Deep learning based reduced order modelling applied to geophysical dynamics modelling

A predictive modelling framework for geophysical dynamical modelling is introduced. New numerical techniques such as, machine learning, reduced order modelling and data assimilation are employed. A parameterized non-intrusive reduced order model (P-NIROM) is introduced based on Proper Orthogonal Decomposition (POD) and machine learning methods. The unique combination of machine learning and POD methods can be used to reduce the computational effort. The predictive modelling capability has been demonstrated by comparing results with those from high fidelity full models and observational data. We have been successfully applied this P-NIROM to flooding prediction, ocean modelling, urban flows and air pollution. In this talk, we will demonstrate the capabilities of this new rapid model by applying it to the realistic cases in the UK and China.