Modelos bayesianos para previsão de chuvas no Ceará

• Ano de defesa: 2023
• Autor(a) principal: Silva, Davi Sousa e
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: LASSO; Modelos Lineares Generalizados (MLGs); t-Student; Análise de séries temporais; Monte Carlo via cadeias de Markov (MCMC)
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/70940

The main objective of this work was to open a discussion on rainfall forecast models for Ceará and its macro-regions. For that, Bayesian models were analyzed, combining Generalized Linear Models (GLMs) and t-Student models with time series models. The proposed approach takes into account not only the historical series, but also dozens of covariates that, according to meteorological studies, are related to the rainfall phenomenon in Ceará. The proposed models use the LASSO algorithm for automatic model selection, in which the variables were selected through the AIC and SBC criteria. Thereafter, generalized linear models and t-Student models were fitted. ARIMA and SARIMA time series models were also fitted. The posterior estimates were obtained through stochastic simulation methods MCMC.