Análise e refinamento estatístico dos prognósticos do vento com o modelo WRF no estado de Alagoas

Abstract

The use of numerical modeling in wind forecasting is an important aspect in various activities of society, which promotes a series of economic benefits in several areas. Faced with the ever-increasing need for more reliable predictions, this work proposes a statistical analysis of WRF mesoscale model performance for the wind forecast in Alagoas State during the year of 2008. In addition to determining the accuracy of the wind speed forecast, it was also carried out a comparative analysis of model estimates with wind profile extrapolations to the levels of 30 and 50m. To mitigate the influence of errors inherent to atmospheric modeling was also performed a statistical post-processing of the simulated test data (for 72 hours). In validation was used data from four anemometric stations: Água Branca (Backwoods “Sertão”), Girau do Ponciano (Rural “Agreste”), Maragogi and Roteiro (Coast). For forecast analysis statistical indices were used: "r" (Pearson correlation), BIAS, MAE and RMSE. The results showed that errors of the zonal wind component are worse for coastal sites, and that it has better correlations with the meridional component in the dry and rainy period. The average error (BIAS) varied between -0.02 m.s-1 and -1.18 m.s-1, with performance ranked as good at Girau do Ponciano and great for the other sites. The MAE and RMSE errors showed satisfactory values. MAE varied between 1.58 m.s-11 and 1.88 m.s-1 and RMSE between 2.03 m.s-1 and 2.31 m.s-1. The extrapolation of the wind was carried out by logarithmic law and power law, the first obtained the worst results for all sites and two vertical levels. Extrapolation by power law was superior in Maragogi to the level of 50m, in other cases the model output was better than the two methods. The statistical adjustment considered the systematic error of WRF, estimated by multiple linear correlations and was applied in one of the experimental sites. The adjustment results were satisfactory, with a reduction of BIAS to less than 1 m.s-1 and gain of around 40% in the MAE and RMSE errors in the levels of 30 and 50m.