Evaluate the Performance of SVR and GEP Models in Predicting the Monthly Fluctuations in Water Level of Urmia Lake

Prediction of lake level fluctuations is one of the most important issues in water resources planning and management. In recent years¡ the significant decline in water level of Urmia Lake had detrimental environmental impacts on this region. In this study¡ the performance of genetic expression programming and support vector regression models for predicting Urmia Lake water level was evaluated based on six different patterns during 1976-2009 to determine the best input pattern. The historical data of water level were used in four patterns¡ and precipitation¡ evaporation¡ seepage and water level were used in two other patterns. The results showed that the genetic expression programming had better performance than SVR model and the model performance improves with increasing input for model training. Also¡ at the best pattern the mean square error and coefficient of determination were calculated 0.08 m and 0.99 for GEP model and 0.60 m and 0.92 for SVM¡ respectively.