Prediction of Evaporation Potential through Data De-noising in Tabriz Synoptic Station

Potential evaporation is a component of the water cycle in nature and its prediction is a complicated and nonlinear practice. In this regard, the purpose of the present study was to provide the time-series prediction model of daily evaporation potential of Tabriz station using the two approaches of neural network and neural network- wavelet through de-noising. Daily time-series data of pan evaporation in Tabriz station consisted of 4309 days for the period of 1992-2011 were considered as the data base for running the above-mentioned models. Neural network prediction model was routed based on the three time series with 4, 7 and 10 days lag time of the normalized original signal. In the second approach, the main time series signal using Meyer wavelet was decomposed to 12 levels and the highest-frequency signal was removed as noise from the time series. Then, Neural network-wavelet model was implemented based on 36 time series with 4, 7 and 10 days delays. The evaluation of the results of these models by statistical and graphical criteria, indicated following A structure of 3-10-1 with correlation coefficient of 0.80 and mean square error of 0.125, and another structure of 36-8-1 with the correlation coefficient 0.917 and mean square error of 0.0858 were known as suitable structures in neural network and neural network-wavelet approaches, respectively.