Estatísticas de cinco variáveis meteorológicas no município de Craíbas no Estado de Alagoas

Abstract

A statistical analysis of meteorological variables (MV) in northeastern Brazil (NEB) is of paramount importance for Brazilian sustainable industrial development, especially those related to clean and renewable energy. Therefore, this research aims to investigate five MV: Wind speed and direction (WS – m.s-1 and WD - °), air temperature (AT - °C), air relative humidity (RH - %) and atmospheric pressure (AP – mm Hg). The study location was in the municipality of Craíbas (geographic coordinates: 09°37’03,4=S, 36°47’40,2=W), semi-arid region (Agreste) in the central part of Alagoas. Two databases were used with a measurement interval of 10 minutes: i) At the reference height of 100m, the MV were evaluated in the total period of 14 months (March/2014 to April/2015) and seasonal periods (transition – PT, rainy – PR, dry - PD); ii) To study the vertical wind profile (VWP) a set of measurements at eight height levels (4, 10, 14, 19, 30, 50, 70 and 100m) was also used. The results showed that the MV in the PD and PT were more intense WS, predominant WD is SE, the highest (smallest) AT (RH) with the greatest tendency to increase (decrease) and the lowest AP. In PR, the variables presented opposite values with predominant WD is ESE. The Wavelet analysis of the MV showed a periodicity of 12 hours, daily and monthly, the spectral power in the PR is lower than in the PD and PT. The MV that were related were: AT and RH were negative and almost perfect (very high) in the total, PT and PD (PR) periods; AT and AP were negative and high in all periods; RH and AP were positive and moderate in both periods. This inverse relationship between AT and RH is stronger in PD and smaller in PR, between AT and AP it is, on the contrary, stronger in PR and smaller in PD. The coherence wavelet (CW) between the MV showed that the AT and RH relationship presented the highest degree of coherence between the pairs of variables with periodicity varying from hours to days, monthly and annual. The WD is the variable that showed the most irregular coherences and periods of phase change. The MV simulation was performed via artificial neural network (ANN): i) Nonlinear Autoregressive (NAR); ii) Nonlinear Autoregressive with eXogenous inputs (NARX); iii) Long and Short Term Memory (LSTM). To investigate the performance of the ANNs, a Taylor diagram and statistical metrics were used: mean error, root mean square error, mean absolute percentage error and Pearson correlation. The best simulation was performed by NAR, followed by NARX and LSTM. The AP, RH and AT variables were simulated more accurately than the WD and WS simulations. In the temporal evolution of WS at different levels, there is a mirroring of WS, gradually intensifying with increasing height, WS increases (decreases) with low (high) variability during the day (night). The lower levels are more turbulent and more susceptible to gusts than the upper levels. The power law coincided more with the mean total of the VWP than the logarithmic law, but both are within the VWP variability range and therefore perfectly applicable. Finally, these results can be used for wind energy and consequently, mitigate the effects of global warming, climate change, among other factors.