This paper applies a hybrid model of wind speed forecasting in short-term to the watershed basin of Paranapanema, Brazil, as strategy to decrease computational demand typically observed in exclusively WRF-based predictions, while deals with an also common lack of measured atmospheric variables in greater spatial and time frame resolution. The model uses adjusted variables from real data simulated WRF outputs for the target area as input of a MultiLayer Perceptron (MLP) Neural Network (ANN) configured with Feed-forward Backpropagation algorithm, tested with different combinations of parameters. The association here proposed aims to match the best of both methods to mitigate each other's typical issues and provide, supported by future works, even better accurate results also for other atmospheric elements.

To assist the planning of the electrical system with reliable information on the temporal variability of solar irradiation and consequently efficient use of this source, the present work aims to estimate and map solar radiation in the micro-region of Barra, located in the Vale São Francisco of the Bahia, using the numerical model WRF-Solar. The period of the simulations was the months of January and July 2019, a period in which the days of perihelion (3th January) and aphelion (4th July) occur. The simulations were satisfactory through statistical analysis between simulated and observed data, with a normalized quadratic error of 0.08 and 0.13, a factor of two of 0.81 and 0.76, and a correlation coefficient of 0.90 and 0.76, respectively, for the months of January and July. Therefore, the computational tool has good forecasting capabilities, with great potential for operational, research, and technological development purposes.

Para estudos mais exatos do potencial energético de um determinado local, o conhecimento da variabilidade temporal da radiação solar é recomendável para obter projetos com melhor aproveitamento da energia solar. Dentro desse contexto, este trabalho avalia as previsões de irradiação e temperatura feitas pelo modelo numérico de previsão meteorológica WRF. A avaliação é realizada em uma escala temporal de 36 horas e os dados previstos pelo modelo são comparados com dados coletados pela estação automática do INMET localizada no município da Barra, situado no Estado da Bahia. Os resultados demonstram que a ferramenta apresenta boa capacidade de previsão, com grande potencial para fins operacionais, de pesquisa e desenvolvimento tecnológico.

No abstract