Prediction of sand rivers bed form using decision trees

Abstract


Background and objectives: The bed forms or in the other word, bed irregularities are structures that form due to stream flow and they have direct impact on roughness and flow resistance in sand bed rivers. Bed forms have different shapes and forms in sand bed rivers. Since river discharge and flow velocity are totally affected by roughness, accurate prediction of the shape of the bed is of great importance. Due to the influence of different parameters in the formation of the river bed form, it is difficult to determine the governing equations and the mathematical models with sufficient precise. Today, the use of artificial intelligence systems has expanded as a new way of analyzing water resources issues. The objective of this research is to introduce a method that can be used to predict the shape of the river bed with high precision.
Materials and methods: In the present study, the data were randomly divided into two parts of the training (70%) including 1647 laboratory data and test (30%) containing 560 laboratory data. The decision trees were coded on the data of the test section in the WEKA programming environment, and finally calibration was performed on the data by using Random Forest and Random Tree algorithms. Then the experimental methods of Van Rijn, Engelund and Hansen and Simons and Richardson were implemented on test data.
Results: Evaluation of the results were done using root mean square error (RMSE), Correctly Classified Instances and Roc area under curve. The results showed that the best performance reached by Random Forest algorithm for experimental data with statistical criteria of CCI=85 (%), RMSE=0.17, ROC=0.97. On the other hand, by examining the results of empirical methods, it was determined that for laboratory data, van Rijn method has better performance with the results of CCI=64 (%), RMSE=1.07. Between different variables, discharge, width, depth, slope of the channel, average diameter of sediment particles and temperature for laboratory data were the most important parameters in predicting bed forms.
Conclusion: In this research, the superiority of soft computing models was evident compared to the empirical methods in modeling and predicting of the bed form. Models performed better in the WEKA environment.
Basically, because of the formation of the river bed form is depended on several factors, and also because of its complex nature, the prediction of this phenomenon is very difficult and sometimes with high errors. Since artificial intelligence methods are used to analyze issues that do not explicitly describe the nature of the problem, so many of the issues of bed form can be solved with these methods.