estimating daily suspended sediment by intelligent and traditional models (case study: kasalian and rood zard watersheds, iran)

Suspended sediment load is an indicator of erosion in watersheds. therefore, accurately estimating the daily suspended sediment load (dssl) is an important issue in river engineering. in this research, artificial neural networks (ann), genetic expression programming (gep) intelligent models, and the traditional sediment rating curve (src) model were used to estimate dssl in the kasilian and rood zard watersheds in iran. the input data to the models included instantaneous flow discharge (q), average daily flow discharge (qi), average daily flow discharge with a delay of three days (qi-1,qi-2,qi-3), average daily precipitation (pi), and average daily precipitation with a delay of three days (pi-1,pi-2,pi-3); the output data was dssl. in this research, the self-organizing map (som) artificial neural network was used for data clustering, and gamma test (gt) methods were used to obtain the best combination of input variables to intelligent models. the results showed that the best models for estimating dssl in the kasilian and rood zard watersheds were respectively the ann model with an activation function of tangent sigmoid with the best combination of input variables (qi-1,qi-2,qi-3,pi,pi-1,pi-2,pi-3) and the gep model with the input variables qi,qi-1,qi-2,pi,pi-1,pi-2,pi-3. the statistical values ​​of the ann-tangent sigmoid model for the kasilian watershed were mae=231.4 (ton day-1), rmse=578.6 (ton day-1), nse =0.98, and r2=0.98; these values for the gep model in the rood zard watershed were mae=475.7 (ton day-1), rmse=1671.9 (ton day-1), nse=0.99, and r2=0.99. the src model in the kasilian watershed with r2=0.34 and nse=0.08 and the rood zard watershed with r2=0.59 and nse=-0.11 showed the low accuracy of this model in estimating dssl.