Forecasting and Sensitivity Analysis of Monthly Evaporation from Siah Bisheh Dam Reservoir using Artificial neural Networks combined with Genetic Algorithm

Abstract

Evaporation process, the main component of the water cycle in nature, is essential in agricultural studies, hydrology and meteorology, the operation of reservoirs, irrigation and drainage systems, irrigation scheduling and management of water resources. Various methods have been presented for estimating evaporation from free surface including water budget method, evaporation from pan and experimental equations that each of these methods is coupled with the restriction and measurement error. Early the new technique using Artificial Neural Networks (ANNs) based on artificial intelligence has been widely used in various scientific fields, particularly water engineering. In this study, the amount of monthly evaporation from the Siah Bisheh dam reservoir was forecasted up 3 next month using Multi-Layer Perceptron (MLP), Radial Basis Function (RBF) and Feed Forward (FF), of ANNs. The genetic algorithm was used for efficient input variables selection and number of neurons in hidden layer of ANNs. The results showed that the correlation coefficient between measured and computed outputs using RBF, MLP and FF models were 0.92%, 0.90% and 0.88% respectively in the estimation and forecasting of evaporation from the dam reservoir. Therefore the RBF model had more precision rather than MLP and FF models in the estimation and forecasting of monthly evaporation. The results of sensitivity analysis showed that the monthly evaporation from the dam reservoir up 3 next month had most sensitivity to the time of evaporation per month, air pressure on ground surface in 2, 3 and 1 months ago, wind speed on 1000mb pressure in 3 and 2 months ago and air temperature on 300mb pressure in current time respectively.