Enhancing sediment transport model reliability through Sensitivity Analysis: A Case Study in the Po River

The limited observational data on river bedload presents a significant challenge in understanding sediment transport processes. However, with recent advancements in computing capability, availability of remotely sensed data, and smart sensors, it is nowadays possible to model these transport processes in river networks at catchment scale. Nevertheless, the results of these models are often not robust due to inherited uncertainty and the stochastic nature of the input parameters. To manage these uncertainties and improve the robustness of model outputs, sensitivity analysis plays a crucial role. Sensitivity analysis is a method to study how changes in a numerical model's input factors contribute to variations in its output.

This project aims to apply Global Sensitivity Analysis (GSA) techniques to the D-CASCADE (Dynamic CAtchment Sediment Connectivity And Delivery) model, for the Po River network in Italy. D-CASCADE is a network-based (or graph-based) model that simulates material movement as distinct transport processes at the reach scale, or ‘cascades,’ defined by their provenance, sediment volume, and interactions downstream, at daily timestep.

To conduct the GSA, we use the SAFE toolbox, supporting both the generation of 5,000 random input factor combinations within defined ranges and distributions, as well as the quantification of the impact of each input factor's variation on the output.

In this work, we focus on the sensitivity estimation of active channel widths and riverbed slopes for every reach of the simulated network. These two input factors are key drivers of the transport capacity and the consequent sediment fluxes generated for the various sediment transport formulas implemented in D-CASCADE. The active transport width (the portion of the channel where bedload transport is active for a specific discharge) is largely unknown, even in data-rich contexts. Hydraulic slopes are also often unknown and generally replaced with topographic slopes which are largely dependent on the quality of the DEM used.  Active widths and slopes are then structurally inherently uncertain although they drive the model results. Through GSA, we evaluate how simultaneous random changes in these two input factors affect the simulated sediment fluxes and budgets. Results are analyzed both at the reach scale (sensitivity to local parameters) and the network scale (sensitivity to upstream parameters).

The presented methodology allows us to obtain important information about the effects of structural uncertainties in sediment transport modelling at network scale. These findings provide a foundation for enhancing the model's accuracy and resolving uncertainty in sediment transport prediction.