A carregar...
Publicação
Determinação da Idade da Floresta Nacional Usando Séries Temporais de Imagens de Satélite
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 16.42 MB | Adobe PDF |
Orientador(es)
Resumo(s)
A floresta é um recurso natural importante a nível ecológico e económico. Dado que a
floresta é o principal uso de solo em área ocupada no território português, o seu desenvol-
vimento merece especial atenção, sendo necessárias ferramentas para a monitorizar. Uma
caraterística importante da floresta é a sua idade pois carateriza o seu aproveitamento
e capacidade de filtração de dióxido de carbono. Sendo um indicador da quantidade de
biomassa presente, poderá fornecer informação adicional no combate a incêndios. Portu-
gal possui fontes de informação florestal, como é o caso do Inventário Florestal Nacional
(IFN), da Carta de Uso e Ocupação do Solo (COS) e da cartografia de áreas ardidas. Quer
o IFN quer a COS fornecem informação sobre a distribuição geográfica da floresta e das
espécies que a compõem. A atualização destas fontes de informação é irregular, tendo no
passado demorado 3 a 12 anos entre edições consecutivas.
A determinação da idade da floresta foi conseguida pela deteção de momentos de
distúrbio, analisando séries temporais provenientes de imagens de satélite. Este assunto
não é trivial, e o estado da arte não oferece soluções concretas, pelo que foi necessário
comparar algumas abordagens para selecionar a que melhor se adequava ao problema. No
campo da análise de séries temporais de deteção remota destacaram-se dois algoritmos:
o LandTrendr e o CCDC. Adicionalmente, no campo da análise de pontos de quebra
em séries temporais genéricas, destacou-se o BOCPD. Foram usados dados das missões
Landsat, que já contam com um historial de 4 décadas. Devido à escassez de dados sobre
a idade da floresta, foi também necessária a criação de um novo conjunto de dados de
referência para avaliar os algoritmos. Este conjunto de dados indica de forma exaustiva
todos os distúrbios entre os anos de 1986 e 2019 para 664 pontos de floresta portuguesa.
Comparando estes dados com as classificações obtidas pelos algoritmos já referidos, foi
possível concluir que o BOCPD obtém os melhores resultados com um F1 de 0.717, False
Negative Rate de 0.365 e False Discovery Rate de 0.175.
The forest is important at an ecological and economical level. Given that the forest is the main land use by area in Portuguese territory, its development deserves special attention, therefore tools are needed to monitor it. An important characteristic of the forest is its age as it characterizes its economical potential and capacity to filter carbon dioxide. As an indicator of the amount of biomass, it can also provide additional infor- mation for fire-fighters. Portugal has forest information sources, such as the Inventário Florestal Nacional (IFN), the Carta de Uso e Ocupação do Solo (COS) and the burned areas cartography. Both the IFN and the COS provide information on the geographic distribution of the forest and its composition. Updates on these information sources are irregular, having taken as little as 3 to as much as 12 years between consecutive editions. Determining the age of the forest was achieved by detecting disturbance moments, while analysing time series from satellite images. This issue is not trivial, and the state of the art does not offer concrete solutions, so it was necessary to compare some approaches and to select the one that best suited the problem. In the field of remote sensing time series analysis, two algorithms stood out: LandTrendr and CCDC. Additionally, in the field of of breakpoint detection in generic time series, the BOCPD stood out. Data from Landsat missions, which already has a history of 4 decades, were used. Due to the scarcity of data on the age of the forest, it was also necessary to create a new reference dataset to evaluate the algorithms. This dataset exhaustively indicates all disturbances between 1986 and 2019 for 664 points of Portuguese forest. Comparing this data with the clas- sifications obtained by the aforementioned algorithms, it was possible to conclude that the BOCPD obtains the best results with an F1 of 0.717, False Negative Rate of 0.365 and False Discovery Rate of 0.175.
The forest is important at an ecological and economical level. Given that the forest is the main land use by area in Portuguese territory, its development deserves special attention, therefore tools are needed to monitor it. An important characteristic of the forest is its age as it characterizes its economical potential and capacity to filter carbon dioxide. As an indicator of the amount of biomass, it can also provide additional infor- mation for fire-fighters. Portugal has forest information sources, such as the Inventário Florestal Nacional (IFN), the Carta de Uso e Ocupação do Solo (COS) and the burned areas cartography. Both the IFN and the COS provide information on the geographic distribution of the forest and its composition. Updates on these information sources are irregular, having taken as little as 3 to as much as 12 years between consecutive editions. Determining the age of the forest was achieved by detecting disturbance moments, while analysing time series from satellite images. This issue is not trivial, and the state of the art does not offer concrete solutions, so it was necessary to compare some approaches and to select the one that best suited the problem. In the field of remote sensing time series analysis, two algorithms stood out: LandTrendr and CCDC. Additionally, in the field of of breakpoint detection in generic time series, the BOCPD stood out. Data from Landsat missions, which already has a history of 4 decades, were used. Due to the scarcity of data on the age of the forest, it was also necessary to create a new reference dataset to evaluate the algorithms. This dataset exhaustively indicates all disturbances between 1986 and 2019 for 664 points of Portuguese forest. Comparing this data with the clas- sifications obtained by the aforementioned algorithms, it was possible to conclude that the BOCPD obtains the best results with an F1 of 0.717, False Negative Rate of 0.365 and False Discovery Rate of 0.175.
Descrição
Palavras-chave
Deteção Remota Idade da Floresta Séries temporais Deteção de alterações Florestais