Análise das propriedades multifractais das séries climáticas no Brasil

Abstract

Research on the dynamics of climate variables can provide important information about their spatio-temporal variability. The understanding of these processes is fundamental for the development of predictive climate models. Climatic variables depend on a variety of natural processes and show random fluctuations at different time and space scales, and linear statistical methods are insufficient for a full time series analysis. Thus, in order to understand the temporal complexity of climatic elements (air temperature (maximum, minimum, average), mean air humidity and mean wind velocity), we used the multifractal detrended fluctuation analysis (MF-DFA) to analyze the multifractal properties of historical series of these variables, provided by the National Institute of Meteorology (INMET), for 265 meteorological stations distributed in Brazil. The results indicate that the process that generates the variability of these climatic variables follows the multifractal dynamics, with greater influence of the seasonal component on the series. It was observed a persistent autocorrelation only in the variables average, maximum and minimum temperature, and this persistence is stronger in the proximity of the line of the Equator. The degree of multifractality indicated by the width of the multifractal spectrum varies between regions, since, for the variables average and minimum temperature, a spatial pattern was observed in which the regions near the equator line presented a lower degree. Regarding the parameter of asymmetry, a homogeneous spatial distribution was observed, and for all variables studied multifractality is more influenced by the small fluctuations, indicating that the temporal component with the greatest influence on these variables was the seasonality. In addition, we investigated the cause of multifractality by analyzing the randomized series. For most meteorological variables studied, multifractality is due to the probability density function and/or to the joint action of the probability density function and long-range correlations. The results indicate that the modeling of long memory in time series of climatic variables should be done through a multifractal model and may contribute to the development of more reliable predictive models.