Generative Adversarial Networks to Forecast Debris Avalanche Hazard

Numerical models of hazardous environmental flows are tools frequently applied to hazard and risk assessment, scenario modelling and understanding of the underlying processes. The complexity and level of detail in numerical models have increased as computational power has become more available as parallelism and new computational techniques have reduced the computational burden of numerical modelling.  

However simulations through these numerical models still pose a large computational challenge. This can limit the applicability of numerical models to probabilistic hazard assessment, where a large number of simulations are required to adequately sample the probability space, and for rapid hazard/crisis assessment where simulation results need to be generated at short lead times.  

One approach to alleviating these limitations is through construction of a surrogate model. A surrogate (or metamodel) is a model of the simulation results within a defined parameter space. Surrogate models typically learn from simulation results to create a fast approximation to the numerical model. Here, we explore the use of neural networks as a surrogate model to generate debris avalanche footprints from Ruapehu volcano, New Zealand.  

Neural networks have shown promise for approximating Partial Differential Equations and may therefore produce a suitable trained surrogate model. A generative adversarial network (GAN) architecture is used as the surrogate, trained using depth-averaged debris avalanche simulation outputs. Our trials of GANs as a surrogate highlight a few key points: (1) well-trained GAN surrogates have high accuracy in reproducing model footprints, (2) the surrogate GANs show better accuracy when constrained to groups with similar simulation inputs (i.e. it is not generalizable), and (3) different loss functions (mean square error vs. mean absolute error) affect trainability and accuracy. These results, and future perspectives on the use of GANs for hazard assessment will be highlighted.