| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 5.42 MB | Adobe PDF |
Orientador(es)
Resumo(s)
Esta dissertação analisa a evolução da temperatura média mensal do mar no Atlântico entre janeiro de 1968 e junho 2023 em quatro regiões: Nazaré, Faro e São Miguel (Portugal) e Praia (Cabo Verde), nas profundidades de 1 m e 5 m, utilizando dados European Marine Observation and Data Network (EMODnet). Observou‑se um padrão sazonal anual bem definido em todas as localidades, bem como um aquecimento estatisticamente significativo, com aumentos entre 0,0196°C e 0,0234°C por ano, equivalentes a 0,20-0,23°C por década.
Praia apresenta temperaturas consistentemente mais elevadas e maior estabilidade térmica.
Foram aplicadas várias metodologias de séries temporais, incluindo decomposição clássica, Seasonal-Trend Decomposition, Holt‑Winters e modelos SARIMA. Os modelos SARIMA apresentaram melhor desempenho, captando de forma robusta a tendência e a sazonalidade, com erros de previsão reduzidos (MAE entre 0,59°C e 0,97°C; MAPE entre 2,94% e 4,97%).
As previsões para julho de 2023 a junho de 2024 reproduzem dequadamente o ciclo sazonal e alinham-se com os valores observados.
Para integrar a dimensão espacial, ajustou-se um modelo STARMA às séries a um metro de profundidade, baseado numa matriz de pesos construída a partir de distâncias geodésicas. O modelo final revelou dependência temporal dominante e influência espacial moderada, apresentando bom desempenho e resíduos próximos de ruído branco. A comparação entre os modelos SARIMA e STARMA revelou desempenhos preditivos globalmente semelhantes, com ligeira vantagem do modelo STARMA na maioria das localidades analisadas (Nazaré, Faro e Praia). As previsões STARMA mantiveram a hierarquia térmica entre regiões e reforçaram a coerência dos resultados univariados.
Globalmente, o estudo confirma o aquecimento progressivo das águas do Atlântico e demonstra que modelos SARIMA e STARMA são ferramentas eficazes para compreender e prever a variabilidade térmica regional, fornecendo suporte científico relevante para ações
de monitorização oceânica e adaptação às alterações climáticas.
This dissertation analyzes the evolution of the monthly mean sea temperature in the Atlantic Ocean between January 1968 and June 2023 in four regions: Nazaré, Faro and São Miguel (Portugal), and Praia (Cape Verde), at depths of 1 m and 5 m, using European Marine Observation and Data Network (EMODnet) data. A well‑defined annual seasonal pattern was observed in all locations, as well as a statistically significant warming trend, with increases ranging from 0.0196°C to 0.0234°C per year, equivalent to 0.20-0.23°C per decade. Praia exhibits consistently higher temperatures and greater thermal stability. Several time series methodologies were applied, including classical decomposition, Seasonal-Trend Decomposition, Holt‑Winters and SARIMA models. SARIMA models showed the best performance, robustly capturing both the trend and seasonality, with low forecasting errors (MAE between 0.59°C and 0.97°C; MAPE between 2.94% and 4.97%). Forecasts for the period from July 2023 to June 2024 adequately reproduced the seasonal cycle and aligned well with the observed values. To incorporate the spatial dimension, a STARMA model was fitted to the time series at one meter depth, based on a spatial weight matrix constructed from geodesic distances. The final model revealed dominant temporal dependence and moderate spatial influence, showing good performance and residuals close to white noise. A comparison between the SARIMA and STARMA models revealed broadly similar predictive performance, with a slight advantage of the STARMA model in most of the locations analysed (Nazaré, Faro, and Praia). STARMA forecasts preserved the thermal hierarchy among regions and reinforced the structural consistency of the univariate results. Overall, the study confirms the progressive warming of Atlantic waters and demonstrates that SARIMA and STARMA models are effective tools for understanding and predicting regional thermal variability, providing relevant scientific support for ocean monitoring and climate adaptation strategies.
This dissertation analyzes the evolution of the monthly mean sea temperature in the Atlantic Ocean between January 1968 and June 2023 in four regions: Nazaré, Faro and São Miguel (Portugal), and Praia (Cape Verde), at depths of 1 m and 5 m, using European Marine Observation and Data Network (EMODnet) data. A well‑defined annual seasonal pattern was observed in all locations, as well as a statistically significant warming trend, with increases ranging from 0.0196°C to 0.0234°C per year, equivalent to 0.20-0.23°C per decade. Praia exhibits consistently higher temperatures and greater thermal stability. Several time series methodologies were applied, including classical decomposition, Seasonal-Trend Decomposition, Holt‑Winters and SARIMA models. SARIMA models showed the best performance, robustly capturing both the trend and seasonality, with low forecasting errors (MAE between 0.59°C and 0.97°C; MAPE between 2.94% and 4.97%). Forecasts for the period from July 2023 to June 2024 adequately reproduced the seasonal cycle and aligned well with the observed values. To incorporate the spatial dimension, a STARMA model was fitted to the time series at one meter depth, based on a spatial weight matrix constructed from geodesic distances. The final model revealed dominant temporal dependence and moderate spatial influence, showing good performance and residuals close to white noise. A comparison between the SARIMA and STARMA models revealed broadly similar predictive performance, with a slight advantage of the STARMA model in most of the locations analysed (Nazaré, Faro, and Praia). STARMA forecasts preserved the thermal hierarchy among regions and reinforced the structural consistency of the univariate results. Overall, the study confirms the progressive warming of Atlantic waters and demonstrates that SARIMA and STARMA models are effective tools for understanding and predicting regional thermal variability, providing relevant scientific support for ocean monitoring and climate adaptation strategies.
Descrição
Tese de Mestrado em Estatística, Matemática e Computação, apresentada à Universidade Aberta
Palavras-chave
Temperatura do mar Modelos SARIMA Modelação Espaço-Temporal (STARMA) Alterações climáticas Previsão Oceano Atlântico Cabo Verde Portugal Sea temperature SARIMA models Spatio‑Temporal Modelling (STARMA) Atlantic Ocean Climate change Forecasting