Predicting the effective diffusivity across the sediment–water interface in rivers

Hyporheic exchange directly controls and regulates the transport of nutrients, heat, and organic matter across the sediment–water interface (SWI), thereby affecting the biochemical processes in rivers, which is critical for maintaining the health of aquatic ecosystems. The interface exchange is controlled by multiple processes, including physical, chemical, and biological processes, which can be modeled by the effective diffusion model using an effective diffusion coefficient, Deff, to quantify the hyporheic exchange rate. In this study, genetic programming (GP), a machine learning (ML) technique based on natural selection, is adopted to search for a robust relationship between the effective diffusion coefficient and surface flow conditions, bedforms, and sediment characteristics on the basis of published broad interfacial mass exchange flux measurements. By utilizing a data set covering a wide range of environmental condition parameters, the effective diffusion coefficient prediction models for the SWI with and without bedforms are developed. Results show that the dimensionless effective diffusion coefficient is not only related to the permeability Reynolds number, ReK, but also to the channel Reynolds number, Re. Compared with the flat bed, ReK has a greater effect on the hyporheic exchange when bedforms present at the SWI by affecting the pumping advection strength. The new Deff predictor with a relatively concise form exhibits considerable improvements with regard to prediction ability and is physically sound relative to the existing predictors. •A genetic programming-based effective diffusion coefficient prediction model is established.•The channel Reynolds number and permeability Reynolds number together characterize the mass transport across the SWI.•The depth-averaged velocity and friction velocity together reflect fluid flow conditions within the hyporheic zone.