Publicação
Análise de componentes independentes da variabilidade mensal do campo da pressão ao nível do mar na região Euro-Atlântica
| Resumo: | The climate system is a dynamic system characterized by its high complexity and dimensionality, governed by non-linear interactions. Over the last few decades, the redefinition of the system’s dimensionality and the extraction of co-related patterns from measurements of atmospheric variables, present challenging tasks. An example is the monthly variability of the pressure field at sea level (SLP) which allows for evidence of certain weather time patterns, characterized by some recurrence (or repetition) and temporal persistence. In this regard, the positive and negative phases of the North Atlantic Oscillation, shows the multimodality of the field’s joint probability distribution. This multivariate variability, whether forced or internally induced, can be separated, using certain multivariate statistical techniques into scalar random components, with as much statistical independence as possible, called independent components. In case of fields having multivariate Gaussian distributions, Independent Component Analysis (ICA) corresponds to Principal Component Analysis (PCA), which is then performed sub-optimally in the non-Gaussian case as it occurs with the geopotential field at the sea level.In this dissertation we intend to apply the SLP field research carried out by NCAR (National Center for Atmospheric Research) reanalysis, wich was provided in a regular grid with a resolution of 2,5º x 2,5º (latitude, longitude) covering the period between January 1948 to December 2017 (70 years). In this dissertation one aims to apply the ICA to the SLP field, obtained from the NCAR reanalysis, thus contributing to the search for independent components within the atmospheric and oceanic phenomena, so that it is possible to deepen the understanding of the fundamental processes behind its variability.As such, a geographic analysis covering the North Atlantic (Atl) and Euro-Atlantic (Eur-Atl)areas was carried out, including a temporal analysis (for both areas), each one including two main climatological seasons: winter (consisting of December, January and February) and summer (consisting of June, July and August). The applied method consists of two phases, in wich the PCA (Principal Component Analysis) is applied in the first and the ICA (Independent Components Analysis) is applied to the PCA result in the second. ICA has proven to be an effective way of extracting significant characteristics from climate data, particularly from monthly pressure field variability data at sea level. By the total negentropy and mutual information it was proved that the HOSVD method worked for the geographic analysis (Atl and Eur-Atl) and for the temporal analysis in winter (in both regions) and for summer only in the Eur-Atl region. The total negentropy is greater in the larger domain (Eur-Atl) than in the smaller domain (Atl), regardless of period. This is intuitive, as a larger domain allows to cover a greater amplitude of the ends of the SLP field. The total negentropy in winter is slightly higher than in summer, this is due to higher extreme values of PCs (in terms of skewness and e-Kurtosis), and therefore to more extreme situations in winter than in summer, in which concerns the SLP anomaly field. |
|---|---|
| Autores principais: | José , Ana Patrícia Assomar |
| Assunto: | Pressão ao nível do mar Análise de Componentes Principais Análise de Componentes Independentes Tese de mestrado - 2022 |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Lisboa |
| Idioma: | português |
| Origem: | Repositório da Universidade de Lisboa |
| Resumo: | The climate system is a dynamic system characterized by its high complexity and dimensionality, governed by non-linear interactions. Over the last few decades, the redefinition of the system’s dimensionality and the extraction of co-related patterns from measurements of atmospheric variables, present challenging tasks. An example is the monthly variability of the pressure field at sea level (SLP) which allows for evidence of certain weather time patterns, characterized by some recurrence (or repetition) and temporal persistence. In this regard, the positive and negative phases of the North Atlantic Oscillation, shows the multimodality of the field’s joint probability distribution. This multivariate variability, whether forced or internally induced, can be separated, using certain multivariate statistical techniques into scalar random components, with as much statistical independence as possible, called independent components. In case of fields having multivariate Gaussian distributions, Independent Component Analysis (ICA) corresponds to Principal Component Analysis (PCA), which is then performed sub-optimally in the non-Gaussian case as it occurs with the geopotential field at the sea level.In this dissertation we intend to apply the SLP field research carried out by NCAR (National Center for Atmospheric Research) reanalysis, wich was provided in a regular grid with a resolution of 2,5º x 2,5º (latitude, longitude) covering the period between January 1948 to December 2017 (70 years). In this dissertation one aims to apply the ICA to the SLP field, obtained from the NCAR reanalysis, thus contributing to the search for independent components within the atmospheric and oceanic phenomena, so that it is possible to deepen the understanding of the fundamental processes behind its variability.As such, a geographic analysis covering the North Atlantic (Atl) and Euro-Atlantic (Eur-Atl)areas was carried out, including a temporal analysis (for both areas), each one including two main climatological seasons: winter (consisting of December, January and February) and summer (consisting of June, July and August). The applied method consists of two phases, in wich the PCA (Principal Component Analysis) is applied in the first and the ICA (Independent Components Analysis) is applied to the PCA result in the second. ICA has proven to be an effective way of extracting significant characteristics from climate data, particularly from monthly pressure field variability data at sea level. By the total negentropy and mutual information it was proved that the HOSVD method worked for the geographic analysis (Atl and Eur-Atl) and for the temporal analysis in winter (in both regions) and for summer only in the Eur-Atl region. The total negentropy is greater in the larger domain (Eur-Atl) than in the smaller domain (Atl), regardless of period. This is intuitive, as a larger domain allows to cover a greater amplitude of the ends of the SLP field. The total negentropy in winter is slightly higher than in summer, this is due to higher extreme values of PCs (in terms of skewness and e-Kurtosis), and therefore to more extreme situations in winter than in summer, in which concerns the SLP anomaly field. |
|---|