Integrating surrogate models into subsurface simulation framework allows computation of complex reactive transport scenarios

摘要

We showcase a flexible, extensible yet efficient framework for reactive transport modelling, including the ability to replace ‘full physics’ geochemical simulations with surrogate models for speedup. Surrogates are data-driven models trained on a set of pre-calculated simulations by means of machine-learning methods. We offer also an input-output-error visualization component for interactive assessment and tuning of their accuracy. Our framework, based on open source or freely available software, makes possible complex reactive transport simulations and ease further research on optimized algorithms to tackle many geoscientific problems.