Least squares support vector machines model for predicting reference evapotranspiration based on weather forecasts

A reference evapotranspiration(ET₀) prediction model is developed based on the least squares support vector machines.Weather forecasts are used for ET₀ predictions.The model can be trained with daily weather parameters including quantified weather types and wind grades,etc.Different combinations of daily weather parameters can be tested in the model training processes.In this study,the daily weather parameters are obtained from the Guangli irrigation district during the period 1997-2007.The 1997-2006 data are used for the model training,and a total of 10 daily ET₀ forecasting schemes are established.Predictions of daily ET₀ using each of the 10 schemes are validated with the observations from 2007.The results show that the scheme using the air temperature and the quantified weather types and wind grades as model predictors is able to give the best model performance; and the corresponding statistics are the root mean square error(E_RMS) of 0.5182,the relative error(E_R) of 0.1878,the coefficient of determination(R²) of 0.8648,the I_A of 0.9669,and the regression coefficient(R_C) of 0.9868.Acceptable daily ET₀ predictions are also obtained with other 6 schemes,among which the simplest scheme using only the air temperature as the model predictor can also produce fair results as revealed by E_RMS=0.6576,E_R =0.2332,R²=0.9866,I_A=0.7747 and R_C=0.9680.The latter scheme shows a strong potential in practical applications.