Evaluation of effective parameters in the estimation of evaporation using artificial neural network model

Abstract: In recent years, Artificial Neural Networks have been applied as a powerful instrument to increase predictability capacity of linear and nonlinear relationships in complex engineering problems. Use of this tool box in different fields of civil engineering, agriculture, environment, and in particular hydrologic matters for a range of significant parameters with complex mathematical equations and variables have been addressed. This study is aimed at an estimation of evaporation using Qnet2000 artificial neural network for Ahvaz, Abadan, and Dezful meteorological stations located in Khouzestan province, Southwestern Iran. Accordingly, the model was respectively examined by 6 input parameters (minimum temperature, maximum temperature, minimum relative humidity, maximum relative humidity, sunny hours, wind velocity), 5 parameters (eliminating minimum relative humidity or maximum relative humidity), and 4 parameters (eliminating minimum relative humidity and maximum relative humidity). Results show that at Ahvaz, and Dezful stations maximum temperature, and at Abadan station first minimum temperature and then maximum temperature have had the most significant effect in evaporation estimation using the software. Also, regarding the best structure determination, results indicate that the best structure was gained for aforementioned stations respectively in following estates: it was gained in secant hyperbolic estate with 5 input parameters (without maximum relative humidity) and 6 knots in hidden layer and correlation coefficient (R2) 0.97, at Ahvaz station; it was gained in sigmoid estate with 6 input parameters (without maximum relative humidity) and 3 knots in hidden layer and correlation coefficient (R2) 0.99, at Abadan station; it was gained in tangent hyperbolic estate with 5 input parameters (without maximum relative humidity) and 9 knots in hidden layer and correlation coefficient (R2) 0.985, at Dezful station .