Applications of Physics-Informed Scientific Machine Learning in Subsurface Science: A Survey

Fast advances in machine learning (ML) algorithms and novel sensing technologies in recent years have presented new opportunities for the subsurface research community to improve the efficacy and transparency of geosystem governance. This chapter argues hat subsurface modeling is inherently a semisupervised generative modeling process, in which joint data distributions are learned via limited observations. In pre-training methods, which are widely used in ML-based surrogate modeling, prior knowledge and process-based models are mainly used to generate training samples for ML from limited real information. The physics is implicitly embedded in the training samples. ML methods originating from the computer vision typically assume the data has a Euclidean structure or can be reasonably made so through resampling. Early works adopting the deep learning methods explored their strong generative modeling capability for geologic simulation. Geosystems are a type of high-dimensional dynamic systems.