Uma experiência com modelo estatístico (MOS) para a previsão da temperatura mínima diária do ar

Nenhuma Miniatura disponível






Curso de graduação

Título da Revista

ISSN da Revista

Título de Volume




Direito de acesso


A MOS (Model Output Statistics) multiple regression equation for the prediction of daily minimum air temperature at the city of Bauru, in São Paulo State, is developed. The multiple regression equation, obtained using stepwise regression analysis, has four predictors, three from the CPTEC (Centre of Weather Forecast and Climate Studies) global model and one from observational data of the meteorological station at IPMet (Institute of Meteorological Research), Bauru. The predictors are the model 24 hours prognosis, valid at 00:00GMT, of 1000hPa temperature, 850hPa meridional wind and 1000hPa relative humidity, and the 18:00GMT observation of temperature. These four predictors account for approximately 80 percent of the total variance of the predictand, with a root mean square error of 1.4°C, i.e., approximately half of the standard deviation of daily mininum temperature observed at the IPMet station. A verification of the MOS equation with an independent sample of 47 cases shows that the forecast value is not significantly deteriorated when the observational predictor is not considered. The MOS equation, with or without this predictor, produces forecast with absolute errors smaller than 7.5°C in 70 percent of the cases studied. This result encourages the use of the MOS technique for operational daily minimum air temperature forecasting and the development of this technique for other weather elements and other localities.




Como citar

Revista Brasileira de Geofisica, v. 18, n. 1, p. 1-11, 2000.

Itens relacionados