A benchmarking study of four numerical runout models for the simulation of tailings flows

Negar Ghahramani,H Joanna Chen,Daley Clohan,Shielan Liu,Marcelo Llano-Serna,Nahyan M Rana,Scott McDougall,Stephen G Evans,W Andy Take
DOI: https://doi.org/10.1016/j.scitotenv.2022.154245
2022-06-25
Abstract:Numerical runout models are important tools for predicting the potential downstream impacts of tailings dam breaches that generate tailings flows, which is a crucial step in emergency preparedness and planning, and risk management. Most existing runout models were originally developed for the analysis of water floods or flow-like landslides (e.g. debris flows). In this study, we back-analyze two well-documented historical tailings dam breaches (1985 Stava, Italy and 1994 Merriespruit, South Africa) using four numerical models (DAN3D, MADflow, FLO-2D and FLOW-3D). The main objective of this multi-case, multi-model benchmarking exercise is to identify collective opportunities to adapt these types of models and associated modelling methods to tailings dam breach runout applications. Comparing best-fit simulation results, we find that all four models are capable of reproducing the bulk behaviour of the real events; however, (i) multiple sets of rheological parameters may produce very similar output results, (ii) the best-fit input parameter combinations are non-transferable between models and inconsistent with independently measured rheological properties of stored tailings, and (iii) choosing an appropriate set requires sufficient understanding of material rheological properties and expert judgment. Using a systematic sensitivity analysis with the First-Order Second-Moment (FOSM) approach, we also find that each model is sensitive to different input parameters, although the total released volume is among the main high-influence parameters in every scenario. We conclude that more case study back-analyses are needed to enhance our understanding of these sensitivities and develop better guidance on the use of these types of numerical models for tailings flow runout prediction.
What problem does this paper attempt to address?