Dynamic linear modeling of homogenized monthly temperature in Lisbon

Abstract

This chapter focuses on the statistical modeling of the homogenized monthly average temperature data of Lisbon from 1856 to 2008. An exploratory analysis was performed using linear regression models which indicates the need of considering the temporal dependency and some flexibility in the trend modeling. In order to incorporate the properties of the data it was adopted a dynamic linear models with a fixed effect component. The model was fitted by a two-step procedure which combines the least squares method and the maximum likelihood estimation in the state space framework. The results indicated an average increase of the homogenized monthly temperature in Lisbon in about 0.427oC per century, between 1856 to 2008. Additionally, smoother predictions of the stochastic slopes indicated that the rise of temperature moderately changes according to the month, higher linear increases occurred in the winter months and lower increases occurred in the summer months.