Optimized Neural Network Prediction Model for Potential Evapotranspiration Utilizing Ensemble Procedure

Abstract

Potential evapotranspiration (ETo) is an essential hydrologic parameter for having better understanding for hydrologic cycle in certain catchment area. In addition, ETo is vital for calculating the agricultural demand. In fact, Penman-Monteith (PM) method is considered as reference method for estimating (ETo), however, this method required a lot of data to be used which is not usually available in many catchment areas. Furthermore, there are several efforts that have been performed as competitor to reach accurate estimation of (ETo) when there is lack of data to utilize (PM) method, but still required numerous research. Recently, methods based on Artificial Intelligence (AI) have been suggested to provide reliable prediction model for several application in engineering and especially for hydrological process. However, time series prediction based on Artificial Neural Network (ANN) learning algorithms is fundamentally difficult and faces problem. One of the major shortcomings is that the ANN model experiences over-fitting problem during training session and also occurs when a neural network loses its generalization. In this research a modification for the classical Multi Layer Preceptron- Artificial Neural Network (MLP-ANN) modeling namely; Ensemble Neural Network (ENN) is proposed and applied for predicting daily ETo. The proposed model applied at two different region with two different climatic conditions, Rasht city located north part of Iran and Johor Bahru City, Johor, Malaysia using maximum and minimum daily temperature collected from 1975 to 2005. The result showed that the ENN outperformed the classical MLP-ANN method and successfully predict daily ETo utilizing maximum and minimum temperature only with satisfactory level of accuracy. In addition, the proposed model could achieve accuracy level better than the traditional competitor methods for ETo.