Bayesian active learning for parameter calibration of landslide run-out models

Abstract Landslide run-out modeling is a powerful model-based decision support tool for landslide hazard assessment and mitigation. Most models contain parameters that cannot be directly measured but rely on back-analysis of past events. As field data events come with certain measurement error, the community developed probabilistic calibration techniques. However, parameter often hindered by high computational costs resulting from long run time single simulation large number required model runs. To address this challenge, work proposes an efficient method integrating modeling, Bayesian inference, Gaussian process emulation, active learning. Here, we present extensive synthetic case study. The results show our new can reduce necessary runs thousands to few hundreds owing emulation It therefore expected advance current practice models.