Capability of hydrodynamic model to achieve sediment transport and deposition dynamics for large flood events: a parameter sensitivity analysis for a tailings dam break real event

Hydromorphological processes are fundamental in shaping water bodies, however, predicting erosion, sediment transport and deposition, and their impacts, becomes particularly challenging under extreme conditions. High-magnitude events of anthropogenic or natural origin can trigger the release of massive sediment concentrations into riverine systems, disrupting their equilibrium and driving significant morphological changes on local, regional, short, and long-term scales. Notable examples include debris flows, mass movements, extreme floods, outburst floods, and dam breaks. Among these, tailings dam failures are particularly relevant due to the severe environmental risks they pose in mining regions and the amount of sediment they release. It is evidenced by recent catastrophic events worldwide (Mount Polley in 2014, in Canada in; Mariana in 2015 and Brumadinho in 2019, in Brazil), that caused severe and long-lasting damage to ecosystems, river systems, and communities. In this context, a significant scientific gap remains in understanding how numerical sediment transport models perform under such conditions. Most sediment transport equations are empirical and were originally developed for natural river ecosystems. They were not built for the hyper-concentrated flows associated with catastrophic flood events, like extreme floods and tailings dam breaks. Consequently, assessing the efficacy and sensitivity of hydrodynamic models coupled with sediment transport for extreme events becomes an important step to evaluate viable tools for impact assessment and prognostic evaluations. In this study, we evaluated a physically-based 2D hydrodynamic model (HEC-RAS 2D) to simulate the sediment dynamics of the 2019 Brumadinho dam break in Brazil. During this event, approximately 9.8 Mm³ of material was released in just 5 minutes. The tailings wave, consisting of 45% sediment particles, propagated downstream in a small subcatchment (Ferro-Carvão stream) for 20 km over approximately 1.5 hours until reaching a major river. It is estimated that nearly half of the released volume was retained in the Ferro-Carvão floodplain through depositional processes, flooding 2.7 km² and significantly reshaping the local morphology. A global sensitivity analysis was performed using a Monte Carlo framework, generating 630 model runs that varied four key sediment-related input parameters: total sediment load, grain size distribution, sediment specific gravity, and model adaptation length. The methodology integrated validation with observed field data, sediment mass balance, and non-parametric statistical tests (Kruskal–Wallis) to assess parameter significance across outputs such as bed change, sediment mass outflow, and volume in the model’s domain. Results show that sediment load strongly influenced all outputs (bed change, sediment outflow, and volume outflow), while specific gravity and adaptation length had moderate effects, particularly on depositional patterns. Grain size showed unexpectedly low sensitivity. Validation results demonstrated model reliability in simulating the case study, with relative errors ranging from 8.1% to 10.1% and accuracy rates of 90–92% for bed change, sediment outflow, and volume outflow. The simulation effectively reproduced spatial sediment dynamics, with depositional zones matching field observations. However, localized overestimations of deposition highlighted limitations in capturing specific erosional processes. These findings underscore the importance of sensitivity analysis for robust calibration and provide critical insights into the uncertainties of simulating morphological changes driven by extreme events.