A carregar...
Publicação
Modelagem preditiva do uso e cobertura do solo com modelo híbrido AC RNA: Simulação de cenários territoriais no Vale do Jamari, Rondônia
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 3.24 MB | Adobe PDF |
Orientador(es)
Resumo(s)
A intensificação da atividade agropecuária na Amazônia Legal tem impulsionado transformações profundas na paisagem, exigindo ferramentas precisas para o monitoramento e o planejamento territorial. Esta dissertação investiga a dinâmica de uso e cobertura do solo no Território Vale do Jamari, em Rondônia, uma região caracterizada por elevada pressão antrópica e rápida conversão florestal. O objetivo principal do estudo foi aplicar e avaliar um modelo híbrido que integra Autômatos Celulares (AC) e Redes Neurais Artificiais (RNA) para simular padrões espaciais de mudança e projetar cenários futuros. A metodologia utilizou séries históricas do projeto MapBiomas para o período de 2014 a 2018 como base para o treinamento de uma Rede Neural do tipo Perceptron Multicamadas (MLP), processada no ambiente do software QGIS através do plugin MOLUSCE. O modelo foi validado simulando o cenário de 2022 e comparando-o com os dados observados, demonstrando elevada concordância estatística e capacidade de capturar a não linearidade dos processos de desmatamento. Com base no modelo calibrado, foram construídos cenários prospectivos para os horizontes de 2030 e 2050, contrastando uma tendência de continuidade histórica (Business as Usual) com cenários de governança que simulam a proteção efetiva de Unidades de Conservação, Terras Indígenas e Áreas de Preservação Permanente. Os resultados indicam que a abordagem híbrida AC-RNA é eficaz para representar a complexidade territorial da região, oferecendo subsídios técnicos robustos e de baixo custo para a formulação de políticas públicas voltadas ao ordenamento territorial e à sustentabilidade ambiental na Amazônia.
The intensification of agricultural activity in the Legal Amazon has driven profound landscape transformations, requiring precise tools for monitoring and territorial planning. This dissertation investigates land use and land cover dynamics in the Vale do Jamari Territory, Rondônia, a region characterized by high anthropic pressure and rapid forest conversion. The main objective of the study was to apply and evaluate a hybrid model integrating Cellular Automata (CA) and Artificial Neural Networks (ANN) to simulate spatial change patterns and project future scenarios. The methodology used historical series from the MapBiomas project for the period 2014 to 2018 as the basis for training a Multi-layer Perceptron (MLP) Neural Network, processed within the QGIS software environment using the MOLUSCE plugin. The model was validated by simulating the 2022 scenario and comparing it with observed data, demonstrating high statistical agreement and the ability to capture the non-linearity of deforestation processes. Based on the calibrated model, prospective scenarios were constructed for the 2030 and 2050 horizons, contrasting a historical continuity trend (Business as Usual) with governance scenarios simulating the effective protection of Conservation Units, Indigenous Lands, and Permanent Preservation Areas. The results indicate that the hybrid CA-ANN approach is effective in representing the territorial complexity of the region, offering robust and low-cost technical subsidies for the formulation of public policies aimed at territorial planning and environmental sustainability in the Amazon.
The intensification of agricultural activity in the Legal Amazon has driven profound landscape transformations, requiring precise tools for monitoring and territorial planning. This dissertation investigates land use and land cover dynamics in the Vale do Jamari Territory, Rondônia, a region characterized by high anthropic pressure and rapid forest conversion. The main objective of the study was to apply and evaluate a hybrid model integrating Cellular Automata (CA) and Artificial Neural Networks (ANN) to simulate spatial change patterns and project future scenarios. The methodology used historical series from the MapBiomas project for the period 2014 to 2018 as the basis for training a Multi-layer Perceptron (MLP) Neural Network, processed within the QGIS software environment using the MOLUSCE plugin. The model was validated by simulating the 2022 scenario and comparing it with observed data, demonstrating high statistical agreement and the ability to capture the non-linearity of deforestation processes. Based on the calibrated model, prospective scenarios were constructed for the 2030 and 2050 horizons, contrasting a historical continuity trend (Business as Usual) with governance scenarios simulating the effective protection of Conservation Units, Indigenous Lands, and Permanent Preservation Areas. The results indicate that the hybrid CA-ANN approach is effective in representing the territorial complexity of the region, offering robust and low-cost technical subsidies for the formulation of public policies aimed at territorial planning and environmental sustainability in the Amazon.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Geographic Information Systems and Science, specialization in Geographic Information Systems and Science
Palavras-chave
Modelagem Espacial Uso e Cobertura do Solo Vale do Jamari Redes Neurais Artificiais Autômatos Celulares MapBiomas Spatial Modeling Land Use and Land Cover Artificial Neural Networks Cellular Automata