Assessment of the effectiveness of statistical forecasting models using data envelopment analysis

Statistical hydrological forecasting models have been widely used for the medium and long term hydrological forecasting.One major concern for these models is how to choose an appropriate one with less imput and higher precision.The Charnes,Cooper and Rhodes(CCR)model for data envelopment analysis is used here for the multi-attributes evaluation of these forecasting models.In the CCR model, the inputs include indexes for forecasting factors and model parameters,the outputs include precision indexes,and the evaluation result is the relative efficiency of each model compared with other models.This method is used to evaluate 20 models of radial basis function networks for forecasting reference evapotranspiration.Results show that the CCR model is feasible for this kind of multi-attributes evaluation of forecasting models.Key factors for forecasting reference evapotranspiration are identified,which are maxmium and minmium temperature,wind speed and sunshine hours.Moreover,the forecasting precision and relative efficiency of a network could be miproved significantly if the date was added as an input of the network.