Deep learning and generative models to predict Manning resistance

The resistance coefficient is a key parameter in free-surface flow modelling, influencing velocity distribution, drainage capacity and energy dissipation. Traditionally derived from Manning’s and Chézy’s equations, its calibration in urban environments is challenging due to the combined effects of viscosity, geometry, wall conditions and slope. This study compares deep neural networks and conditional generative adversarial networks for predicting the Manning coefficient using discharge, kinematic viscosity, surface roughness, slope, and filling ratio. The deep neural network achieved higher accuracy, with a mean relative error of 1.27% and a coefficient of determination of 0.9753, compared to 2.26% error and 0.9159 for the conditional generative adversarial network. Only 4.34% of deep neural network predictions exceeded a 5% error tolerance, compared to 15.8% for the conditional generative adversarial network. Both models demonstrated reliable uncertainty quantification and generalization, highlighting viscosity and roughness as dominant factors shaping hydraulic resistance behaviour.