Unsupervised Spatio-Temporal Analysis of Coastal Upwelling from Sea Surface Temperature Images

• Autor(a) principal: Martins, Alexandre Guerreiro
• 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/10362/141080
• topic: coastal upwelling; spatio-temporal clustering; image segmentation validation; sea surface temperature images; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

Oceanic dynamics are a hot topic in the scientific literature. When equatorward winds blow parallel to the coastline, these can form oceanic currents that can consequently create many phenomena, one of these being coastal upwelling. The automatic analysis, segmentation, study and understanding of such structures and phenomena is of major importance in various application areas mainly due to the time-consuming work needed to manually analyze such data. The segmentation and tracking of the dynamics of such structures can be performed by the use of a novelty clustering concept introduced in [20] where dynamic Spatio-Temporal (ST) clusters are studied and identified using consecutive Sea Surface Temperature (SST) images over a period of time. The aim of this dissertation was to develop a new algorithm called Core-Shell clustering algorithm, which is an extension of the previously developed Sequential Self Tuning Seeded Expanding Cluster (S-STSEC) algorithm, proposed in [56, 59]. This new algorithm aims at the automatic recognition, definition and ST characterization of coastal upwelling from SST images. A suitable experimental protocol for SST image preprocessing, tests and validation of the Core-Shell clustering algorithm through two unsupervised measures was developed. Collections of images from the years 2007, 2015 and 2019 were chosen to total 69 SST images. The segmentations’ results obtained by the S-STSEC algorithm of the various SST instants were manually evaluated as accurate and of high quality. The two unsupervised evaluation measures, when applied, were used to evaluate the quality of the Core-Shell clusters created by the Core-Shell clustering algorithm when compared to the correspondent SST instants’ segmentations, with mean values higher than 85%. Time series were extracted and segmented in an unsupervised manner for the computation of upwelling spans. It was concluded that the results obtained correctly represent the behavior of coastal upwelling regions and their dynamics.