Uso de redes neurais artificiais para análise multitemporal da dinâmica do uso e cobertura da terra em bacias hidrográficas
Carregando...
Arquivos
Data
2022-01-17
Autores
Silva, Camille Vasconcelos
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
As atividades antrópicas se relacionam de forma direta com o uso dos recursos naturais, promovendo de forma acelerada um cenário de degradação ambiental. Essas alterações são oriundas principalmente do crescimento populacional desordenado e das atividades agropecuárias e industriais. Dentre as consequências dessas mudanças, tem-se a redução das florestas nativas, em especial a Mata Atlântica, que apresenta apenas 13% de remanescente florestal. Nesse cenário, as geotecnologias, como os Sistemas de Informação Geográficas (SIG) e o Sensoriamento Remoto, têm apresentado papel importante como ferramentas para o monitoramento ambiental. Diante disso, esta pesquisa tem como objetivo avaliar o potencial das ferramentas de SIG associadas às técnicas de Inteligência Artificial na análise temporal da dinâmica do uso e cobertura da terra de 2000 a 2020 na Bacia Hidrográfica do Rio Sorocabuçu (BHRS), em Ibiúna, São Paulo, Brasil. Para o desenvolvimento deste estudo foi organizada uma base cartográfica com informações planimétricas e altimétricas, como hidrografia, malha viária, curvas de nível, além da utilização de imagens multiespectrais do satélite Landsat 7, sensor ETM+, e Landsat 8, sensor OLI. O processamento e análise das imagens e dados foram efetuados nos softwares ArcGis e Qgis. Para o mapeamento do uso e cobertura da terra foram realizados os métodos por interpretação visual, onde as feições são identificadas de acordo com seu padrão de cor, textura e forma, e o de classificação por Redes Neurais Artificiais (RNA) realizada no software IDRISI Selva, através do módulo Multi-Layer-Perpectron (MLP), operando com o algoritmo Error Backpropagation. Para a verificação da acurácia da classificação foi utilizado o coeficiente kappa (k). Foram caracterizadas seis classes de uso e cobertura da terra: Áreas Urbanizadas, Áreas Agrícolas, Área Vegetacional Perturbada, Reflorestamento, Mata e Corpos Hídricos. Os mapas de uso da terra obtidos pela classificação RNA MLP apresentaram índice kappa de 0,82 apontando excelente classificação. Foi constatada variação ao longo do tempo analisado de áreas com atividades antrópicas de 44,71% para 47,74% e áreas naturais de 55,29% para 52,26% pela Interpretação Visual. Pela RNA MLP as áreas antropizadas mudaram de 50,57% para 56,11% e as áreas naturais de 49,43% para 43,89%. Uma fragilidade identificada no método da RNA MLP foi a superestimação de algumas classes, como Área Urbanizada, Áreas Agrícolas, Área Vegetacional Perturbada e Reflorestamento. Como potencialidade, essa metodologia possibilitou gerar um diagnóstico ambiental mais rápido e de menor custo e pode ser replicada para o reconhecimento de padrões e classificação de imagens de satélite em outros ecossistemas florestais. Esta pesquisa pode contribuir para a identificação de potenciais impactos socioambientais e auxiliar no planejamento e gestão ambiental da BHRS.
Human activities are directly related to the use of natural resources, rapidly promoting a scenario of environmental degradation. These changes are mainly due to disordered population growth and agricultural and industrial activities. Among the consequences of these changes, there is the reduction of native forests, especially the Atlantic Forest, which has only 13% of remaining forest. In this scenario, geotechnologies, such as Geographic Information Systems (GIS) and Remote Sensing, have played an important role as tools for environmental monitoring. Therefore, this research aims to evaluate the potential of GIS tools associated with Artificial Intelligence techniques in the temporal analysis of the dynamics of land use and land cover from 2000 to 2020 in the Sorocabuçu River Basin (BHRS), in Ibiúna, São Paulo, Brazil. For the development of this study, a cartographic base was organized with planimetric and altimetric information, such as hydrography, road network, contour lines, in addition to the use of multispectral images from the Landsat 7 satellite, ETM+ sensor, and Landsat 8, OLI sensor. The processing and analysis of images and data were performed using ArcGis and Qgis software. For the mapping of land use and land cover, the methods by visual interpretation were performed, where the features are identified according to their color, texture and shape pattern, and the classification by Artificial Neural Networks (ANN) performed in the IDRISI Selva software, through the Multi-Layer-Perpectron (MLP) module, operating with the error backpropagation algorithm. The kappa coefficient (k) was used to verify the accuracy of the classification. Six classes of land use and land cover were characterized: Urbanized Areas, Agricultural Areas, Disturbed Vegetation Area, Reforestation, Forest and Water Bodies. The land use maps obtained by the RNA MLP classification presented a kappa index of 0.82 indicating an excellent classification. It was found variation over the analyzed time of areas with anthropic activities from 44.71% to 47.74% and natural areas from 55.29% to 52.26% by Visual Interpretation. By RNA MLP, anthropized areas changed from 50.57% to 56.11% and natural areas from 49.43% to 43.89%. A weakness identified in the MLP RNA method was the overestimation of some classes, such as Urbanized Area, Agricultural Areas, Disturbed Vegetation Area and Reforestation. As a potential, this methodology made it possible to generate a faster and cheaper environmental diagnosis and can be replicated for pattern recognition and classification of satellite images in other forest ecosystems. This research can contribute to the identification of potential social and environmental impacts and assist in BHRS environmental planning and management.
Human activities are directly related to the use of natural resources, rapidly promoting a scenario of environmental degradation. These changes are mainly due to disordered population growth and agricultural and industrial activities. Among the consequences of these changes, there is the reduction of native forests, especially the Atlantic Forest, which has only 13% of remaining forest. In this scenario, geotechnologies, such as Geographic Information Systems (GIS) and Remote Sensing, have played an important role as tools for environmental monitoring. Therefore, this research aims to evaluate the potential of GIS tools associated with Artificial Intelligence techniques in the temporal analysis of the dynamics of land use and land cover from 2000 to 2020 in the Sorocabuçu River Basin (BHRS), in Ibiúna, São Paulo, Brazil. For the development of this study, a cartographic base was organized with planimetric and altimetric information, such as hydrography, road network, contour lines, in addition to the use of multispectral images from the Landsat 7 satellite, ETM+ sensor, and Landsat 8, OLI sensor. The processing and analysis of images and data were performed using ArcGis and Qgis software. For the mapping of land use and land cover, the methods by visual interpretation were performed, where the features are identified according to their color, texture and shape pattern, and the classification by Artificial Neural Networks (ANN) performed in the IDRISI Selva software, through the Multi-Layer-Perpectron (MLP) module, operating with the error backpropagation algorithm. The kappa coefficient (k) was used to verify the accuracy of the classification. Six classes of land use and land cover were characterized: Urbanized Areas, Agricultural Areas, Disturbed Vegetation Area, Reforestation, Forest and Water Bodies. The land use maps obtained by the RNA MLP classification presented a kappa index of 0.82 indicating an excellent classification. It was found variation over the analyzed time of areas with anthropic activities from 44.71% to 47.74% and natural areas from 55.29% to 52.26% by Visual Interpretation. By RNA MLP, anthropized areas changed from 50.57% to 56.11% and natural areas from 49.43% to 43.89%. A weakness identified in the MLP RNA method was the overestimation of some classes, such as Urbanized Area, Agricultural Areas, Disturbed Vegetation Area and Reforestation. As a potential, this methodology made it possible to generate a faster and cheaper environmental diagnosis and can be replicated for pattern recognition and classification of satellite images in other forest ecosystems. This research can contribute to the identification of potential social and environmental impacts and assist in BHRS environmental planning and management.
Descrição
Palavras-chave
Geoprocessamento, Inteligência Artificial, Mata Atlântica, Geoprocessing, Artificial intelligence, Atlantic Forest