Remote Sensing for Land Use / Land Cover Mapping in Almada

• Autor(a) principal: Oliveira, Inês Cardoso Leitão Barata de
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/151807
• topic: Remote sensing; Sentinel-1; Sentinel-2; Machine Learning; LULC Mapping; Object-based LULC Classification; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

Monitoring land use and land cover is an extremely important task which, if properly carried out, can assist in decision making about urban and territorial planning, thus pro- viding an improvement in the citizens’ quality of life. In Portugal, and more specifically in the Almada municipality , the main tool used in this task is Carta de Ocupação de Solo (COS), a map which represents 83 classes of land use and land cover. Despite its useful- ness, COS has certain limitations, such as low spatial resolution, due to the minimum mapping unit of 1 hectare, and low temporal resolution, as it is developed through the analysis of orthophotos and released every 3 to 5 years. These constraints lead to a map which is not adequate to continuously track land-use and land-cover changes, especially with the increasingly fast pace of urbanization. This research work investigated the application of machine learning classification algorithms with Sentinel-1 and Sentinel-2 imagery, and derived products, to LULC map- ping in Almada. As such, maps were developed for 2018 using the two most common approaches to LULC classification: pixel-based (PBIA) and object-based (OBIA). Multiple combinations of satellite data and derived products, as well as two classifiers were tested for each approach. A comparison of two methods of collecting ground truth data, manual and semi-automatic, was also produced. The best results were obtained in the PBIA approach, using the manually collected ground truth and the Extreme Gradient Boosting (XGBoost) classifier with the combina- tion of Sentinel-1 and Sentinel-2 imagery and textural features obtained through Sentinel- 2 data. The classification model obtained a kappa score of 0.994, and produced an ac- curate LULC map, which has some limitations in separating Agriculture and Other Vegetation, but is able to identify with great precision Artificial Territories, Forests and Bare and sparsely vegetated areas.