Are the source and sink behaviour simulated by a network-scale sediment transport model credible?

In this work, we employ Global Sensitivity Analysis (GSA) to quantify the robustness of sediment fluxes and sediment budgeting simulated at the river network scale. These simulations are subject to significant uncertainties when determining key drivers of transport capacity estimation, such as active transport width, river slope, or grain size distribution. These parameters, already difficult to estimate accurately at the scale of individual river cross-sections, become even more challenging to constrain at the network or entire river scale.  

We achieve this by applying the network scale sediment transport model, D-CASCADE (Dynamic CAtchment Sediment Connectivity And DElivery), to the 528 km long Po River basin in Northern Italy. D-CASCADE divides a river network into discrete units, termed ‘reaches’, and uses an empirical sediment transport formula to estimate how sediment is transported through this network. The river Po was divided into 35 reaches of homogenous geomorphic characteristics of a few kilometers each, and the model includes the input from 21 of its main tributaries.  We simulated the sediment transfer across this network for five hydrological years (2017-2021), with a daily time step.

Monte Carlo simulations were conducted using 1000 realizations, jointly perturbing the nominal value of 9 input parameters, such as active transport width, slope, roughness coefficient, and initial grain size distribution (GSD), for all the 35 reaches composing the Po River course. Specifically, the proposed methodology aims at assessing whether the plausible variations of the input parameters (that we established from field data and expert judgement) affect the nature of a river reach in the simulations, as sediment source (negative sediment budgeting), sink (positive sediment budgeting), or at equilibrium (sediment budget under a threshold), hence impacting its geomorphic behavior. The results indicated a strong robustness in classification, as no transition from source to sink was observed in any reach in the network. Nonetheless, some reaches shifted from being classified as either source or sink to

Regional Sensitivity Analysis (RSA) was then applied to identify which uncertain parameters have the greatest influence on such transitions. The RSA results showed that the active width, slope, roughness coefficient, and active layer depth are the primary input parameters affecting the river's state in gravel-dominated reaches, while the initial grain size distribution (GSD) is important in its sand-dominated reaches.

The presented approach applied to such network models allows for testing whether modelled river reaches maintain similar geomorphic behaviour despite the input uncertainties, while also identifying river network segments that exhibit an increased sensitivity and to which parameters. These insights can help prioritize efforts in data collection (known to be resource and time-demanding) and/or guide model calibration (known to be computationally expensive) towards parameters and locations whose improvements would most effectively reduce the final model uncertainty.