Computational experiments on the 1962 and 1970 landslide events at Huascarán (Peru) with r.avaflow: Lessons learned for predictive mass flow simulations

Mass flow simulations are considered important tools for hazard analysis. For the simulation of single process mass flows such as debris flows, robust tools and reasonable parameter range estimates are available. However, this is much less the case for more complex mass flows, e.g. involving process chains and flow transformation. We explore the challenges of simulating complex flow-dominated landslides by back-calculating the Huascarán events of 1962 and 1970 with r.avaflow, a two-phase mass flow model (Pudasaini, 2012) in a GIS-based open source simulation framework. Both events started as rock-ice falls on the western slope of the north summit of Nevado Huascarán (Cordillera Blanca, Peru) and entrained large volumes of glacial till at lower elevation, resulting in highly mobile debris avalanches. Whereas the 1962 event badly affected the village of Ranrahirca when spreading over a debris cone, the 1970 event overtopped a ridge and led to the complete destruction of the town of Yungay. Well documented in the literature, these events provide an opportunity as a natural laboratory for testing innovative mass flow simulation tools and their features. In a first step, we consider (i) the 1962 event and (ii) the 1970 event separately, for each of them optimizing the key input parameters in terms of empirical adequacy. In a second step, we apply the optimized parameter set for (i) to the 1970 event and the parameter set derived for (ii) to the 1962 event. In a third step, we explore the sensitivity of the model outcomes to selected key parameters (basal friction angle and entrainment coefficient). The results (a) demonstrate the general ability of r.avaflow to reproduce the spatio-temporal evolution of flow heights and velocities as well as travel times and volumes of these complex mass flow events reasonably well; and (b) highlight the challenges and uncertainties involved in predictive simulations with parameter sets obtained from back-calculations. We suggest a strategy to appropriately deal with uncertain outcomes by superimposing the results of multiple simulations.