Loading...
Publication
Análise de Espécies Invasoras e Respetiva Proposta de Intervenção
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 5.15 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
Em Portugal, de acordo com o Decreto-Lei 92/2019 de 10 de julho, existem cerca
de 300 espécies Invasoras. As espécies exóticas, sejam animais, plantas ou outros
organismos não representam por si só um problema. Essa questão surge quando uma
espécie apresenta um comportamento invasor. O concelho de Gavião, à semelhança de
outros em Portugal, é atingido pelo fenómeno das espécies invasoras, nomeadamente
espécies de Acacia e Hakea.
Os objetivos deste trabalho consistiram, numa primeira fase, na localização e
caracterização de uma seleção de plantas invasoras presentes no concelho de Gavião e,
numa segunda fase, na definição de uma proposta de intervenção para controlo das
mesmas com vista a minimizar os seus impactes. Utilizei a deteção remota (DR) e os
métodos de aprendizagem automática para a classificação das imagens.
Complementarmente à informação de campo, foram delimitadas várias áreas
aleatoriamente no sentido de potenciar a validação efetuada. Posteriormente efetuouse a validação de campo das áreas delimitadas com base nos classificadores Random
Forest e Máquina Vetor Suporte, que revelaram valores de eficácia diferentes.
A aplicação do algoritmo da floresta aleatória (Ramdom Forest) alcançou um
índice Kappa de 65% o que significa que a sua classificação é substancial. A área ocupada
por espécies invasoras através do algoritmo em causa, foi de aproximadamente 4% do
território, o que equivale a mais de 1000 hectares de acácias. O algoritmo Máquina
Vetor Suporte alcançou um índice de 51%, obtendo uma classificação moderada.
A aprendizagem automática revela-se uma excelente ferramenta de deteção e
de classificação de imagens. Através da deteção remota foi possível identificar e definir
um plano estratégico. Foi possível através da DR extrair as áreas de acácias, para as
háqueas não foi possível definir as áreas devido à confusão no espectro com os matos.
De acordo com as espécies indicadas, foi elaborada uma proposta abrangendo
as várias etapas de controlo, incluído as metodologias mais adequadas, bem como o
custo possível para execução de ações de controlo numa área piloto
PALAVRAS-CHAVE: Espécies Invasoras, Deteção Remota, Índice Kappa, Floresta
Aleatória, Aprendizagem automática, Sentinel-2
In Portugal, according to Decree-Law 92/2019 of 10 July, there are about 300 Invasive species. Exotic species, whether animals, plants or other organisms are not in themselves a problem. This question arises when a species exhibits invasive behavior. The municipality of Gavião, like others in Portugal, is affected by the phenomenon of invasive species, namely Acacias and Hakeas. The methodology used in the present work consists, in a first phase, in the localization and characterization of the invasive species present in Gavião county and, in a second phase, in the definition of an intervention proposal to control them in order to minimize their impacts. The information used comes from field data collection, complemented by satellite images obtained by Sentinel. The images analyzed date from May 25, 2019, at a time of year when most species are in flowering, which facilitates the treatment of their image and further analysis. In addition to the field information, several areas were randomly delimited in order to enhance the validation performed. Subsequently, the field validation of the delimited areas was performed based on two classifiers, which revealed different efficacy values. The application of the random forest algorithm reached a 65% Kappa index which means that its classification is substantial. The area occupied by invasive species through the algorithm in question was approximately 4% of the territory, which is equivalent to over 1000 hectares of acacia trees. According to the indicated species, the most appropriate forms of control were suggested, as well as the possible cost of carrying out control actions in a pilot area. Machine learning proves to be an excellent image detection and classification tool. Through remote sensing it was possible to identify and define a strategic plan. KEYWORDS: Invasive species, Acacia dealbata, Acacia mearnsii, Hakea sericea, Machine Learning, Sentinel-2
In Portugal, according to Decree-Law 92/2019 of 10 July, there are about 300 Invasive species. Exotic species, whether animals, plants or other organisms are not in themselves a problem. This question arises when a species exhibits invasive behavior. The municipality of Gavião, like others in Portugal, is affected by the phenomenon of invasive species, namely Acacias and Hakeas. The methodology used in the present work consists, in a first phase, in the localization and characterization of the invasive species present in Gavião county and, in a second phase, in the definition of an intervention proposal to control them in order to minimize their impacts. The information used comes from field data collection, complemented by satellite images obtained by Sentinel. The images analyzed date from May 25, 2019, at a time of year when most species are in flowering, which facilitates the treatment of their image and further analysis. In addition to the field information, several areas were randomly delimited in order to enhance the validation performed. Subsequently, the field validation of the delimited areas was performed based on two classifiers, which revealed different efficacy values. The application of the random forest algorithm reached a 65% Kappa index which means that its classification is substantial. The area occupied by invasive species through the algorithm in question was approximately 4% of the territory, which is equivalent to over 1000 hectares of acacia trees. According to the indicated species, the most appropriate forms of control were suggested, as well as the possible cost of carrying out control actions in a pilot area. Machine learning proves to be an excellent image detection and classification tool. Through remote sensing it was possible to identify and define a strategic plan. KEYWORDS: Invasive species, Acacia dealbata, Acacia mearnsii, Hakea sericea, Machine Learning, Sentinel-2
Description
Keywords
Espécies Invasoras Deteção Remota Índice Kappa Floresta Aleatória Aprendizagem automática Sentinel-2