Machine Learning Approaches for Estimating Roughness Coefficients in Two-Stage Sinuous Channels

Estimating the flow rate of streams during floods is rarely successful due to a lack of dependable data from the field in natural areas. Predicting discharge depends greatly on the value of Manning’s roughness coefficient. The study uses Artificial Neural Networks (ANN), Gene Expression Programming (GEP), Extreme Learning Machines (ELM) and Gaussian Process Regression (GPR) as AI techniques to determine Manning’s roughness coefficient in two-stage sinuous channels. Relative width, depth ratio, sinuosity ratio, channel bed slope and sinuous belt width ratio are the input variables used for the model. They were taken into account to determine the roughness adjustment factor. Several experiments were run to test how different parameters affect roughness prediction. Results of the developed AI models were compared to those of established analytical methods using various statistical tricks. During the testing phase, the highest accuracy was achieved by GPR and ANN (R² = 0.82). After that, GEP reached R² = 0.79 and ELM R² = 0.65. Predictions of Manning's n using the GPR method are quite accurate. However, the GEP model's ability to generate a generalized model equation for Manning's n makes it particularly valuable. Based on the above, the GEP model’s expression was applied to estimate Manning’s n for measurements of a flood from the Baitarani River in India. It was confirmed that the proposed models can work in experimentally small-scale environments and real rivers.