Wavelet network model for reference crop evapotranspiration forecasting

Reference Crop Evapotranspiration(RCE) plays an important role in the agricultural production management. RCE characterized by its nonlinearity and multi-time scale feature, may vary with the change of time under the influence of stochastic variation of meteorological factors such as temperature, sunlight, wind speed and so on. In order to reveal the evolutionary law of RCE time series process, a forecasting model is established in this paper for RCE forecasting by combining the wavelet transform and neural networks. Firstly the RCE time series is decomposed to different frequency components with wavelet analysis. Then the artificial neural network is used in multi-scale forecasting of these coefficients. Finally, based on the formula reconstructed, the forecasted RCE time series is obtained. The effectiveness of this method is verified by an example. A comparison indicated that the error of the simulation adopting this model was smaller than that by using artificial neural network model or AR model only. The application of the wavelet network model for RCE forecasting is encouraging.