Uso da transformada em ondeletas para projeções de médio prazo das vazoes na bacia do rio São Francisco

Abstract

This work aims to create projection model future scenarios 3 to 10 years of streamflow using the wavelet transform in the São Francisco River Basin. We used the stremflow data provided by Operador Nacional do Sistema (ONS), divided into two periods: 1931-2006 and 2007-2016, for calibration and verification, respectively. The annual series is standardized and an analysis is made of global power spectrum to identify the most energetic bands. The time series is decomposed into three bands the residue and reconstructed whereas bands are orthogonal. It was made a auto regressive (AR) model for band and residue, with noise of the AR model obtained by the difference between the model results in the calibration period and the value of the band, and then used to generate scenarios. The projection is obtained by adding the auto regressive models. For performance evaluation is made a qualitative analysis of the accumulated probability distribution of projected years period and its verisimilitude. The model identified the distribution probabilities function of projected year and achieved verisimilitude greater than 1, indicating that this methodology can capture the variability medium term.