Predicting submerged vegetation drag with a machine learning-based method

Abstract

Accurate estimation of the drag forces generated by vegetation stems is crucial for the comprehensive assessment of the impact of aquatic vegetation on hydrodynamic processes in aquatic environments. The coupling relationship between vegetation layer flow velocity and vegetation drag makes precise prediction of submerged vegetation drag forces particularly challenging. The present study utilized published data on submerged vegetation drag force measurements and employed a genetic programming (GP) algorithm, a machine learning technique, to establish the connection between submerged vegetation drag forces and flow and vegetation parameters. When using the bulk velocity, U, as the reference velocity scale to define the drag coefficient, C_d, and stem Reynolds number, the GP runs revealed that the drag coefficient of submerged vegetation is related to submergence ratio (H*), aspect ratio (d*), blockage ratio (ψ*), and vegetation density (λ). The relation between vegetation stem drag forces and flow velocity is implicitly embedded in the definition of C_d. Comparisons with experimental drag force measurements indicate that using the bulk velocity as the reference velocity, as opposed to using the vegetation layer average velocity, U_v, eliminates the need for complex iterative processes to estimate U_v and avoids introducing additional errors associated with U_v estimation. This approach significantly enhances the model’s predictive capabilities and results in a simpler and more user-friendly formula expression.