Modelos decomposição temporal e de regressão harmônica: Uma comparação para a série mensal da temperatura mínima e máxima média de Presidente Prudente (SP)

Abstract

In this work two regression models, constructed to estimate the monthly minimum and maximum average temperature (Tmin and Tmax) series at Presidente Prudente (SP) city are presented. The temporal decomposition model with linear trend and seasonality and the harmonic model with linear trend and harmonic components were built for periods of 50 years (1961-2010) and of 30 years (1981-2010). The models were compared by accuracy measures on the 50 and 30 years series. The precision and performance measurements were evaluated for the validation data set (2011-2014). It was proposed the Spearman correlation coefficient instead of the Pearson coefficient in the case of non-normality of the original observations and model estimates, and to assess the accuracy and the performance of the models through the CS index. It was observed that the harmonic model showed better results than the temporal decomposition model, especially when it was built using the longer Tmax (50 years) and shorter Tmin series (30 years).