Construção de um Modelo Preditivo de Longo Prazo: previsão da temperatura atmosférica máxima média em Campos dos Goytacazes utilizando Redes Neurais Artificiais

previsão da temperatura atmosférica máxima média em Campos dos Goytacazes utilizando Redes Neurais Artificiais

Abstract

The increasing complexity of climatic phenomena has challenged traditional methods of weather forecasting, needing the optimization of conventional techniques and the development of approaches that keep pace with the technological advancements in the State of The Art. In this context, it is argued that artificial neural networks (ANNs) hold enormous potential as tools for long-term climate prediction, as they can be applied to capture complex and non-linear patterns present in meteorological data. Therefore, this article proposes the construction of a climate prediction model capable of forecasting the average maximum temperature for the coming decades. To achieve this objective, the paper discusses the preprocessing process, which involves the development of a linear regression model for data imputation of missing values, as well as analyses of linearity, normality, homoscedasticity, and outliers and leverage points to assess the quality of the treatment. Additionally, the work describes the architecture and topology of the constructed neural network, along with the calculation of uncertainty measures and performance metrics to assess the level of model performance and ensure the quality of the forecast conducted.