Modelos prognósticos da temperatura mínima média numa região homogênea do Rio Grande do Sul

Abstract

This work aims to study the minima mean monthly temperatures behave in RS state, Brazil, in 90 years (1913 – 2002) of observation. The data were cllected in 40 meteorological station distributed in the state, and the faults were completed by the correlation method. On the first step, five regions were determinated by “Cluster Analysis”, these regions present similar temperatures designated “homogeneous regions”. In this study it has been evidenced an increasing of temperature east-west against a gradient of altitude on the north part of the state. The influence of the highest altitude on north is evident. So, it is noticeable the similarity among quarterly minimum medium temperatures in a large area, evidencing the spatial homogeneity of this variability. After this study, the homogeneous region which is included the meteorological station of Pelotas was detached. It was determinated the quarterly medium temperatures of this region; the result was four series with ninety data each, one for each year of observation. These series were submitted a Spectral Analysis, based on the “Discrete Fourier Transform”, with the intention of identifing extents occasionally in these temporal series. In each series, has been used 84 years, 1913 to 1996, to determinate the model that describes the behavior of the variable and which is used at the same time for projection to the future. The six years remaining, 1997 to 2002, was used for forecasting. The periodogram, estimator of the spectrum evidenced quarterly some detached peak, indicating the most important frequencies selected to compose the models. In this way, the estimated models were constituted by: three waves with periods of 28,0; 6,5 and 3,5 years, for the first quarterly, for waves for the second quarterly with periods of 6,5; 5,6; 4,2 and 3,4 years, six waves with periods of 9,3; 7,6; 6,5; 4,7; 3,8 and 3,5 years, for the third quarterly and, finally, five waves for the fourth quarterly, with the same period by 84,0; 16,8; 12,0; 9,3 and 6,5 years. It was noticed by the autocorrelation of the residues that the established models were valid to represent the data and can be used to forecast. It was also used a smoothing model, “Brown’s linear exponential smoothing method”, that was equally useful for forecasting.