmospheric concentration of polluting gases worries the international community, which has set targets for controlling climate change. Forests are an important component 1n this process, as they capture part of these gases. Traditional inventory methods are costly and infeasible over large areas, while estimates based on remote sensing can provide an accurate accounting of carbon stocks. This study aimed to evaluate the potential of high spatial resolution remote sensing variables in the estimation of carbon stock in Atlantic Forest fragments. The native areas of the Universiade Federal de Vicosa, Campus Vicosa, were classified into three successive stages, where 10 plots (20x50 m) were established. Total height (Ht) and diameter at 1.3 m in height (dgh) of arboreal-shrubby individuals with dgh > 5.0 cmwere obtained. The carbon stock was estimated using the Schumacher and Hall model adjusted for secondary vegetation 1n the region. Sentinel-2 satellite images, with a resolution of 10 m, of the dry and rainy periods were used. Seven spectral variables were obtained: the reflectance of the Blue, Green, Red and Near Infrared bands, in addition to the Vegetation Indices (VIs) Enhanced Vegetation Index, Normalized Difference Vegetation Index and Soil Adjusted Vegetation Index. The correlation between variables and carbon stock was verified by Pearson's correlation. Multilayer Perceptron (MLP) Artificial Neural Networks (ANN) were trained with different combinations of variables. ANN validation was performed with 1,000 simulated values for each input variable. Carbon was modeled with the ANN with the best performance in training and validation. 4,114 individuals and 4,579 stems were inventoried. The intermediate stratum had a greater number of individuals (36.83%) and stems (36.34%). The average Ht was 9.74, 9.55 and 11.98 m and the quadratic diameter (Dg) of 11.33, 10.68 and 12.91 cm, in the initial, intermediate, and advanced strata, respectively. In this order, the average carbon stock sampled was 24.99, 35.79 and 82.28 Mg ha-¹ , and overall average 47.68 Mg ha™!. Diametric distribution presented the inverted-J pattern in all successional strata. Diametric classes < 25 cm contributed 62.92% of the carbon stock in the initial stratum and 52.20% in the intermediate. In the advanced stratum, individuals > 25 cm represented 79.73% of the stock. The rainy period variables showed better correlations with the stored carbon. The ANN trained with the rainy season reflectances performed well and was selected for validation and estimation of the area's carbon. The performance of ANNs in training was not improved with the addition of IVs. The simulated spectral data were consistent and adequate to validate the selected ANN. The total carbon stock obtained by ANN modeling was 41,962.15 Meg. Values ranged from 6.68 to 108.29 Mg ha-¹ , with an average of 48.70 Mg ha-¹ . The methodology used 1s efficient and can guide accurate carbon accounting and the establishment of practices and policies aimed at mitigating climate change. Keywords: Forest inventory. Forest structure. Forest succession. Non-parametric modeling. Sentinel-2.
Descrição
Palavras-chave
Levantamentos florestais, Estatística não-paramétrica, Sucessão florestal
Citação
VERLY, Otávio Miranda. Sensoriamento remoto na estimativa do estoque de carbono em fragmentos florestais da Mata Atlântica. 2021. 55 f. Dissertação (Mestrado em Ciência Florestal) - Universidade Federal de Viçosa, Viçosa. 2021.