Physics-constrained climate emulator of the moist global atmosphere learnt from ERA5 reanalysis and CMIP projections
Vassili Kitsios
A reduced order model of the global atmosphere is developed, in which the moist hydrostatic equations of motion are symbolically projected onto a set of empirical orthogonal functions (EOF). This approach transforms a system of partial differential equations dependent upon time and space, into a set of computationally cheap ordinary differential equations dependent upon time and EOF index. The required three-dimensional EOFs are calculated from the ERA5 atmospheric reanalysis to represent the dynamics of the anomalies. Reanalyses in general well capture climate variability and relationships between the teleconnections. However, diagnosing the climate change signal is not trivial due to the short observational record and slow natural processes in the Earth system. The EOF basis is then augmented with a climate change pattern (or shift mode) extracted from the multi-model ensemble average of Coupled Model Inter-comparison Project (CMIP) simulations. The reduced-order model coefficients are then calculated by solving a regression problem where the input factors are justified by those derived from the physical equations of motion. We have previously adopted similar approaches to deliver reduced-order models reproducing the large-scale variability of the dry atmosphere at a fraction of the computational cost required to numerically simulate the flow.
Please use this thread for discussion about this talk.