Estimation model for reference evapotranspiration by neural network based on principal components

In order to improve the performance of neural network model for the prediction of reference crop evapotranspiration, principal component analysis are applied to the weather data including the maximum, minimum and average daily temperature, sunshine duration, air pressure, humidity of exposure field, air relative humidity and wind velocity, and a three-layer BP(back-propagation) neural network model is constructed based on the principal components. Based on ten-day average weather data from 2001 to 2004 in the Water Conservancy Science Experimental Station in Chongchuan, the principal-component-based model was trained and predicted with Matlab neural network toolbox, and compared with tranditional BP neural network. Results show that the principal-component-based BP network model can well reflect the relationship between environmental factors and ET0, and is superior to the general BP network model in the prediction of ET0, especially for the validation samples outsides training dataset, which shows better performance of the principal component BP network model in comparison with the traditional BP network model.