Prediction of daily suspended sediment load using the Genetic Expression Programming and Artificial Neural Network models

Because of the quantitative and qualitative problems of Daily Suspended Sediment Load (SSL) data with direct measurement, it is important to use methods for predicting it in watersheds. In this research, two methods consisting of the artificial neural network (ANN) and Genetic Expression Programming (GEP) were used to predict SSL. The studied area was a watershed in north of Iran. Input data included instantaneous flow discharge (Q), average daily flow discharge (Qi), average daily precipitation (Pi) and the output was SSL. A clustering method was used to homogenize data for the self-organizing map (SOM) method and then, all data were divided into three groups including 70, 15 and 15% for training, validating and testing, respectively. Also, the gamma test method was used to determine the best combination of input variables. In all combinations of inputs to the ANN and GEP models, the ANN model with tangent sigmoid activation function and input variables combination including Q, Qi, Qi-2, Qi-3, Pi, Pi-2, Pi-3 was the best for estimating SSL in the area with a root mean square error of 1995.3 (ton day -1 ) and the Nash-Sutcliff efficiency of 0.96. In general, the results of this study showed that intelligent models are capable of accurately estimating the SSL value. Also, using SOM preprocessing techniques and gamma tests increased the generalization power of the models. We also found that choosing the most influential variables and their best combination increased the modeling power and accuracy of SSL estimation, respectively.