| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 11.29 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
O sequestro de carbono assume-se como sendo um dos principais mecanismos naturais de mitigação das alterações climáticas, sendo os solos e as florestas reservatórios essenciais neste processo. Esta dissertação tem como objetivo principal modelar o potencial de sequestro de carbono em Portugal continental para o cenário futuro de 2050, por meio de algoritmos lineares e de inteligência artificial, de forma a melhorar o conhecimento sobre a dinâmica do carbono e o apoio a estratégias de gestão sustentável do território. Os objetivos específicos passam por: 1) elaborar uma nova cartografia de solos à escala 1:100 000, através da harmonização dos dados disponíveis nas bases de dados INFOSOLO e LUCAS, e na cartografia existente; 2) avaliar o potencial de sequestro de carbono cruzando a nova carta de solos com a distribuição da vegetação florestal, recorrendo a algoritmos de machine learning (Random Forest) e ao software InVEST. Os resultados demonstram que a integração de novas bases cartográficas melhora a capacidade preditiva dos modelos, destacando fatores edafoclimáticos, como o índice de precipitação positiva e o pH do solo, como determinantes no sequestro de carbono. O Random Forest evidenciou um forte desempenho, permitindo identificar áreas com maior potencial de sequestro, enquanto o InVEST revelou padrões espaciais heterogéneos, ainda que condicionados pelas simplificações do modelo. Verificou-se que as florestas constituem os principais sumidouros de carbono, contrastando com o contributo mais limitado dos sistemas agroflorestais. Conclui-se que a articulação entre cartografia atualizada, algoritmos de inteligência artificial e ferramentas de modelação espacial representam um avanço significativo para a compreensão do potencial de sequestro de carbono em Portugal continental. Este trabalho fornece não só contributos científicos relevantes, como também instrumentos de apoio à decisão para políticas de ordenamento, gestão florestal e mitigação das alterações climáticas.
Carbon sequestration is recognised as one of the primary natural mechanisms for mitigating climate change, with soils and forests serving as essential reservoirs in this process. This dissertation aims to model the carbon sequestration potential in mainland Portugal for the 2050 scenario, utilising linear algorithms and artificial intelligence, to enhance knowledge of carbon dynamics and support strategies for sustainable land management. The specific objectives are: 1) to develop a new soil map at a 1:100,000 scale, by harmonising data available in the INFOSOLO and LUCAS databases, as well as existing cartography; 2) to assess the carbon sequestration potential by combining the new soil map with the distribution of forest vegetation, using machine learning algorithms (Random Forest) and the InVEST software. The results show that integrating new cartographic bases improves the predictive capacity of the models, highlighting edaphoclimatic factors such as the positive precipitation index and soil pH as key determinants in carbon sequestration. Random Forest demonstrated strong performance, enabling the identification of areas with greater sequestration potential, while InVEST revealed heterogeneous spatial patterns, albeit constrained by the model’s simplifications. Forests were found to be the main carbon sinks, in contrast with the more limited contribution of agroforestry systems. It is concluded that the combination of updated cartography, artificial intelligence algorithms and spatial modeling tools represents a significant advance in understanding the carbon sequestration potential in mainland Portugal. This study provides not only relevant scientific contributions but also decision-support tools for land-use policies, forest management, and climate change mitigation.
Carbon sequestration is recognised as one of the primary natural mechanisms for mitigating climate change, with soils and forests serving as essential reservoirs in this process. This dissertation aims to model the carbon sequestration potential in mainland Portugal for the 2050 scenario, utilising linear algorithms and artificial intelligence, to enhance knowledge of carbon dynamics and support strategies for sustainable land management. The specific objectives are: 1) to develop a new soil map at a 1:100,000 scale, by harmonising data available in the INFOSOLO and LUCAS databases, as well as existing cartography; 2) to assess the carbon sequestration potential by combining the new soil map with the distribution of forest vegetation, using machine learning algorithms (Random Forest) and the InVEST software. The results show that integrating new cartographic bases improves the predictive capacity of the models, highlighting edaphoclimatic factors such as the positive precipitation index and soil pH as key determinants in carbon sequestration. Random Forest demonstrated strong performance, enabling the identification of areas with greater sequestration potential, while InVEST revealed heterogeneous spatial patterns, albeit constrained by the model’s simplifications. Forests were found to be the main carbon sinks, in contrast with the more limited contribution of agroforestry systems. It is concluded that the combination of updated cartography, artificial intelligence algorithms and spatial modeling tools represents a significant advance in understanding the carbon sequestration potential in mainland Portugal. This study provides not only relevant scientific contributions but also decision-support tools for land-use policies, forest management, and climate change mitigation.
Descrição
Este trabalho foi desenvolvido no âmbito do concurso Science4Policy 2024 (S4P-24), o concurso de Estudos de Ciência para as Políticas Públicas, uma iniciativa do Centro de Planeamento e Avaliação de Políticas Públicas em parceria com a Fundação para a Ciência e Tecnologia, I.P., e financiada pelo Plano de Recuperação e Resiliência. Projeto ML-SOIL (2024.00178.S4P24), Modelação Coparticipativa de Distritos de Solo com base em Machine learning.
Palavras-chave
Carbon Sequestration Forests Soils Machine Learning Geographic Information Systems Sequestro de Carbono Florestas Solos Machine Learning Sistemas de Informação Geográfica
